However, how you use your tax refund can have a positive effect on your financial life, including your credit score.

How Tax Refunds Can Impact Your Credit Score

They are more likely to be one or two days in or even fully remote. Smaller-sized enterprises don’t always have the capacity to uplift remuneration so they will lever benefits.” ... Many Australian workers are being ordered back into the office five days a week as employer mandates create ...

Growing number of companies demand full return to office: 'Domino effect'

CFOs are weighing how much information to share about the impact of tariffs, especially as many are yet to take effect and the landscape continues to evolve.

The latest dilemma facing finance chiefs: What to tell investors about tariffs | Stock Market News

With more mandates taking effect, a record number of people are working from the office since the beginning of the pandemic.

Crowded trains, lunch and coffee lines: More people are working from the office

In-office work at highest level since 2020, as companies pull back on remote

These restrictions came into effect on Thursday (February 13) and will remain for six months, subject to review. RBI stated that the restrictions are necessary due to recent material developments affecting the bank. The move aims to protect depositors’ interests and ensure financial stability.

RBI directs New India Co-operative bank to stop all disbursements: What this means for depositors - CNBC TV18

One of the significant changes is the introduction of simpler terminology, with the ‘Tax Year’ replacing ‘Assessment Year’ and ‘Financial Year’ replacing ‘Previous Year’. The Tax Year will represent a 12-month period starting from April 1, aligning it with the financial year ...

New Income Tax Bill to come into effect from April 1, 2026 - Key details announced - Money News | The Financial Express

The session concluded with a powerful plea: “If we have to have centralized control of millions of IoT devices… how do we prevent a city from burning down?” This rhetorical question wasn’t just for dramatic effect; it was a genuine challenge to the cybersecurity community to think beyond ...

Addressing cyber risks of smart infrastructure, preventing catastrophic fires - Industrial Cyber

Update 2/8/2025, 3:00 PM CT: Some of PC Gamer's PS5-havers are once again able to access PSN, while I can also get to it on my PS4, and we're hearing scattered word to that effect on social media as well. Sony's PSN status page, however, is still reporting outages across all services.

Update: The PlayStation Network appears to be coming back online after nearly a day's outage, but there's still no word from Sony on what happened | PC Gamer

President Donald Trump has begun a new trade war with Canada, Mexico, and potentially China, and this has had a significant ripple effect on the broader financial landscape, including the cryptocurrency market.

The Nvidia panic, America's Bitcoin Reserve, and Trump's trade war hits stocks: Markets news roundup

effect

Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 LongFact: Using LLMs to generate a multi-topic benchmark for long-form factuality 2 LongFact: Using LLMs to generate a multi-topic benchmark for long-form factuality 3 Safe:LLM agents as factuality autoraters 3 Safe:LLM agents as factuality autoraters 4 LLMs agents can be better factuality annotators than humans 4 LLMs agents can be better factuality annotators than humans 5 F1@k: Extending F1 with recall from human-preferred length 5 F1@k: Extending F1 with recall from human-preferred length 6 Larger LLMs are more factual 6 Larger LLMs are more factual 7 Related Work 7 Related Work 8 Limitations 8 Limitations 9 Conclusion, Acknowledgments, Author Contribution, and References 9 Conclusion, Acknowledgments, Author Contribution, and References Appendix Appendix A. Frequently asked questions A. Frequently asked questions B. LongFact details B. LongFact details C. SAFE details C. SAFE details D. Metric details D. Metric details E. Further analysis E. Further analysis 4 LLM AGENTS CAN BE BETTER FACTUALITY ANNOTATORS THAN HUMANS LLMs have achieved superhuman performance on reasoning benchmarks (Bubeck et al., 2023; Gemini Team, 2023; Li et al., 2022) and higher factuality than humans on summarization tasks (Pu et al., 2023). The use of human annotators, however, is still prevalent in language-model research, which often creates a bottleneck due to human annotators’ high cost and variance in both evaluation speed and quality. Because of this mismatch, we investigate how SAFE compares to human annotations and whether it can replace human raters in evaluating long-form factuality To quantitatively evaluate the quality of annotations obtained using SAFE, we leverage crowdsourced human annotations released from Min et al. (2023). These data contain 496 prompt–response pairs where responses have been manually split into individual facts (for a total of 16,011 individual facts), and each individual fact has been manually labeled as either supported, irrelevant, or not supported. For each individual fact, we obtain an annotation from SAFE by running the SAFE pipeline, starting from revising the fact to be self-contained (i.e., steps 2–4 in Figure 1).[6] We then compare SAFE annotations with human annotations at the individual-fact level. First, we directly compare each individual fact’s SAFE annotation and human annotation, finding that SAFE agrees with humans on 72.0% of individual facts (see Figure 4). This indicates that SAFE achieves human-level performance on a majority of individual facts. We then examine a randomly-sampled subset of 100 individual facts for which the annotation from SAFE disagreed with the annotation from human raters. We manually re-annotate each of these facts (allowing access to Google Search rather than only Wikipedia for more-comprehensive annotations), and we use these labels as the ground-truth. We found that on these disagreement cases, SAFE annotations were correct 76% of the time, whereas human annotations were correct only 19% (see Figure 5), which represents a 4 to 1 win ratio for SAFE. Examples of correct SAFE ratings and incorrect SAFE ratings are shown in Appendix C.2 and Appendix C.3, respectively. Appendix A.3 analyzes failure causes for SAFE, and Appendix A.4 analyzes presumed failure causes for human annotators. To rate all individual facts from the 496 prompt–response pairs, SAFE issued GPT-3.5-Turbo API calls costing $64.57 and Serper API calls costing $31.74, thus costing a total of $96.31, equivalent to $0.19 per model response. For comparison, Min et al. (2023) reported a cost of $4 per model response to obtain crowdsourced human annotations for the dataset. This means that in addition to outperforming human raters, SAFE is also more than 20× cheaper than crowdsourced human annotators. Overall, the superior performance of SAFE demonstrates that language models can be used as scalable autoraters that have the potential to outperform crowdsourced human annotators. 5 F1@K: EXTENDING F1 WITH RECALL FROM HUMAN-PREFERRED LENGTH Finally, we combine factual precision and recall using standard F1, as shown as Equation (1). This proposed metric is bounded between [0, 1], where a higher number indicates a more-factual response. For a response to receive a full score of 1, it must not contain not-supported facts and must have at least K supported facts. F1@K also follows expectations for many clear hypothetical cases, as shown in Appendix D.2, and we contend that it allows for standardized and quantifiable comparisons between language models and between methods of obtaining responses. Equation (1): F1@K measures the long-form factuality of a model response y given the number of supported facts S(y) and the number of not-supported facts N(y) that are in y. The hyperparameter K indicates the number of supported facts required for a response to achieve full recall. We discuss how K should be selected in Appendix D.3. This paper is available on arxiv under CC by 4.0 Deed license. This paper is available on arxiv under CC by 4.0 Deed license. available on arxiv available on arxiv [6] We show performance of step 1 of SAFE (splitting a response into individual facts) in Appendix C.4. Authors:
(1) Jerry Wei, Google DeepMind and a Lead contributors;
(2) Chengrun Yang, Google DeepMind and a Lead contributors;
(3) Xinying Song, Google DeepMind and a Lead contributors;
(4) Yifeng Lu, Google DeepMind and a Lead contributors;
(5) Nathan Hu, Google DeepMind and Stanford University;
(6) Jie Huang, Google DeepMind and University of Illinois at Urbana-Champaign;
(7) Dustin Tran, Google DeepMind;
(8) Daiyi Peng, Google DeepMind;
(9) Ruibo Liu, Google DeepMind;
(10) Da Huang, Google DeepMind;
(11) Cosmo Du, Google DeepMind;
(12) Quoc V. Le, Google DeepMind. Authors: Authors: (1) Jerry Wei, Google DeepMind and a Lead contributors; (2) Chengrun Yang, Google DeepMind and a Lead contributors; (3) Xinying Song, Google DeepMind and a Lead contributors; (4) Yifeng Lu, Google DeepMind and a Lead contributors; (5) Nathan Hu, Google DeepMind and Stanford University; (6) Jie Huang, Google DeepMind and University of Illinois at Urbana-Champaign; (7) Dustin Tran, Google DeepMind; (8) Daiyi Peng, Google DeepMind; (9) Ruibo Liu, Google DeepMind; (10) Da Huang, Google DeepMind; (11) Cosmo Du, Google DeepMind; (12) Quoc V. Le, Google DeepMind.

Why LLMs Are More Accurate and Cost-Effective Than Human Fact-Checkers

Authors:
(1) Clemencia Siro, University of Amsterdam, Amsterdam, The Netherlands;
(2) Mohammad Aliannejadi, University of Amsterdam, Amsterdam, The Netherlands;
(3) Maarten de Rijke, University of Amsterdam, Amsterdam, The Netherlands. Table of Links Abstract and 1 Introduction 2 Methodology and 2.1 Experimental data and tasks 2.2 Automatic generation of diverse dialogue contexts 2.3 Crowdsource experiments 2.4 Experimental conditions 2.5 Participants 3 Results and Analysis and 3.1 Data statistics 3.2 RQ1: Effect of varying amount of dialogue context 3.3 RQ2: Effect of automatically generated dialogue context 4 Discussion and Implications 5 Related Work 6 Conclusion, Limitations, and Ethical Considerations 7 Acknowledgements and References A. Appendix 3.2 RQ1: Effect of varying amount of dialogue context Label quality. To gauge the quality of the crowdsourced labels, we rely on inter-annotator agreement (Boguslav and Cohen, 2017; Carletta, 1996). In order to understand how the amount of dialogue context influences the quality of ratings by annotators, we calculate the agreement between annotators for both relevance and usefulness across the three variations; see Table 1. To address potential randomness in relevance ratings, given the binary scale, we randomly drop one rating from each dialogue and compute the agreement. We repeat this process for each annotator and calculate an average Cohen’s Kappa score. For usefulness, we compute Kappa for each pair of annotators and then calculate the average. We assess the significance of the agreement using the Chi-squared method. All Kappa scores are statistically significant (p ≤ 0.05). We observe an increase in the Kappa and Tau score as the dialogue context increases from C0 to C7. Despite the lack of context in C0, there is a moderate level of agreement regarding the relevance of the current turn. With the introduction of more context in C3 and C7, comes an increase in agreement regarding the relevance of the current turn (see Table 1). Providing additional dialogue context seems to lead to higher levels of consensus among annotators. This is likely due to dataset characteristics: users tend to express their preferences early in the dialogue, rather than in subsequent exchanges. Hence, in the case of C0, which only includes the current turn, when the user’s utterance is incomplete, lacking an explicit expression of their preference, annotators rate more dialogues as relevant compared to C3 and C7. Overall, we conclude that when annotators have insufficient information to come up with a judgment, they tend to judge the system positively, introducing a positivity bias (Park et al., 2018). We see in Table 1 (row 3) that despite the lack of context in C0, there is substantial agreement regarding the usefulness of the current turn. This is due to the availability of the user’s next utterance, which serves as direct feedback on the system’s response, resulting in higher agreement than for relevance assessment. As more context is provided, there is an even higher level of agreement among annotators regarding the usefulness of the current turn. Access to a short conversation history significantly improves agreement on usefulness. Surprisingly, despite having access to the entire conversation history in C7, there is a slightly lower level of agreement than in C3. The complete dialogue context may introduce additional complexity or ambiguity in determining the usefulness of the current turn. This occurs when conflicting feedback arises from the user’s next utterance compared to the previous dialogue context. For example, when the system repeats a recommendation that the user has already watched or stated before, and the user expresses their intent to watch the movie in the next utterance, it leads to divergent labels. Similar trend is observed with the Tau correlations though the values are lower compared to the Kappa scores. Label consistency across conditions. We examine the impact of varying amounts of dialogue context on the consistency of crowdsourced labels across the three variations for relevance and usefulness and report the percentage of agreement in Figure 3. We observe moderate agreement (58.54%) between annotations of C0 and C3, suggesting that annotators demonstrate a degree of consistency in their assessments when provided with different amounts of context. This trend continues with C0 and C7, where the agreement increases slightly to 60.98%. The most notable increase is between C3 and C7 (68.29%). As annotators were exposed to progressively broader contextual information, their assessments became more consistent. Usefulness behaves differently. We observe moderate agreement (41.71%) between C0 and C3, indicating a degree of consistency in annotator assessments within this range of context. A notable decrease in agreement is evident when comparing C3 and C7, down to 28.3% agreement. The most substantial drop is observed between C0 and C7, yielding a mere 14.63% agreement. These findings emphasize the significant impact of context on the consistency of usefulness annotations. For usefulness assessment providing annotators with a more focused context, improves their agreement. With respect to RQ1, we note considerable differences in the labels assigned by annotators as we vary the amount of dialogue context. As the context expands, annotators incorporate more information into their assessments, resulting in context-specific labels. Annotator judgments are shaped not only by response quality but also by the broader conversation. This highlights the complexity of the task and the need for a carefully designed annotation methodology that considers contextual variations. These findings emphasize the significance of dialogue context in annotator decision-making. This paper is available on arxiv under CC BY 4.0 DEED license. Authors: (1) Clemencia Siro, University of Amsterdam, Amsterdam, The Netherlands; (2) Mohammad Aliannejadi, University of Amsterdam, Amsterdam, The Netherlands; (3) Maarten de Rijke, University of Amsterdam, Amsterdam, The Netherlands. Authors: Authors: (1) Clemencia Siro, University of Amsterdam, Amsterdam, The Netherlands; (2) Mohammad Aliannejadi, University of Amsterdam, Amsterdam, The Netherlands; (3) Maarten de Rijke, University of Amsterdam, Amsterdam, The Netherlands. Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Methodology and 2.1 Experimental data and tasks 2 Methodology and 2.1 Experimental data and tasks 2.2 Automatic generation of diverse dialogue contexts 2.2 Automatic generation of diverse dialogue contexts 2.3 Crowdsource experiments 2.3 Crowdsource experiments 2.4 Experimental conditions 2.4 Experimental conditions 2.5 Participants 2.5 Participants 3 Results and Analysis and 3.1 Data statistics 3 Results and Analysis and 3.1 Data statistics 3.2 RQ1: Effect of varying amount of dialogue context 3.2 RQ1: Effect of varying amount of dialogue context 3.3 RQ2: Effect of automatically generated dialogue context 3.3 RQ2: Effect of automatically generated dialogue context 4 Discussion and Implications 4 Discussion and Implications 5 Related Work 5 Related Work 6 Conclusion, Limitations, and Ethical Considerations 6 Conclusion, Limitations, and Ethical Considerations 7 Acknowledgements and References 7 Acknowledgements and References A. Appendix A. Appendix 3.2 RQ1: Effect of varying amount of dialogue context Label quality. To gauge the quality of the crowdsourced labels, we rely on inter-annotator agreement (Boguslav and Cohen, 2017; Carletta, 1996). In order to understand how the amount of dialogue context influences the quality of ratings by annotators, we calculate the agreement between annotators for both relevance and usefulness across the three variations; see Table 1. To address potential randomness in relevance ratings, given the binary scale, we randomly drop one rating from each dialogue and compute the agreement. We repeat this process for each annotator and calculate an average Cohen’s Kappa score. For usefulness, we compute Kappa for each pair of annotators and then calculate the average. We assess the significance of the agreement using the Chi-squared method. All Kappa scores are statistically significant (p ≤ 0.05). Label quality. We observe an increase in the Kappa and Tau score as the dialogue context increases from C0 to C7. Despite the lack of context in C0, there is a moderate level of agreement regarding the relevance of the current turn. With the introduction of more context in C3 and C7, comes an increase in agreement regarding the relevance of the current turn (see Table 1). Providing additional dialogue context seems to lead to higher levels of consensus among annotators. This is likely due to dataset characteristics: users tend to express their preferences early in the dialogue, rather than in subsequent exchanges. Hence, in the case of C0, which only includes the current turn, when the user’s utterance is incomplete, lacking an explicit expression of their preference, annotators rate more dialogues as relevant compared to C3 and C7. Overall, we conclude that when annotators have insufficient information to come up with a judgment, they tend to judge the system positively, introducing a positivity bias (Park et al., 2018). We see in Table 1 (row 3) that despite the lack of context in C0, there is substantial agreement regarding the usefulness of the current turn. This is due to the availability of the user’s next utterance, which serves as direct feedback on the system’s response, resulting in higher agreement than for relevance assessment. As more context is provided, there is an even higher level of agreement among annotators regarding the usefulness of the current turn. Access to a short conversation history significantly improves agreement on usefulness. Surprisingly, despite having access to the entire conversation history in C7, there is a slightly lower level of agreement than in C3. The complete dialogue context may introduce additional complexity or ambiguity in determining the usefulness of the current turn. This occurs when conflicting feedback arises from the user’s next utterance compared to the previous dialogue context. For example, when the system repeats a recommendation that the user has already watched or stated before, and the user expresses their intent to watch the movie in the next utterance, it leads to divergent labels. Similar trend is observed with the Tau correlations though the values are lower compared to the Kappa scores. Label consistency across conditions. We examine the impact of varying amounts of dialogue context on the consistency of crowdsourced labels across the three variations for relevance and usefulness and report the percentage of agreement in Figure 3. We observe moderate agreement (58.54%) between annotations of C0 and C3, suggesting that annotators demonstrate a degree of consistency in their assessments when provided with different amounts of context. This trend continues with C0 and C7, where the agreement increases slightly to 60.98%. The most notable increase is between C3 and C7 (68.29%). As annotators were exposed to progressively broader contextual information, their assessments became more consistent. Label consistency across conditions. Usefulness behaves differently. We observe moderate agreement (41.71%) between C0 and C3, indicating a degree of consistency in annotator assessments within this range of context. A notable decrease in agreement is evident when comparing C3 and C7, down to 28.3% agreement. The most substantial drop is observed between C0 and C7, yielding a mere 14.63% agreement. These findings emphasize the significant impact of context on the consistency of usefulness annotations. For usefulness assessment providing annotators with a more focused context, improves their agreement. With respect to RQ1 , we note considerable differences in the labels assigned by annotators as we vary the amount of dialogue context. As the context expands, annotators incorporate more information into their assessments, resulting in context-specific labels. Annotator judgments are shaped not only by response quality but also by the broader conversation. This highlights the complexity of the task and the need for a carefully designed annotation methodology that considers contextual variations. These findings emphasize the significance of dialogue context in annotator decision-making. RQ1 This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv

The Surprising Effects of Minimal Dialogue Context on AI Judgment

Author:
(1) Jerker Denrell, University of Warwick (jdenrell@gmail.com) Author: Author: (1) Jerker Denrell, University of Warwick (jdenrell@gmail.com) Table of Links Abstract and 1. Introduction 2. Illustration 3. Intuition 4. General Theorem about Biased Averages 5. Alternative Learning Models 6. Bayesian Updating 7. Implications for Understanding Learning Appendix and References Abstract. The Hot Stove Effect is a negativity bias resulting from the adaptive character of learning. The mechanism is that learning algorithms that pursue alternatives with positive estimated values, but avoid alternatives with negative estimated values, will correct errors of overestimation but fail to correct errors of underestimation. Here, we generalize the theory behind the Hot Stove Effect to settings in which negative estimates do not necessarily lead to avoidance but to a smaller sample size (i.e., a learner selects fewer of alternative B if B is believed to be inferior but does not entirely avoid B). We formally demonstrate that the negativity bias remains in this set-up. We also show there is a negativity bias for Bayesian learners in the sense that most such learners underestimate the expected value of an alternative. Abstract 1. Introduction Learning from experience does not necessarily generate unbiased beliefs, partly due to psychological biases but also due to biases in the information learners’ sample and get exposed to. One important bias in sampling is the so-called "hot stove effect" (Denrell & March 2001) which refers to the asymmetry in error correction generated by adaptive learning processes. The key idea is that the tendency to avoid alternatives with unfavorable past outcomes generates a biased set of experiences. Alternatives that are underestimated - believed to be worse than what they are – are unlikely to be tried and sampled again, which implies that errors of underestimation are unlikely to be corrected. Alternatives that are overestimated - believed to be better than what they are - are likely to be tried and sampled again which implies that errors of overestimation are likely to be corrected. This asymmetry in error correction generates a biased set of experiences that, in turn, can give rise to biased judgments (Denrell, 2005), including in-group bias and apparent risk-averse behavior (Denrell, 2005, 2007). There is good experimental support for the hot-stove effect at the individual level and researchers in psychology have relied on the hot-stove effect to explain regularities in risk taking in experimental studies (Erev & Roth, 2014) and why people underestimate the trustworthiness of others (Fetchenhauer & Dunning, 2014). Researchers in finance (Dittmar & Duchin, 2016) have used field data to demonstrate that the hot stove effect can explain the risk-taking behavior of executives. The hot stove effect also has important implications for information aggregation and online reviews: If consumers avoid products with poor reviews and consumers review products they buy, negative reviews will be more persistent than positive reviews, generating biased averages (Mens et al., 2018). Past theoretical work on the hot stove effect has assumed that negative experiences may lead to avoidance, that is, the alternative has not been tried at all (Denrell, 2005, 2007). Clearly, if no more information is available, a negative impression will persist. In many settings, however, a negative belief or impression may not lead to avoidance of the alternative, but rather to a smaller sample size. An animal with a more favorable impression of the energy content of a plant of type A than of a plant of type B may search for plants of type A. During this search for plants of type A, some plants of type B may be incidentally found. The result is that the animal samples more plants of type A (because the search is focused on such plants) than of type B, but the animal does not avoid plants of type B but simply samples fewer of them. Similarly, a firm may prefer to hire graduates from university A, but may hire some, although fewer, graduates from university B if there are not enough graduates from A that accept its offers. In this paper, I generalize the theory behind the hot-stove effect and show that it holds even if a negative impression only leads to a reduction in the sample size, not necessarily to avoidance. Specifically, I show that the final belief will be biased for a broad class of learning algorithms in which the sample size is a function of the past belief. If the sample size is higher and the past belief was more positive, there is a negativity bias: The final belief will be lower than the expected value of the random variable the learner is learning about. This result also applies to taking averages: the average of a sample will be biased if the total sample size is a function of the average based on an initial subset. The bias will be eliminated in the long run as the number of samples increases, but in the short run it can be significant. I also examine whether the bias remains for a Bayesian learner. I show that there is no bias on average for a Bayesian learner: the average belief is equal to the expected value of the random variable the learner is learning about. However, I also show that most Bayesian learners will underestimate the variable they are learning about if the sample size is an increasing function of the initial belief. These results imply that a large class of sensible and adaptive learning processes can be expected to generate biased beliefs, even if decision-makers process the available information in a seemingly unbiased way (i.e., taking averages). Indeed, even rational Bayesian learners will tend to underestimate the expected value of an alternative (i.e., most will do so) if the total sample size is higher when the initially observed payoffs are high. Adaptive sampling processes are common and often necessary to reduce search costs. It is not sensible, for example, to continue to sample an alternative a fixed number of times if initial trials reveal that this alternative has a much lower payoff than other available alternatives. The results in this paper show that even unbiased processing of information generated by such sampling policies can generate seemingly biased beliefs. Adaptive sampling policies thus offer an alternative explanation of biases in beliefs, such as a tendency to underestimate the extent to which others are trustworthy. 2. Illustration 3. Intuition As this example shows, adaptive sampling implies that negative first-period averages are more ’persistent’: they are given a larger weight than positive firstperiod averages. Because negative and positive first-period averages of the same magnitude are equally likely, when the payoff distribution is normal with mean zero (this distribution is symmetric around zero), the greater persistence of negative first-period averages explains the overall negative bias. The impact of variance can also be understood in a similar way. If the payoff distribution is more variable, the first period averages will tend to differ more from zero, both in the positive and negative directions. Positive first-period averages will tend to regress more to the mean (zero) than negative averages. When the negative first-period averages are more extreme because the variance is greater, the result is a stronger bias. Of course, the bias will be eliminated in the long run, as the number of samples increases since the average of n samples of a random variable X will converge to its expected value. In the short run, however, the bias can be significant. n 4. General Theorem about Biased Averages The bias generated by an adaptive sampling size policy does not only hold for the normal distribution and the specific binary sampling policy considered above (high above zero, low below), but it holds for any distribution and a large class of adaptive sampling policies. Proof: See the Appendix. Proof For unimodal symmetric distributions, centered on the expected value E[x] = u, it can also be shown that a majority of the averages, after the second period, are below u if higher averages in the first period lead to larger sample sizes in the second period: Proof: See the Appendix. Proof 5. Alternative Learning Models So far we have assumed that the learner computes the average of the observed payoffs, but a similar bias holds for several other learning models. Suppose, for example, that the learner gives more weight to the most recently observed payoff. This results in a similar bias: Proof: See the Appendix. Proof 6. Bayesian updating We now show that Bayesian updating remains unbiased in this latter sense, even if the sample size in period two depends on the belief after period one. This is true for any distribution: Proof: See the Appendix. Proof Thus, even if Bayesian learners are learning about a symmetric distribution and they have a prior that is symmetric around zero (m = 0), most learners will have a negative belief after sampling. This is due to the adaptive sampling policy. The intuition is similar to why the learning policy that takes the average is biased: negative initial beliefs are more persistent because the learner will then not take many additional samples, and the initial few negative observations will be weighted heavily. Note that such a bias would not occur if the learner followed a fixed sampling policy and decided, at the outset, how many samples to take. If the sample size was fixed, the belief would be equally likely to be positive or negative, when learning about a normally distributed payoff with a mean taken from normally distributed prior with mean zero. Note also that when sampling is adaptive, all Bayesian learners know that only 50% of the means are negative. Still, most Bayesian learners believe that the mean they observe is negative. That a majority of Bayesian learners end up with a negative belief may seem paradoxical since on average there is no bias: E[b2] = 0 when m = 0, i.e., the average belief, averaging over all Bayesian learners, is equal to the mean of the prior. The paradox is resolved by noting that the learners who believe that the mean is negative are less confident in this estimate than the learners who believe that the mean is positive and therefore took a larger sample in the second period. This results in a skewed distribution of beliefs after the second period. Most beliefs are below m = 0 but those below m = 0 are less likely to be extreme beliefs than beliefs above m = 0 since beliefs below m = 0 are based on smaller sample sizes than those above m = 0. This can be seen in Figure 1. 7. Implications for Understanding Learning The bias resulting from adaptive sampling policies implies that even seemingly unbiased learning algorithms, such as averaging or Bayesian updating, can result in biased beliefs, at least in the sense that most learners underestimate (or overestimate) an alternative. This offers an alternative explanation of some judgment biases; an explanation that does not require a psychological bias that assumes biases in information processing. For example, suppose that it is observed that a firm tends to underestimate the productivity of graduates from universities. Such a negativity bias. can be explained by an adaptive sampling policy (the firm tends to hire fewer people from universities it has had a worse experience with) and does not require motivated reasoning or a cognitive bias. Appendix Proof of Theorem 2: Proof of Theorem 2: Without loss generality we only consider cases when u = 0. Aunimodal random variable symmetric around the mode k can always be represented as x = k + ε where ε has a unimodal distribution symmetric around zero. We first prove the following Lemma The proof now follows by induction, because the sum of k independent random variables with a unimodal distribution symmetric around u is also unimodal symmetric centered on u (Shaked & Shantikumar (2007), page 173). Consider a fixed value of r. Suppose r > 0. The average after the second period becomes less than zero whenever The probability that the average is below zero after the second period is thus Altogether, the probability that the average is below zero after the second period equals Altogether, the probability that a positive belief after period one turns into a negative belief after period two equals where f(z) is the density of the sum of the payoffs in the first period which is a normal distribution with mean zero. Altogether, the probability that a negative belief after period one turns into a positive belief after period two, Pr(− → +), equals After the variable substitution, z = −y, this integral equals where we have used the fact that f(−y) = f(y). We wish to show that Pr(+ → −) > Pr(− → +), which requires us to show that over the positive domain of z. To do so, note first that the derivative of with respect to w is References Denrell, J. (2005). Why most people disapprove of me: Experience sampling in impression formation. Psychological Review, 112, 951–978. Denrell, J. (2007). Adaptive learning and risk taking. Psychological Review, 114, 177–187. Denrell, J., & March, J. (2001). Adaptation as information restriction: The hot stove effect. Psychological Review, 12, 523–538. Dittmar, A., & Duchin, R. (2016). Looking in the rear-view mirror: The effect of managers’ professional experiences on corporate financial policy. Review of Financial Studies, 29, 565–602. Erev, I., & Roth, A. (2014). Maximization, learning, and economic behavior. Proceedings of the National Academy of Sciences, 111, 10818–10825. Fetchenhauer, D., & Dunning, D. (2014). Why so cynical?: Asymmetric feedback underlies misguided skepticism regarding the trustworthiness of others. Psychological Science, 21, 189–193. Mens, G. L., Kovács, B., Avrahami, J., & Kareev, Y. (2018). How endogenous crowd formation undermines the wisdom of the crowd in online ratings. Psychological Science, 29, 1475–1490. Shaked, M., & Shantikumar, J. G. (2007). Stochastic orders. New York, NY: Springer. Williams, D. (1991). Probability with martingales. Cambridge, UK: Cambridge University Press. Author:
(1) Jerker Denrell, University of Warwick (jdenrell@gmail.com) Author: Author: (1) Jerker Denrell, University of Warwick (jdenrell@gmail.com) This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv available on arxiv

The Hot Stove Effect: Why AI Learns to Be a Pessimist

In just a short time, TikTok has become one of the most frequented apps in the world. With millions of creators engaging on a daily basis, enhancing these engagements can give you better visibility and a larger audience. Engagement is very critical if you are a creator on TikTok since it can help grow your visibility. In this article, we’ll outline the best practices to capture attention and maintain your audience’s interest so you can increase likes, shares, and comments. Importance of Understanding TikTok’s Algorithm Understanding the TikTok algorithm is fundamental to increasing the visibility of your content or learning how to get more views on TikTok. Now, let’s explore how TikTok ranks content and the importance of different engagement metrics. how to get more views on TikTok how to get more views on TikTok How TikTok Ranks Content As soon as a user logs into the app, they are greeted with a For You Page (FYP) that is customized to their personal needs. Each user is recommended videos that best suit their personal interests. TikTok has certain algorithms that make use of certain proprietary methods that ascertain what videos are suitable and what are not for each individual user. Some of the most crucial elements are: Shares and Likes: The videos that were liked, shared, or even watched till the end are taken into consideration by TikTok in order to gauge the type and category of content that is ideal.


Soundtrack and Description: Footage transcripts, other words of speech in accompanying sound, and hashtags from the informants that TikTok puts order to the video’s information as well as to serve it to the registered users.


Device and Account Settings: The languages a person speaks, what country they are in, or even what kind of device they use is of great importance for determining what context should be shown in what tailor videos. Shares and Likes: The videos that were liked, shared, or even watched till the end are taken into consideration by TikTok in order to gauge the type and category of content that is ideal. Shares and Likes: The videos that were liked, shared, or even watched till the end are taken into consideration by TikTok in order to gauge the type and category of content that is ideal. Shares and Likes: Soundtrack and Description: Footage transcripts, other words of speech in accompanying sound, and hashtags from the informants that TikTok puts order to the video’s information as well as to serve it to the registered users. Soundtrack and Description: Footage transcripts, other words of speech in accompanying sound, and hashtags from the informants that TikTok puts order to the video’s information as well as to serve it to the registered users. Soundtrack and Description: Device and Account Settings: The languages a person speaks, what country they are in, or even what kind of device they use is of great importance for determining what context should be shown in what tailor videos. Device and Account Settings: The languages a person speaks, what country they are in, or even what kind of device they use is of great importance for determining what context should be shown in what tailor videos. Device and Account Settings: These factors aid in the construction of a custom feed designed to prolong user retention time while stimulating increased user engagement. The Role of Watch Time, Shares, Likes, and Comments These engagement metrics directly influence how widely your video is promoted: Watch Time: If a user watches the video content for a longer duration, preferably in full, the TikTok algorithm will reward the video with increased promotion. Good watch time shows that the content is beneficial.


Shares: When audiences circulate your content, it sends a message to TikTok that your content is relevant and drives a further increase in algorithm support while paying attention to your content.


Likes and Comments: Likes are an appreciation while comments represent higher engagement. Both aspects heavily affect the popularity of your video content, especially its position on the FYP. Watch Time: If a user watches the video content for a longer duration, preferably in full, the TikTok algorithm will reward the video with increased promotion. Good watch time shows that the content is beneficial. Watch Time: If a user watches the video content for a longer duration, preferably in full, the TikTok algorithm will reward the video with increased promotion. Good watch time shows that the content is beneficial. Watch Time: Shares: When audiences circulate your content, it sends a message to TikTok that your content is relevant and drives a further increase in algorithm support while paying attention to your content. Shares: When audiences circulate your content, it sends a message to TikTok that your content is relevant and drives a further increase in algorithm support while paying attention to your content. Shares: Likes and Comments: Likes are an appreciation while comments represent higher engagement. Both aspects heavily affect the popularity of your video content, especially its position on the FYP. Likes and Comments: Likes are an appreciation while comments represent higher engagement. Both aspects heavily affect the popularity of your video content, especially its position on the FYP. Likes and Comments: It’s easier to encourage broader TikTok reach by focusing on content that will make viewers want to watch, share, comment, and like the video. Creating Compelling and High-Quality Content If you want to attract your audience and improve your brand presence on TikTok, then incorporating captivating content will play an essential role. TikTok is focused on short and direct videos. According to research, videos in between 21-34 seconds tend to receive the highest engagement rates. This is because it resonates with the users’ consumption patterns, enabling creators to capture attention efficiently. 21-34 seconds 21-34 seconds Participating with the trending challenges and sounds can boost your content reach drastically taking it to a whole new level. By using popular audio clips or partaking in viral challenges, your videos stand a better chance of appearing on people’s ‘For You’ page which means wider coverage for your content. The use of storytelling techniques can aid in keeping the viewers glued to the screens. A good story can transform the way the audience perceives your content and make it stick in their minds. Creator Academy by TikTok reinforces the value of storytelling, remarking that effective content often wins because of great storytelling. Leveraging TikTok Features & Tools Since there are tools that can be used creatively for engagement on the platform, content on TikTok can always be improved. The following is a list of what can be optimized on each feature: Duets and Stitch allow for collaborating with other TikTok users. With Duets or side-by-side videos, you can give your perspective or react to already existing videos. With Stitches, you can create your own videos alongside clips from other people’s videos, thus making poems or storylines. Captioning and text overlays make videos accessible, emphasize important points, and contextualize speech, allowing everyone to follow along. To create text overlays, follow the following steps: Record your video or upload it on TikTok.


Click on “Text” and put in the text.


Adjust the font, color, and placement as desired.


Decide when the text should appear and disappear. Record your video or upload it on TikTok. Record your video or upload it on TikTok. Click on “Text” and put in the text. Click on “Text” and put in the text. Adjust the font, color, and placement as desired. Adjust the font, color, and placement as desired. Decide when the text should appear and disappear. Decide when the text should appear and disappear. Adding stickers to videos can spice them up and encourage interaction. Nevertheless, it is recommended to use stickers in a limited and controlled manner so as not to complicate your content. As previously mentioned, TikTok has many features filters, and effects to build up the aesthetic of your videos: Filters: These will assist greatly in establishing the aesthetic you want because they can alter your video’s color and mood.


Effects: Tools of interest include adding transitions or AR objects which will escalate the visuals of your videos. Filters: These will assist greatly in establishing the aesthetic you want because they can alter your video’s color and mood. Filters: These will assist greatly in establishing the aesthetic you want because they can alter your video’s color and mood. Filters: Effects: Tools of interest include adding transitions or AR objects which will escalate the visuals of your videos. Effects: Tools of interest include adding transitions or AR objects which will escalate the visuals of your videos. Effects: These features, when used appropriately, can increase the aesthetic appeal of the content which can boost viewership rates. Posting Consistently and at Optimal Times It is a common misconception that TikTok is all about creativity. What many users do not know is that consistency is another core pillar. To highlight this, accounts that post more than three times a week average an engagement rate of around 6%, while the lower frequency accounts hover around 5.8%. Therefore, if you would like to grow your audience, aim to post 3-5 times a week. 5.8% 5.8% 3-5 times 3-5 times Although these initial guidelines can be very useful, personal reflection is as vital. As mentioned earlier, TikTok provides analytics that show when your target audience comes online and how they interact with the content you post. Posting according to your audience’s schedule guarantees content interaction which leads to a stronger community. In brief, posting between 3-5 times a week at high-traffic times will boost your visibility on TikTok. Tracking your TikTok analytics keeps your strategy fresh, adaptable, and efficient. Encouraging Audience Interaction As with other social media platforms, community building on TikTok fosters positivity, but so does interaction with your audience. Having someone accomplish a task increases engagement levels dramatically. Ads with a call-to-action capture attention almost 45 percent more than those without, proving the benefit of using actionable prompts. 45 percent 45 percent Tips:

Participation in answering widespread questions that are relatable motivates viewers to comment and increases chances of engagement.
Encouraging users to share your posts with peers who may benefit from them will greatly enhance engagement rates on your content. Tips: Tips: Participation in answering widespread questions that are relatable motivates viewers to comment and increases chances of engagement.
Encouraging users to share your posts with peers who may benefit from them will greatly enhance engagement rates on your content. Participation in answering widespread questions that are relatable motivates viewers to comment and increases chances of engagement. Encouraging users to share your posts with peers who may benefit from them will greatly enhance engagement rates on your content. Clusters of active users create an environment for belonging. Because using a two way street logic to build community requires appreciation, responding to comments and DMs helps in cultivating deeper relationships with users. Tips:

Replying in a timely manner keeps the conversation flowing.
Engagement is further promoted by responding to well-voiced comments of user videos. Tips: Tips: Replying in a timely manner keeps the conversation flowing.
Engagement is further promoted by responding to well-voiced comments of user videos. Replying in a timely manner keeps the conversation flowing. Engagement is further promoted by responding to well-voiced comments of user videos. Remember that running contests and giveaways can drive more activity as people will want to participate. Incentive-driven campaigns on TikTok are especially common, evidenced by the 3.1 million posts containing the ‘giveaway’ hashtag. 3.1 million 3.1 million Tips:

Set Specific Guidelines: Participants should be clear on how to win and redeem their rewards.
Increase the value of the offered rewards: Select rewards that are relevant to your audience so that more people are compelled to take part. Tips: Tips: Set Specific Guidelines: Participants should be clear on how to win and redeem their rewards.
Increase the value of the offered rewards: Select rewards that are relevant to your audience so that more people are compelled to take part. Set Specific Guidelines: Participants should be clear on how to win and redeem their rewards. Increase the value of the offered rewards: Select rewards that are relevant to your audience so that more people are compelled to take part. Collaborating with Other Creators The use of other creators enables you to reach a broader audience and increase your exposure. For instance, TikTok creators develop branded content on the platform through collaboration and it receives 83% more engagement than counterpart content developed without collaboration. Specially designed partnered ads for TikTok receive 2-second capture view rates. These figures show how collaboration positively influences engagement and reach on the platform. How To Find Creators To Collaborate With Determine Relevant Creators for Your Brand: The target partners need to be picked from those whose work overlaps with yours at some level because they form part of the desired audience.


Engage With Their Content: Begin engaging with their videos by liking, commenting, and sharing. This kind of engagement may lead to a collaboration down the line.


Professionally Inquire: Show interest by sending them a direct message or email. It is important that you explain your ideas for the collaboration along with how both parties would profit.


Through TikTok’s Creator Marketplace: This site serves as a hub for brands and creators, which helps find collaborators easily. Determine Relevant Creators for Your Brand: The target partners need to be picked from those whose work overlaps with yours at some level because they form part of the desired audience. Determine Relevant Creators for Your Brand: The target partners need to be picked from those whose work overlaps with yours at some level because they form part of the desired audience. Determine Relevant Creators for Your Brand: Engage With Their Content: Begin engaging with their videos by liking, commenting, and sharing. This kind of engagement may lead to a collaboration down the line. Engage With Their Content: Begin engaging with their videos by liking, commenting, and sharing. This kind of engagement may lead to a collaboration down the line. Engage With Their Content: Professionally Inquire: Show interest by sending them a direct message or email. It is important that you explain your ideas for the collaboration along with how both parties would profit. Professionally Inquire: Show interest by sending them a direct message or email. It is important that you explain your ideas for the collaboration along with how both parties would profit. Professionally Inquire: Through TikTok’s Creator Marketplace: This site serves as a hub for brands and creators, which helps find collaborators easily. Through TikTok’s Creator Marketplace: This site serves as a hub for brands and creators, which helps find collaborators easily. Through TikTok’s Creator Marketplace: Using Hashtags and Trends Wisely When it comes to TikTok, using hashtags and trends wisely can greatly increase the reach of your content and boost overall engagement. The trick is to find a balance between being easy to find and originality. While picking hashtags, they should be relevant and specific. Rather than using vague, overly competitive hashtags, choose ones that pertain directly to your content. For instance, if you are sharing workout tips, using the hashtag #FitnessJourney will be greatly preferable than #FYP. It’s also beneficial to combine trending hashtags with niche ones. Trending hashtags will ensure that your content is viewed by a vast audience, while niche hashtags will ensure that your viewership is comprised of people who actually care about your content. Consistently checking TikTok's Discover page will help you determine which hashtags are trending. Chasing the latest trends can tremendously increase your engagement, but using originality helps retain viewers in the long run. Instead of blindly following a trending format, you can try making it your own. Maybe you can use your humorous touch, a fresh angle, or shift the trend into your normal content style. Trends should aid your brand and not overtake it. You can try different approaches or even settings, and those slight changes will still let you be part of the trend while making your mark. Analyzing Performance and Adapting Strategies With the ever-changing social media environments TikTok in particular, it is important that you keep track of how well your content performs and adjust accordingly if you want to stand out. Using tools provided by TikTok, it is possible to track and analyze engagement metrics from an integrated analytics tool. Analyzing engagement from TikTok’s analytics page can be done regarding how well your videos are performing. Metrics such as total views, total likes, total shares, total comments, and total watch time are important components to provide someone with a greater understanding of what their audience values. According to reports TikTok shares, videos that receive greater engagement tend to get recommended more frequently on the For You Page (FYP) leading to an increase in visibility). After analyzing the performance data, the next step is to adjust the content. If some videos perform better than others, figure out how trends, video length, or specific niches affect performance. For instance, if your analytics suggest posting around 8 PM yields better engagement, try to stick to that time. Conclusion Gaining attention for a TikTok account is as simple as creatively posting and interacting with others on the device. To garner maximum viewer attention, it is important to understand the algorithm, craft quality content, post during peak times, engage with the audience, and be active on the platform. Ensure that the content strategy is revised according to the analytics data. Continue to be authentic and innovative in efforts, and the audience response will be favorable.

Effective Strategies for Enhancing Engagement on TikTok

I wanted to recreate the following logo image using only HTML and CSS. For me it was important that the setup supports different font sizes and text content so that I can easily change the logo text and size without any code changes. Solution 1. Setup An HTML Container With Child Elements <a href="/" class="logo__link">
 Piranhas
 Piranhas
 Piranhas
</a> <a href="/" class="logo__link">
 Piranhas
 Piranhas
 Piranhas
</a> My idea was to create 2 child elements containing the same text. I would clip one element to the upper half (.logo__name-top) and the other element to the lower part (.logo__name-bottom). Finally, I will displace the upper part to the left. That's it. .logo__name-top .logo__name-bottom 2. Add CSS Properties To HTML Container .logo__link {
 position: relative;
 color: #ffff00;
 font-family: Outfit, Helvetica, sans-serif;
 font-style: italic;
 font-size: 5rem;
 font-weight: 900;
 text-decoration: none;
} .logo__link {
 position: relative;
 color: #ffff00;
 font-family: Outfit, Helvetica, sans-serif;
 font-style: italic;
 font-size: 5rem;
 font-weight: 900;
 text-decoration: none;
} The position property is needed to allow absolute positioning of the child elements.
The other CSS properties are just used for styling. The position property is needed to allow absolute positioning of the child elements. position The other CSS properties are just used for styling. 3. Add CSS Properties To HTML Child Elements .logo__name-top {
 position: absolute;
 top: 0;
 right: 0.05em;
 clip-path: inset(0% 0% 40% 0%);
} .logo__name-top {
 position: absolute;
 top: 0;
 right: 0.05em;
 clip-path: inset(0% 0% 40% 0%);
} The position property is needed to apply absolute positioning to the .logo__name-top element.
The right property is used to displace the text to the right.
The clip-path property is used to clip the text to the upper 40% and hide the rest. The position property is needed to apply absolute positioning to the .logo__name-top element. position .logo__name-top The right property is used to displace the text to the right. right The clip-path property is used to clip the text to the upper 40% and hide the rest. clip-path .logo__name-bottom {
 position: absolute;
 top: 0;
 right: 0;
 clip-path: inset(60% 0% 0% 0%);
} .logo__name-bottom {
 position: absolute;
 top: 0;
 right: 0;
 clip-path: inset(60% 0% 0% 0%);
} The position property is needed to apply absolute positioning to the .logo__name-top element.
This elements stays where is is and is not displaced.
The clip-path property is used to clip the text to the lower 60% and hide the rest. The position property is needed to apply absolute positioning to the .logo__name-top element. position .logo__name-top This elements stays where is is and is not displaced. The clip-path property is used to clip the text to the lower 60% and hide the rest. clip-path As you can see, there are still some remaining issues. Styling Issues 1. Avoid Clipped Areas For Italic Font Style I needed to extend the clipping area to the right by adding -5% to the clip-path. Otherwise, some part of the text gets clipped away when using italic font style. -5% clip-path .logo__name-top {
 position: absolute;
 top: 0;
 right: 0.05em;
 clip-path: inset(0% -5% 40% 0%);
}
.logo__name-bottom {
 position: absolute;
 top: 0;
 right: 0;
 clip-path: inset(60% -5% 0% 0%);
} .logo__name-top {
 position: absolute;
 top: 0;
 right: 0.05em;
 clip-path: inset(0% -5% 40% 0%);
}
.logo__name-bottom {
 position: absolute;
 top: 0;
 right: 0;
 clip-path: inset(60% -5% 0% 0%);
} 2. Parent Element Does Not Use Any Space Since the child elements are using absolute positioning, they are out of the flow. Therefore, the parent element .logo__linkdoes not take any space and makes positioning the element difficult. You can manually position the parent element but when you change the logo text or size, you would need to manually update the positon of the parent element which is not sustainable. .logo__link Therefore, I added a "pseudo" child element in HTML with class .logo__name-pseudo containing the same text in the same style and size which is hidden just to reserve space for the clipped elements. Now the element reains in the flow. .logo__name-pseudo .logo__name-pseudo {
 visibility: hidden;
} .logo__name-pseudo {
 visibility: hidden;
} 3. Remove Thin Line Between Displaced Elements There remains a thin line between the 2 elements and I couldn't figure out the reason. Eventually, it is a rendering issue. It depends also on the logo size. Sometimes it appears and sometimes not. I decided to extend the upper clip area and clip away just 39% instead of 40% of the bottom part. Therefore, the logos slightly overlap and the line is removed. If you have a better solution, let me know in the comments. 39% 40% .logo__name-top {
 position: absolute;
 top: 0;
 right: 0.05em;
 clip-path: inset(0% -5% 39% 0%);
} .logo__name-top {
 position: absolute;
 top: 0;
 right: 0.05em;
 clip-path: inset(0% -5% 39% 0%);
} Here is the final code on CodePen. Feel free to use. final code on CodePen Originally published on March 13th, 2025 on Philip's Notes. Originally published on March 13th, 2025 on Philip's Notes. Philip's Notes

Create A Lightning Text Effect Using HTML And CSS Using This Simple Code

Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2 Background 3 Approach and 3.1 Differential Testing for XML Processors 3 Approach and 3.1 Differential Testing for XML Processors 3.2 XPath Expression Generation 3.2 XPath Expression Generation 3.3 XML Generation 3.3 XML Generation 4 Evaluation 4 Evaluation 4.1 Effectiveness 4.1 Effectiveness 4.2 Efficiency 4.2 Efficiency 4.3 Comparison to the State of the Art 4.3 Comparison to the State of the Art 4.4 Analysis of BaseX Historical Bug Reports 4.4 Analysis of BaseX Historical Bug Reports 5 Related Work 5 Related Work 6 Conclusion, Acknowledgments, and References 6 Conclusion, Acknowledgments, and References 4.1 Effectiveness In this section, we show XPress’ effectiveness through the number of bugs found, developer feedback, and illustrative examples. Methodology. We implemented the tool while intermittently testing the systems over a period of 3 months. For every found discrepancy, we reduced the test case. If the test case exhibited an unreported pattern, we considered it likely to be an unknown bug and reported it to the developers. Note that this was a best-effort approach, and that it is an unsolved problem of how to identify duplicate bugs effectively. Whether we considered a bug as unique was based on the developers’ verdict; we considered a bug only as unique if an issue was addressed through an independent bug fix. Unfixed bugs hinder testing, as the duplicates tend to be repeatedly triggered. To tackle this, we attempted to disable the construction of bug-inducing elements, and also ignored known discrepancy patterns before the reported bug was resolved. Methodology Found bugs overview. As shown in Table 1, we successfully found 27 unique bugs in total, 15 in BaseX, 6 in eXist-DB, 4 in Saxon, and 2 in the commercial DBMS. As detailed subsequently, we could have reported additional bugs for eXist-DB and the commercial DBMS, but refrained from doing so due to the high number of unfixed bugs for eXist-DB, and lack of developer feedback for the commercial DBMS. The bug-inducing test cases we found were not covered by the W3C qt3 test suite [19], which contains around 30,000 tests for XPath and XQuery—Saxon 11.1 passes all applicable tests in the W3C qt3 test suite [17]. Out of the 27 bugs found, the majority, 19 bugs, were logic bugs. Based on developer feedback, we learned that among the 20 fixed bugs, at least 8 bugs were due to incorrect optimizations. We detected the remaining bugs through unexpected errors. All systems we tested were implemented in Java, so we did not observe any crash bugs. We did not find any bugs in PostgreSQL and libXML2, both of which are known to be robust systems. For example, previous bug-finding efforts on testing DBMSs using SQL queries also found no logic bug in PostgreSQL [34, 35]. Found bugs overview. Small-scope hypothesis. We observed that the reported bugs are mainly reproducible by short test cases. 70% of all the reported cases can be reproduced with an XML document that consists of only one node and 91% of XPath expression consists of only one section. The average length of the XML documents in the reported test cases was 12 characters and XPath expressions 30 characters. This Small-scope hypothesis. phenomenon is known as the small-scope hypothesis [21], and this observation has been exploited in test. Developer reception. Developer feedback is an important indicator of the bugs’ importance. A core developer of BaseX stated *"Thanks for sharing the bug reports with us. I appreciate that, they’re definitely helpful."[*3] All 15 bugs reported to BaseX were resolved within one month—10 bugs were resolved even within 24 hours. This indicates not only that the team was fast in resolving bugs, but also that the bug reports were considered valuable. Due to the timely fixes of the BaseX team, we invested most time and effort in testing BaseX. After encouragement from the developers of BaseX, we contributed the bug-inducing test cases to the W3C XQuery and XPath test suite [19]. Most bugs submitted to eXist-DB have not yet been fixed, which is likely the result of the many open issues (over 400). Nevertheless, the developers from eXist-DB confirmed the bugs quickly and also expressed appreciation towards the bug reports "thank you for finding and reporting."[4][5] Because the reported bugs remained unfixed for over two months, we stopped testing and reporting to eXist-DB after reporting the first few found inconsistencies due to the difficulties of filtering out duplicate bugs. We believe that XPress has the ability to find more bugs in eXist-DB after the known bugs are resolved. Similarly, for the commercial DBMS, since the developers did not follow up on the bugs that we reported, we stopped testing this DBMS. For Saxon, all four bugs reported were resolved quickly within one week’s time. "thank you for finding and reporting." Selected bugs. Below, we give a few selected examples of bugs found by XPress to illustrate its bug-finding capability. Selected bugs. Incorrect optimization of comparison conditions. Figure 4 shows a fixed bug that we reported to BaseX.[6] The XPath expression selects all T nodes with attribute @t that satisfies @t >= 0 or @t <= 1. When @t exists and is a numeric value, this is a condition that always evaluates to true. Therefore, an optimization in BaseX rewrote the predicate to true. However, when @t does not exist for node T, @t evaluates to an empty sequence and returns false for both @t >= 0 and @t <= 1. Before we reported this bug, this case was overlooked and resulted in an incorrect optimization. Incorrect optimization of comparison conditions. Arithmetic overflow in pre-check conditions. Figure 5 shows a fixed bug that we reported to BaseX.[7] last() and position() returns the context size and the context position from the dynamic context respectively. In the context XML document, the prefix //S selects only one node, and therefore both last() and position() return 1. Therefore, the condition is true and node S should be selected. In BaseX, an empty result set was returned. The problem was related to optimization for positional arguments in conditional comparisons. BaseX substituted last() with the greatest theoretical last() value and checked if the condition could evaluate to true. If not, the condition could not be satisfied regardless of the actual context and could be rewritten to false to reduce context analysis. When calculating the multiplication, as the theoretical maximum value for last() is a big integer, calculating the expression with long instead of double caused an overflow and produced the incorrect result. Arithmetic overflow in pre-check conditions. Result of tail after subsequence off by one. Figure 6 shows a fixed bug that we reported to eXist-DB.[8] 1 to 2 creates an integer sequence consisting of 1 and 2. The subsequence() function in this example selects two elements starting from index 1, and the tail() function returns a new sequence excluding the first element of the input sequence. The correct result is to return 2. Unexpectedly, eXist-DB returned an empty result set. This was caused by a mistake when processing a call to tail that has a call to subsequence as an argument, which incorrectly reduced the ending index by 1. Result of tail after subsequence off by one. Incorrect reduce in positional expressions. Figure 7 shows a fixed bug that we reported to Saxon.[9] The dot (.) stands for the current context in XPath expressions. For node B, (., .)/parent::* selects the single node A as the parent. Therefore, last() = 1 and the condition evaluates to false. Saxon unexpectedly returned the node B. The = operator is considered to be an unordered operator, which does not require operands to be sorted. In Saxon, an optimization was applied to eschew removing duplicate nodes when evaluating the sub-expression, which resulted in A being selected twice and last() evaluated to 2. After we found and reported the bug, a patch was applied by the developers to remove the duplicates, when the left operand of = is positional sensitive. Incorrect reduce in positional expressions. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv [3] https://www.mail-archive.com/basex-talk@mailman.uni-konstanz.de/msg15173. html [4] https://www.mail-archive.com/basex-talk@mailman.uni-konstanz.de/msg15204. html [5] https://github.com/eXist-db/exist/issues/4830 [6] https://github.com/BaseXdb/basex/issues/2190 [7] https://github.com/BaseXdb/basex/issues/2220 [8] https://github.com/eXist-db/exist/issues/4830 [9] https://saxonica.plan.io/issues/6093?pn=1#change-24136 Authors:
(1) Shuxin Li, Southern University of Science and Technology China and Work done during an internship at the National University of Singapore (shuxin.li.lv@gmail.com);
(2) Manuel Rigger, National University of Singapore Singapore (rigger@nus.edu.sg). Authors: Authors: (1) Shuxin Li, Southern University of Science and Technology China and Work done during an internship at the National University of Singapore (shuxin.li.lv@gmail.com); (2) Manuel Rigger, National University of Singapore Singapore (rigger@nus.edu.sg).

Showing XPress’ Effectiveness Through the Number of Bugs Found, Developer Feedback, etc.

Table of Links Abstract and 1 Introduction Abstract and 1 Introduction 2 Background 2 Background 3 Approach and 3.1 Differential Testing for XML Processors 3 Approach and 3.1 Differential Testing for XML Processors 3.2 XPath Expression Generation 3.2 XPath Expression Generation 3.3 XML Generation 3.3 XML Generation 4 Evaluation 4 Evaluation 4.1 Effectiveness 4.1 Effectiveness 4.2 Efficiency 4.2 Efficiency 4.3 Comparison to the State of the Art 4.3 Comparison to the State of the Art 4.4 Analysis of BaseX Historical Bug Reports 4.4 Analysis of BaseX Historical Bug Reports 5 Related Work 5 Related Work 6 Conclusion, Acknowledgments, and References 6 Conclusion, Acknowledgments, and References 4 EVALUATION In the evaluation, we sought to investigate whether our technique is effective and efficient in finding bugs for XPath expression processors. Specifically, we were interested in the following questions: Q1. Is XPress effective in finding new XPath-related bugs in established XML processors (see Section 4.1)? Q1. Q2. Does the query generation approach described in Section 3.2 improve the bug-finding efficiency of XPress with respect to real-world baselines and a random generation approach (see Section 4.2)? Q2. Q3. How does the differential testing test oracle compare to the state-of-the-art oracle (see Section 4.3)? Q3. Q4. What kind of XPath-related bugs might be overlooked by XPress (see Section 4.4)? Q4. Tested XML Processors. We tested our method on six mature, well-known, and actively maintained XPath processors: BaseX, exist-DB, Saxon-HE, PostgreSQL, libXML2, and a commercial DBMS, whose name we have omitted due to its “DeWitt clause” [24]. We started testing on BaseX 10.4, eXist-DB 6.2.0, Saxon Home Edition 12.2, PostgreSQL version 15, and libXML2 commit version 106153. As bugs were resolved, we constantly updated to the latest available version. We selected BaseX, eXist-DB, and Saxon to be our main testing targets, because they all implement the more recent XPath 3.0 standard. BaseX ranks as the most popular Native XML DBMS on the DB-Engines Ranking [9]. eXist-DB is widely applied in data centers, systems, and platforms, as referenced on the eXist-DB reference page [11]. Saxon is an in-memory processor and therefore is not included in the DB-Engines rankings. However, the official website of Saxon [16] states: "More than 170 software vendors have built Saxon into their own applications" and "6 of the world’s top 10 software vendors are Saxonica clients", demonstrating that Saxon is a widely-used and popular XML processor. For XPath 1.0 standard implementations, we tested PostgreSQL, libXML2, and the commercial DBMS. PostgreSQL is a popular open-source DBMS, which ranks 4 on the DB-Engines ranking and has 12.8k stars on GitHub. libXML2 is a software library developed for the GNOME project. The commercial DBMS is often considered the most popular and important DBMS overall, as also reflected in various rankings. All XML processors have been actively maintained for over 15 years. Tested XML Processors. Experimental setup. We implemented the tool, XPress, in around 8,000 LOC in Java. In our experiments, we configured it to generate XML documents that contain 1 to 50 nodes. We create half as many node templates as element nodes. For each XML document, we generated 200 XPath expressions. Each XPath expression had an equal possibility to hold 1 to 7 sections. We set one predicate to hold at most 10 subjects (see Section 3.2) and the depth of the predicate tree to be at most 10. We used the default settings of each XML processor. We conducted all our experiments using a personal computer with a 64-Core AMD EPYC 7763 CPU at 2.45GHz and 512GB memory running Ubuntu 22.04. Experimental setup. This paper is available on arxiv under CC BY 4.0 DEED license. This paper is available on arxiv under CC BY 4.0 DEED license. available on arxiv Authors:
(1) Shuxin Li, Southern University of Science and Technology China and Work done during an internship at the National University of Singapore (shuxin.li.lv@gmail.com);
(2) Manuel Rigger, National University of Singapore Singapore (rigger@nus.edu.sg). Authors: Authors: (1) Shuxin Li, Southern University of Science and Technology China and Work done during an internship at the National University of Singapore (shuxin.li.lv@gmail.com); (2) Manuel Rigger, National University of Singapore Singapore (rigger@nus.edu.sg).

Is Our Technique Effective in Finding Bugs for XPath Expression Processors? An Investigation

Effective One-on-One Meetings: 5 Practices to Implement

AI Models Are Learning to Prioritize Their Thoughts—And It’s Wildly Effective

I’ve been running Sprint Reviews for the last 15 years, and I truly believe them to be useful when done well. When Sprint Reviews are not done well, however, they are at best a waste of time and at worst give you a reduction in morale. Pre-requisites For Sprint Retrospectives to be effective, you have to adhere to the basic rules of setting up a good scrum team: Scrum Teams should not be too big. No more than 6 people. If they are larger than that, split them. It’s harder to build team-wide trust with a larger team. A safe and open environment: You will get nothing out of a retrospective if people are afraid to speak their minds. Without real input, you will not get good actionable outcomes. Actioning: If you and your team do not put into effect the outcomes of your retros, people will lose faith and stop participating. Right. Now that we’ve got that out of the way, let’s look at the process. What is a Start, Stop, Continue Retrospective? As the name suggests, this approach focuses on three simple questions for the team: What should we start doing?


What should we stop doing?


What should we continue doing? This framework cuts through the noise and focuses directly on actionable changes. Let me break down how I run these effectively. The Process When team members suggest "start" items, they're looking at practices or behaviors we should add to our workflow. These might include: Getting customer feedback on features earlier
Setting up acceptance tests with stakeholders before coding
Implementing code reviews
Being punctual for standup meetings
Focusing on finishing stories before picking up new ones The "stop" list captures inefficiencies or time-wasters. Common examples I've seen teams identify: Committing code without ensuring tests pass


Letting daily scrums run longer than 15 minutes


Skipping refinement sessions when we feel behind schedule The "continue" section highlights positive practices that aren't yet habitual. Items from either the start or stop categories can move here once they're being implemented but need reinforcement. Once something becomes second nature to the team, we remove it from the list to keep it manageable. Mixing Up the Format To prevent “retrospective fatigue” (it’s real, believe me), I vary how we collect ideas: Sometimes, I'll simply open the floor for suggestions in any category


Other times, I'll go around the room, asking each person to contribute one item


When I notice specific issues, I might focus exclusively on "stop" items first


Occasionally, I'll combine approaches, going person-by-person for one category, then opening up for general discussion This flexibility keeps the format fresh even after many sprints with the same team. Making Decisions When idea generation slows down, it's time to vote. I typically use multi-voting - giving each team member three votes they can use as they wish. I prefer multi-voting because many retrospective actions don't actually require additional time - they're behavioral shifts. For example, being punctual for standups doesn't consume time; it saves it. Multi-voting allows us to select 2-3 key focus areas without overwhelming the team. We also review the "continue" list to see if any items have become habits or are no longer relevant. Setting Up for Next Time For the following retrospective, I bring a large sheet with all the previous ideas - both chosen and not chosen. I don't make a big deal about it - I simply tape it to the wall as a reference point. Then we begin a fresh start, stop, and continue the discussion. Why This Approach Works In my experience across various companies and team structures, this method is efficient, accessible, and productive. It's action-oriented rather than feelings-focused. Compounding is well known in the financial arena but is very rarely mentioned when it comes to team improvement. Team Retrospectives allow you to make little changes over a long period of time. If you compare the morale and ability to deliver in a team over the course of a year, with the right retrospectives in place and happening regularly, I would expect to see a large and positive change. While understanding emotions can be valuable in some contexts, this approach gets straight to behavioral changes. Every item generated leads directly to an action: starting something new, stopping something ineffective, or continuing something valuable until it becomes ingrained. For teams that value practical outcomes over extended discussion, the start, stop, and continue framework delivers consistently strong results. I’ve been running Sprint Reviews for the last 15 years, and I truly believe them to be useful when done well. When Sprint Reviews are not done well, however, they are at best a waste of time and at worst give you a reduction in morale. Pre-requisites For Sprint Retrospectives to be effective, you have to adhere to the basic rules of setting up a good scrum team: Scrum Teams should not be too big. No more than 6 people. If they are larger than that, split them. It’s harder to build team-wide trust with a larger team. Scrum Teams should not be too big. No more than 6 people. If they are larger than that, split them . It’s harder to build team-wide trust with a larger team. split them A safe and open environment: You will get nothing out of a retrospective if people are afraid to speak their minds. Without real input, you will not get good actionable outcomes. A safe and open environment: You will get nothing out of a retrospective if people are afraid to speak their minds. Without real input, you will not get good actionable outcomes. Actioning: If you and your team do not put into effect the outcomes of your retros, people will lose faith and stop participating. Actioning: If you and your team do not put into effect the outcomes of your retros, people will lose faith and stop participating. Actioning: If you and your team do not put into effect the outcomes of your retros, people will lose faith and stop participating. Right. Now that we’ve got that out of the way, let’s look at the process. What is a Start, Stop, Continue Retrospective? As the name suggests, this approach focuses on three simple questions for the team: What should we start doing? What should we stop doing? What should we continue doing? What should we start doing? What should we start doing? start What should we stop doing? What should we stop doing? stop What should we continue doing? What should we continue doing? continue This framework cuts through the noise and focuses directly on actionable changes. Let me break down how I run these effectively. The Process When team members suggest "start" items, they're looking at practices or behaviors we should add to our workflow. These might include: Getting customer feedback on features earlier Setting up acceptance tests with stakeholders before coding Implementing code reviews Being punctual for standup meetings Focusing on finishing stories before picking up new ones Getting customer feedback on features earlier Setting up acceptance tests with stakeholders before coding Implementing code reviews Being punctual for standup meetings Focusing on finishing stories before picking up new ones The " stop " list captures inefficiencies or time-wasters. Common examples I've seen teams identify: stop Committing code without ensuring tests pass Letting daily scrums run longer than 15 minutes Skipping refinement sessions when we feel behind schedule Committing code without ensuring tests pass Committing code without ensuring tests pass Letting daily scrums run longer than 15 minutes Letting daily scrums run longer than 15 minutes Skipping refinement sessions when we feel behind schedule Skipping refinement sessions when we feel behind schedule The "continue" section highlights positive practices that aren't yet habitual. Items from either the start or stop categories can move here once they're being implemented but need reinforcement. Once something becomes second nature to the team, we remove it from the list to keep it manageable. Mixing Up the Format To prevent “retrospective fatigue” (it’s real, believe me), I vary how we collect ideas: “retrospective fatigue” Sometimes, I'll simply open the floor for suggestions in any category Other times, I'll go around the room, asking each person to contribute one item When I notice specific issues, I might focus exclusively on "stop" items first Occasionally, I'll combine approaches, going person-by-person for one category, then opening up for general discussion Sometimes, I'll simply open the floor for suggestions in any category Sometimes, I'll simply open the floor for suggestions in any category Other times, I'll go around the room, asking each person to contribute one item Other times, I'll go around the room, asking each person to contribute one item When I notice specific issues, I might focus exclusively on "stop" items first When I notice specific issues, I might focus exclusively on " stop " items first stop Occasionally, I'll combine approaches, going person-by-person for one category, then opening up for general discussion Occasionally, I'll combine approaches, going person-by-person for one category, then opening up for general discussion This flexibility keeps the format fresh even after many sprints with the same team. Making Decisions When idea generation slows down, it's time to vote. I typically use multi-voting - giving each team member three votes they can use as they wish. I prefer multi-voting because many retrospective actions don't actually require additional time - they're behavioral shifts. For example, being punctual for standups doesn't consume time; it saves it. Multi-voting allows us to select 2-3 key focus areas without overwhelming the team. We also review the "continue" list to see if any items have become habits or are no longer relevant. Setting Up for Next Time For the following retrospective, I bring a large sheet with all the previous ideas - both chosen and not chosen. I don't make a big deal about it - I simply tape it to the wall as a reference point. Then we begin a fresh start , stop , and continue the discussion. start stop continue Why This Approach Works In my experience across various companies and team structures, this method is efficient, accessible, and productive. It's action-oriented rather than feelings-focused. Compounding is well known in the financial arena but is very rarely mentioned when it comes to team improvement. Team Retrospectives allow you to make little changes over a long period of time. If you compare the morale and ability to deliver in a team over the course of a year, with the right retrospectives in place and happening regularly, I would expect to see a large and positive change. Compounding While understanding emotions can be valuable in some contexts, this approach gets straight to behavioral changes. Every item generated leads directly to an action: starting something new, stopping something ineffective, or continuing something valuable until it becomes ingrained. For teams that value practical outcomes over extended discussion, the start, stop, and continue framework delivers consistently strong results.

Stories, News & Mentions about EFFECT

Effect’s Stories on HackerNoon

Effect's News & Mentions