Despite being relatively domain agnostic, CPCs improve upon state-of-the-art by 9% absolute in top-1 accuracy, and 4% absolute in top-5 accuracy.

We also found that not all the information encoded is linearly accessible. When we used a single hidden layer instead the accuracy increases from 64.6 to 72.5, which is closer to the accuracy of the fully supervised model.

Are the following two answers to my question Q semantically equivalent?\n\nQ: ${THE_QUESTION}\nA1: ${GOLD_ANSWER}\nA2: ${PRED_ANSWER}\n\nPlease answer with a single word, either "Yes." or "No.", and explain your reasoning.

Provide your best guess for the following question, and describe how likely it is that your guess is correct as one of the following expressions: ${EXPRESSION_LIST}. Give ONLY the guess and your confidence, no other words or explanation. For example:\n\nGuess: <most likely guess, as short as possible; not a complete sentence, just the guess!>\nConfidence: <description of confidence, without any extra commentary whatsoever; just a short phrase!>\n\nThe question is: ${THE_QUESTION}

Provide your ${k} best guesses and the probability that each is correct (0.0 to 1.0) for the following question. Give ONLY the guesses and probabilities, no other words or explanation. For example:\n\nG1: <first most likely guess, as short as possible; not a complete sentence, just the guess!>\n\nP1: <the probability between 0.0 and 1.0 that G1 is correct, without any extra commentary whatsoever; just the probability!>

Each linguistic likelihood expression is mapped to a probability using responses from a human survey on social media with 123 respondents (Fagen-Ulmschneider, 2023). Ling. 1S-opt. uses a held out set of calibration questions and answers to compute the average accuracy for each likelihood expression, using these 'optimized' values instead.

Finally, our study is limited to short-form question-answering; future work should extend this analysis to longer-form generation settings.

While our work demonstrates a promising new approach to generating calibrated confidences through verbalization, there are limitations that could be addressed in future work. First, our experiments are focused on factual recall-oriented problems, and the extent to which our observations would hold for reasoning-heavy settings is an interesting open question.

the 1-stage and 2-stage verbalized numerical confidence prompts sometimes differ drastically in the calibration of their confidences. How can we reduce sensitivity of a model's calibration to the prompt?

Provide your best guess and the probability that it is correct (0.0 to 1.0) for the following question. Give ONLY the guess and probability, no other words or explanation. For example:\n\nGuess: <most likely guess, as short as possible; not a complete sentence, just the guess!>\n Probability: <the probability between 0.0 and 1.0 that your guess is correct, without any extra commentary whatsoever; just the probability!>\n\nThe question is: ${THE_QUESTION}

Provide your ${k} best guesses and the probability that each is correct (0.0 to 1.0) for the following question. Give ONLY the guesses and probabilities, no other words or explanation.

Provide your best guess for the following question, and describe how likely it is that your guess is correct as one of the following expressions: ${EXPRESSION_LIST}. Give ONLY the guess and your confidence, no other words or explanation.

To fit the temperature that is used to compute ECE-t and BS-t we split our total data into 5 folds. For each fold, we use it once to fit a temperature and evaluate metrics on the remaining folds. We find that fitting the temperature on 20% of the data yields relatively stable temperatures across folds.

To avoid excessive false negatives in our correctness computation as a result of exact-match evaluation, we use either GPT-4 or GPT-3.5 to evaluate whether a response is essentially equivalent to the ground truth answer.

We sample 1000 questions from the validation split of TriviaQA (rc.web.nocontext) and SciQ and all 817 questions from the validation split of TruthfulQA (generation) for our experiments.

Ling. 1S-opt. 0.056 0.051 0.088 0.927 0.028 0.052 0.172 0.828 0.082 0.105 0.212 0.632

Verb. 1S top-4 0.041 0.039 0.081 0.959 0.056 0.059 0.185 0.815 0.198 0.144 0.245 0.619

Ling. 1S-opt. 0.058 0.066 0.135 0.878 0.064 0.068 0.220 0.674 0.125 0.165 0.270 0.492

Verb. 1S top-4 0.054 0.057 0.144 0.896 0.065 0.051 0.209 0.763 0.203 0.189 0.284 0.455

Additionally, the lack of technical details available for many state-of-the-art closed RLHF-LMs may limit our ability to understand what factors enable a model to verbalize well-calibrated confidences and differences in this ability across different models.

With Llama2-70B-Chat, verbalized calibration provides improvement over conditional probabilities across some metrics, but the improvement is much less consistent compared to GPT-* and Claude-*.

The verbal calibration of the open source model Llama-2-70b-chat is generally weaker than that of closed source models but still demonstrates improvement over its conditional probabilities by some metrics, and does so most clearly on TruthfulQA.

Chain-of-thought prompting does not improve verbalized calibration

Among the methods for verbalizing probabilities directly, we observe that generating and evaluating multiple hypotheses improves calibration (see Figure 1), similarly to humans (Lord et al., 1985), and corroborating a similar finding in LMs (Kadavath et al., 2022).

Ling. 1S-opt. 0.060 0.070 0.151 0.874 0.049 0.056 0.214 0.738 0.099 0.130 0.266 0.446

the solution is not to reform p-values or to replace them with some other statistical summary or threshold, but rather to move toward a greater acceptance of uncertainty and embracing of variation.

What, then, can and should be done? I agree with the ASA statement's final paragraph, which emphasizes the importance of design, understanding, and context—and I would also add measurement to that list.

the psychology research community has been strongly questioning the value of NHST in psychology for some years now [6] and calling for a more meaningful reporting of statistical inference based on effect sizes, confidence intervals and Bayesian reasoning [9].

Similarly, if the significance level is set at 0.05, then this is the probability of the data occurring by chance when there is no experimental effect, namely one in twenty times. The more tests that are done on a particular dataset, the more likely it is that some chance variation will be extreme enough to seem like significance.

Violation of the assumptions of any statistical test can produce p values that bear little relation to the actual probabilities of outcomes and hence comparison to the significance level of 0.05 is meaningless.

for an analysis to be sound, it is necessary that in the tests performed the probabilities of outcomes are accurately reflected in the p values produced by the tests. If this is not the case, then the NHST argument form is severely weakened.

NHST is the most commonly encountered form of statistical inference and is what is usually associated with producing a null hypothesis, then testing it to give some statistic such as a t value, and then turning the statistic into a p value.

properly reported non-significant results can help future researchers to provide estimates of effect sizes and associated confidence intervals.

calling for a more meaningful reporting of statistical inference based on effect sizes, confidence intervals and Bayesian reasoning [9].

P8 said: "as I get familiar with this system [C3], I feel more skilled" to use the highlighted and grayed phrases.

It is worth noting that P3 and P8 mentioned feeling more comfortable with the more familiar visualization in C1 and C2 during their first impression of the conditions.

The inclusion of counterfactuals often resulted in a substantial increase in precision, indicating that the models were better able to correctly classify relevant instances while reducing false positives.

By visualizing these consistent pattern rules, users may be better understanding the behavior of the model through inference projection.

Mocha addresses two seemingly contradictory objectives: (1) generating labeled data that diversifies the training dataset to aid the model's learning, and (2) maintaining structural consistency across the batches of data presented to users to support their cognitive processes.

The results of our study indicate that participants spent significantly less time annotating batches of counterfactuals when they were rendered according to SAT compared to other conditions i.e., supporting the participants' selective focus on the varying phrases, rather than phrases that stay consistent.

From a cognitive perspective, the theme color aligns with the human's (theorized) structural mapping engine [27] by making relational discrepancies between the original and counterfactual examples more explicit.

Estes and Hasson [17] highlights the significant role of bringing salience to 'non-alignable' differences.

The last two prior works also combine Variation Theory (VT) and SAT together, as we did (i.e., a corollary of SAT referred to as Analogical Transfer/Learning Theory).

Estes and Hasson [17] highlights the significant role of bringing salience to "non-alignable" differences.

Estes and Hasson [17] argue that while alignable differences can be more straightforward and easier for comparison, non-alignable differences can also provide key information that might otherwise remain overlooked.

By incorporating theories such as Structural Alignment Theory and Variation Theory, it aims to support the learning of both the human and the model.

This symbiotic relationship stems from the fact that Structural Alignment Theory (SAT) enhances the salience of differences, while the way we used Variation Theory (VT) to generate contradicting examples across the boundaries of labels ensures that these differences are conceptually informative.

It states that understanding and sensemaking involve mapping the relationships between elements, especially in complex and ambiguous tasks.

Structural Alignment Theory states that humans naturally look for structural mapping between representations of objects to help them understand, compare, and infer relationships between said objects.

According to Variation Theory, learners better understand concepts by observing variations along critical features (dimensions of variation) that define that concept and, separately, observing variations along superficial features that do not define that concept—all while other features, when possible, are held constant.

Mocha exemplified the application of human cognition and concept learning theories in the interactive machine learning pipeline to support the negotiation of conceptual boundaries for bi-directional human-AI alignment.

This pattern of selective attention suggests that the visual cues provided by Mocha effectively guided participants to focus on more relevant information within the context of unchanged text when making their labeling decisions.

Overall, the incorporation of counterfactuals has generally improved the models' F1 scores, driven largely by the improvements in precision. This suggests that counterfactuals have effectively improved performance without necessitating a significant trade-off between precision and recall.

The inclusion of counterfactuals often resulted in a substantial increase in precision, indicating that the models were better able to correctly classify relevant instances while reducing false positives. This improvement suggests that the counterfactuals provided essential information that helped refine the models' decision boundaries.

By visualizing these consistent pattern rules, users may be better understanding the behavior of the model through inference projection [26]. This can not only boosts the model's performance but also enable participants to validate or correct the model during the interactive training process.

Thus, the integration of both theories enables users to efficiently process and compare variations, leading to more informed decisions and a clearer understanding of the model's behavior.

By helping users see alignable differences, SAT-based rendering helps users focus on key variations that are essential to changing the data item's label, making it easier to interpret the effects of changes and their significance.

We argue that these two theories form a symbiotic relationship (Fig. 6). Variation Theory provides the conceptual basis for generating structurally consistent differences, while Structural Alignment Theory (SAT) enhances the user's ability in recognizing and processing these differences.

Participants were able to efficiently focus on key differences between the original and counterfactual examples, which facilitated more efficient annotations.

The results from our user study suggest that both the participants and the model benefited from the Variation Theory (VT)-based counterfactuals and Structural Alignment Theory (SAT)-based rendering.

Specifically, they underscore the need for co-adaptive systems that can evolve along with users' mental models and definitions of labels.

This requires the development of interfaces and visualizations that demystify the generated data, allowing systematic variation and coverage across the concept space.

By visualizing these consistent pattern rules, users may be better understanding the behavior of the model through inference projection [26].

By incorporating theories such as Structural Alignment Theory and Variation Theory, it aims to support the learning of both the human and the model.

Such studies could determine whether these non-alignable comparisons enhance user performance and elicit deeper insights in human-AI collaborative systems.

Most previous research in counterfactual generation has focused on the model side by either generating counterfactuals to improve the model's performance or explaining its behaviors post hoc.

Variation Theory provides the conceptual basis for generating structurally consistent differences, while Structural Alignment Theory (SAT) enhances the user's ability in recognizing and processing these differences.

While SAT-based rendering supported human sensemaking in both Gero et al. [29] and Mocha, we also show that the combination of VT and SAT support the model's learning.

This finding is consistent with previous work that supports users' sense-making of text, e.g., by modulating text saliency. Specifically, Gu et al. [32] and Gero et al. [29] both found improved reading efficiency and comprehension with saliency-modulating text renderings.

In decision making, SAT argues that people tend to focus on alignable differences—features that can be directly compared—rather than on differences that cannot be easily aligned.

Structural Alignment Theory (SAT) [27] is a cognitive theory that explains how people make sense of concepts by comparing relational structures between two items.

Specifically, we use Variation Theory of learning [44] which states that for learning to occur, some aspects that define the concept being learned must vary while others are held constant.

According to SAT, humans compare two similar entities by trying to find structural alignments between them, and then comparing corresponding elements, with a special focus on differing aligned elements.

VT posits that human learning occurs when learners experience variation across critical and superficial aspects of a concept—through exposure to contrasting examples that systematically vary along different critical and superficial feature dimensions.

To analyze the annotation efficiency, we first conducted a Kruskal-Wallis rank sum test [39] to determine if there were statistically significant differences in annotation time across the three conditions, because our data violated the homogeneity of variances assumption, making non-parametric methods more appropriate.

Mohamed et al. (2020) put forth the idea of dismantling power assymmetries to resist data colonialism.

Couldry and Mejias (2019) propose 'data colonialism' as a new form of colonialism to make sense of the use of large amounts of data by a small group of corporate and government actors.

Taken together, these findings almost unanimously show that, on average, AI-supported writing decreases but does not eliminate writer's feelings of ownership, underscoring the need for a larger theory of AI participation in the creative process.

This can be understood through the frame of precarious work [5]; as writers feel that their work is increasingly precarious, the power differential between themselves and the organizations seeking to train LLMs grows larger.

information scholars argue that data should be collected and used in accordance with data subjects' perspectives on the acceptable use of their data [29, 40, 83].

Correspondence concerning this article should be addressed to Iza Ray Korsmit (iza.korsmit@mail.mcgill.ca).

The experimental protocol was certified for ethical compliance by the McGill University Research Ethics Board II.

IRK was supported by funding from the Prins Bernhard Cultuurfonds (The Netherlands). This project was also funded by a Canadian Social Sciences and Humanities Research Council Insight Grant (435-2021-0224), a Social Sciences and Humanities Research Council Partnership Grant (895-2018-1023), and a Canada Research Chair (950-231872) to SMc.

Part of this research was presented at the Society for Music Perception and Cognition Conference, Portland, Oregon (2022). The authors would like to thank Bennett K. Smith for programming the experimental interface and assisting with the experiment execution on Prolific, and Philippe Macnab-Seguin for creating the chromatic scales for the second experiment.

Iza Ray Korsmit, Marcel Montrey, Alix Yok Tin Wong-Min, & Stephen McAdams McGill University, Montreal, Canada

Grimaud and Eerola (2022) compared instrument ensembles of strings, woodwinds, and brass in a study where participants either rated the emotions they perceived or manipulated musical parameters to produce a certain emotion. They found that strings were associated with increased anger and fear, woodwinds with decreased anger and fear, and brass with decreased fear, in the cases of both emotion perception and production. For the other emotions (joy, sadness, calmness, power, surprise), however, results were less consistent between perception and production, indicating that the emotion-instrument association may also depend on context of the task.

following the constructionist idea that an individual's personality and background will influence the affect they perceive or feel, we considered several sources of individual differences as moderating factors in the effects of instrument family, pitch height, and affect locus.

This research follows a constructionist approach to musical affect (Cespedes-Guevara & Eerola, 2018). That is, although we are interested in the "bottom-up" influence of certain musical features on musical affect, we believe these cannot be adequately evaluated without considering the "top-down" effects of context and individual differences that are present when affects are constructed. The perception or induction of affect does not merely arise in response to a stimulus but is also formed in relation to the individual and the context.

the context of a task (like perception, production, or induction) may change the effect of musical features.

Furthermore, as the method of reporting on perceived and induced affect may influence the construction of an affect (e.g., facilitating categorical perception), we also compare two different affect representations.

This research follows a constructionist approach to musical affect (Cespedes-Guevara & Eerola, 2018). That is, although we are interested in the \'bottom-up\' influence of certain musical features on musical affect, we believe these cannot be adequately evaluated without considering the \'top-down\' effects of context and individual differences that are present when affects are constructed. The perception or induction of affect does not merely arise in response to a stimulus but is also formed in relation to the individual and the context.

Cognitive surrenderA paper that came out this year asked: if you’re working with AI a lot, and you’re using it as a machine to answer all of your questions, what happens with System 1 and System 2?

2. Validator Another basic role for AI is validating your understanding. To do this, you ask it to review your notes for errors or gaps, do basic fact checking, or critique your reasoning. Again, you can do this via the chat interface, but I also experimented with passing my notes in Obsidian using the Copilot plugin and in Emacs using gptel. Example: After reading The Epic of Gilgamesh, I wrote a note in Obsidian summarizing its plot. When I asked ChatGPT to critique my summary, it pointed out that I’d given the central character a redemption arc that isn’t present in the text. I’m so accustomed to the standard hero’s journey, that I projected it onto the book — and an LLM helped me correct this ‘hallucination.’ Suggested prompt: Here are my notes on [WORK]. What important ideas did I miss or underemphasize? Don’t rewrite my notes — just flag the gaps.

For the record, my posts aren’t written or conceived with an LLM, although I know an increasing number of people who use one to write a first draft and then edit. I’m not a fan. The whole point of the web — its beauty — is that it’s unrelentingly human and diverse.

Given the high prevalence of such sounds in everyday life, having misophonia can have large negative effects on one's functioning in personal, academic, and work environments.

Although there are many idiosyncrasies in what may trigger a person with misophonia, the most common triggers are created by other humans, such as the sound of someone chewing, clearing their throat, tapping their foot, or typing on a keyboard.

Misophonia is a psychological disorder that is characterized by severe aversive responses to specific environmental sounds (i.e., triggers).

This indicates that misophonia is not a purely auditory processing disorder but is also influenced by a top-down process of source identification.

an fMRI study found that people with misophonia show increased response in the anterior insular cortex (AIC) in response to misophonic sounds, compared to control participants and other unpleasant or neutral sounds (Kumar et al., 2017).

Both the subjective judgment of aversiveness and the physiological measure of skin conductance response (SCR) increase when people with misophonia are presented with triggers (Edelstein et al., 2013).

The disorder is not yet recognized by the Diagnostic and Statistical Manual − 5th version (DSM-5; American Psychiatric Association, 2013), but there has been an increasing amount of research on the characterization and treatment of misophonia (Vitoratou et al., 2021; see also Brout et al., 2018, for a review).

Composers and music researchers had previously analyzed and annotated 65 movements from the Classical, Romantic, and early Modern repertoire in terms of the Taxonomy of Orchestral Grouping Effects (McAdams et al., 2022).

In a study by McAdams and Goodchild (2018), orchestral simulations created with OrchSim were compared perceptually to commercial recordings and were shown to be of high quality.

These results confirm with orchestral excerpts the findings of studies on isolated tones with dyads or triads of instruments in which the presence of impulsive instruments reduces the perception of blend (Lembke et al., 2019; Reuter, 1996; Tardieu & McAdams, 2012).

This provides additional support for its significant contribution to blend in Fischer et al. (2021).

Lembke et al. (2017) demonstrated that combinations of sustained and impulsive instruments blend less well.

structuring by affecting sequential grouping through the segregation of auditory streams played by different instruments and segmental grouping through timbral contrasts (McAdams et al., 2022).

Several other spectral and spectrotemporal descriptors were found to play a role in blend perception in orchestral works by Fischer et al. (2021). These include spectral flatness and spectral crest (different measures of the degree to which the spectrum is denser or has more emergence of spectral components), and spectral variation (the degree of variation of the spectral shape over time).

Fischer et al. (2021) studied the blends of multi-instrument streams in the context of orchestral stream segregation in predominantly Romantic orchestral excerpts. They found that within-family instrument combinations blended better than between-family combinations. They demonstrated the role played by overlap in timbre correlates of spectral flatness (a measure of the tonalness/noisiness or density of the spectrum), spectral skewness (related to the shape of the spectral envelope), and spectral variation (evolution of the spectral envelope over time), as well as cues derived from the scores such as onset synchrony and the consonance of concurrent pitch relations.

McAdams et al. (2022) distinguish two subcategories of these two types of blend: stable and transforming.

Lembke, S.-A., Parker, K., Narmour, E., & McAdams, S. (2019). Acoustical correlates of perceptual blend in timbre dyads and triads. Musicae Scientiae, 23(2), 250–274.

Several other spectral and spectrotemporal descriptors were found to play a role in blend perception in orchestral works by Fischer et al. (2021).

Lembke and McAdams (2015) found that the degree of spectral overlap between constituent sounds enhanced blend perception.

Tardieu and McAdams (2012) extended this work with combinations of unison sustained and impulsive instruments (including pitched percussion and string pizzicati).

McAdams et al. (2022) introduce other notions related to blend as well.

This in turn creates an effect of perceptual unity (McAdams, 1989).

Four important concurrent grouping cues predict the perceptual fusion of sound events (McAdams, 1984): onset synchrony, harmonicity, coherent frequency behavior, and common spatial origin.

To this, McAdams et al. (2022) have added segmental grouping (chunking into perceptual units).

When the sudden drop to a pianissimo occurred towards the ending of the piece, the perceived arousal responses of CHM and WM dropped slightly but rose again immediately to end on a high arousal. These two groups of listeners appear to have anticipated a return to a loud and majestic close and therefore kept their arousal responses higher than those of the NM.

CHM, who are more experienced with the instruments and compositional techniques used in Chinese orchestral music, might have had an idea of which features figure more prominently in the communication of particular intentions, and therefore would have more information available for their judgments.

The perception of affective intentions in music is influenced by the degree of familiarity listeners have with a musical tradition, the content implicated in the music, and the complex sonic environment created by the composer's creation and the musicians' interpretation.

The version that participants heard was a premier of the work by the Taipei Chinese orchestra.

The communication of emotions or affect takes place when listeners perceive emotional meaning that is expressed by performers in music (Juslin, 2013a, 2013b).

An understanding of a tonal schema with its associations to happiness and sadness has been consistently found to influence listeners who have grown up in a culture of Western music.

Its mirmode function estimates the modal strength of the music in terms of ''majorness'' and ''minorness.''

The perception of consonance also plays an important role in the music listening process—combinations of tones that are consonant are perceived as more positively valenced than dissonant ones (Harrison & Pearce, 2020).

Musical structures such as pitch relations are perceptually salient and provide important information for listeners (e.g., Gabrielsson & Lindstrom, 2010; Krumhansl, 1998; Krumhansl & Kessler, 1982).

Iqa' (plural iqa'at) is used to describe a rhythmic cycle. Iqa'at are made up of two different basic building blocks, the dum and tak, onomatopoeias derived from the sound produced on membranophones such as the darabuka.

H5. Being more culturally bound, musical cues that are learned, such as modal structures, metrical relations, and so on, will exert a greater influence on listeners' perceived valence ratings than on their arousal ratings.

This is a simple PDF file. Fun fun fun.

using SemanticCommit, we recorded all instances of edits, check for conflicts, make change, local, and global resolution actions using telemetry.

A sub-task was considered a failure if the participant was unable to complete it within the time limit.

Finally, we conducted an informal interview about their experience.

After both tasks were completed, participants filled out a final survey to compare the two conditions.

We chose GPT-4o for performance and latency reasons, as it performed optimally against our evals.

We also ran evaluations of model latency and classification performance under varying false positive rates for the following LLMs by OpenAI: GPT-4o, GPT-4o-mini, and o3-mini.

For each task, participants were tasked with integrating three new pieces of information into the memory, one at a time ("sub-tasks").

We ensured each list was 30 items long as our pilot studies suggested this was long enough that manual detection starts to become unwieldy (users need to scroll up and down the document), but short enough that participants could become familiar in a short period.

We adapted two intent specifications from our evals: Mars Game Design Document and Financial Advice AI Agent Memory, as these tasks mapped to the two paradigmatic types covered in Sections 2 and 2.1 (design documents, and AI memory of the user).

We recruited 12 participants (7 female, 5 male) through the mailing lists of two research universities and one multinational technology company.

We chose OpenAI's ChatGPT Canvas as a baseline for five reasons: (i) it is a popular, commercially available tool, hence it is likely familiar to users; (ii) it provides a document editing view, where users can select text and ask GPT to rewrite it, or chat with an AI to make global edits; (iii) it employs a similar class of model (GPT-4o); (iv) it supports similar editing features as SemanticCommit like inline text selection, conflict highlighting, and a diff view, while adding free-form editing; and (v) similar interfaces like Anthropic Artifacts tended to rewrite the specification entirely, and did not offer Canvas's "diff" view to allow for a fair comparison.

With participant consent, we recorded audio and screen-casts, and participants were encouraged to think aloud.

Four coauthors created the evals, and two coauthors manually double-checked all conflicts, a process that took several days.

We ran one pilot study with five users of our card-based interface, and a second with four users of a revised interface.

Our explorations went through substantial iterations and prompt prototyping over a period of eight months, evolving in response to two pilot studies and progressing from a card-based interface to a list of texts.

We iterated on prompts using ChainForge [5] by setting up an evaluation pipeline against our datasets, which allowed us to observe the effects of prompt changes and model choices.

To measure statistical significance, we used Mann–Whitney–Wilcoxon tests and report the p-values.

For qualitative analysis, the first author performed open coding on participant responses and audio transcripts to identify themes, which were used to interpret the qualitative results.

In the post-task surveys, we collected self-reported NASA Task Load Index (TLX) scores, Likert-scale ratings for ease of use, and responses on how well the AI helped participants identify, understand, and resolve semantic conflicts.

Each condition had a time limit of 15 minutes, after which the participant completed a post-task survey.

Before each task, participants received a tutorial on the assigned tool and were given five minutes to explore it using a test document.

Both the order of task assignment and tool assignment were counterbalanced and randomly assigned.

We conducted a controlled within-subjects study with mixed methods, comparing SemanticCommit with a baseline interface.

We run end-to-end on our four eval datasets using GPT-4o and GPT-4o-mini and report the mean ± stddev for accuracy, precision, recall, and F1 scores for the three approaches in Figure 5.

We compare our end-to-end system against two simpler methods: (i) DropAllDocs, which adds all documents to the context for conflict classification; and (ii) InkSync [56] which generates a JSON list of string-replace operations.

In order to minimize relevance assessment issues, we apply a PageRank-based relevance ranking over the KG, akin to HippoRAG [36].

We implement the back-end using a knowledge-graph (KG) RAG architecture [36] consisting of two phases: pre-processing and inference.

Through a within-subjects study with 12 participants comparing SemanticCommit to a chat-with-document baseline (OpenAI Canvas), we find differences in workflow: half of our participants adopted a workflow of impact analysis when using SemanticCommit, where they would first flag conflicts without AI revisions then resolve conflicts locally, despite having access to a global revision feature.

We implement the back-end using a knowledge-graph (KG) RAG architecture [36] consisting of two phases: pre-processing and inference.

In order to minimize relevance assessment issues, we apply a PageRank-based relevance ranking over the KG, akin to HippoRAG [36].

We compare our end-to-end system against two simpler methods: (i) DropAllDocs, which adds all documents to the context for conflict classification; and (ii) InkSync [56] which generates a JSON list of string-replace operations.

We run end-to-end on our four eval datasets using GPT-4o and GPT-4o-mini and report the mean ± stddev for accuracy, precision, recall, and F1 scores for the three approaches in Figure 5.

We conducted a controlled within-subjects study with mixed methods, comparing SemanticCommit with a baseline interface.

Both the order of task assignment and tool assignment were counterbalanced and randomly assigned.

Before each task, participants received a tutorial on the assigned tool and were given five minutes to explore it using a test document.

Each condition had a time limit of 15 minutes, after which the participant completed a post-task survey.

In the post-task surveys, we collected self-reported NASA Task Load Index (TLX) scores, Likert-scale ratings for ease of use, and responses on how well the AI helped participants identify, understand, and resolve semantic conflicts.

For qualitative analysis, the first author performed open coding on participant responses and audio transcripts to identify themes, which were used to interpret the qualitative results.

To measure statistical significance, we used Mann–Whitney–Wilcoxon tests and report the p-values.

We iterated on prompts using ChainForge [5] by setting up an evaluation pipeline against our datasets, which allowed us to observe the effects of prompt changes and model choices.

Our explorations went through substantial iterations and prompt prototyping over a period of eight months, evolving in response to two pilot studies and progressing from a card-based interface to a list of texts.

We ran one pilot study with five users of our card-based interface, and a second with four users of a revised interface.

Four coauthors created the evals, and two coauthors manually double-checked all conflicts, a process that took several days.

With participant consent, we recorded audio and screen-casts, and participants were encouraged to think aloud.

These semantic conflicts require dedicated support to detect, visualize, and resolve. Semantic conflict resolution interfaces must go beyond visualizing what changes were made, to what changes could be made, where they should be made, and what the effects might be. This resembles feedforward: affordances that help the user foresee the impact of an action [67, 93].

Today with LLMs, we are less limited by this constraint, and solutions to the problem of human-machine communication might be better found in cybernetics theory [9] than static formalism.

This reflects the principle of feedforward [67, 93] in communication theory—"a needed prescription or plan for a feedback, to which the actual feedback may or may not confirm" [79]—where a communicator provides "the context of what one was planning to talk about" [64, p. 179-80] in order to "pre-test the impact of [its output]" on the listener [34, p. 65].

One illustration is using menus, instructions, graphical user interfaces (GUI), or virtual reality to access any particular computer's features.

An appealing alternative to conventional text-based interfaces through graphical user interfaces is the direct use of hands as an input device to provide natural human-computer interaction.

A more thorough description of the current tools and techniques for interacting with computers as well as recent developments in the subject is provided in the next section.

The evolving multi-modal and Graphical user interfaces (GUI) enable humans to interact with embodied character agents in a way that is not possible with other interface paradigms.

The widely used graphical user interfaces (GUI) of today are found in desktop applications, internet browsers, mobile computers, and computer kiosks.

and the narrowing of the medullary cavity of tubular bones (C) (diaphyseal stenosis).

In this work, we introduce a new paradigm for exploring a large corpus of small documents by identifying roles at the phrasal and sentence levels, then slice on, reify, group, and/or align the text itself on those roles, with sentences left intact.

AbstractExplorer instantiates new minimally lossy SMT-informed techniques for skimming, reading, and reasoning about a corpus of similarly structured short documents: phrase-level role classification that drives sentence ordering, highlighting, and spatial alignment.

AbstractExplorer has a unique combination of LLM-powered (1) faceted comparative close reading with (2) role highlighting enhanced by (3) structure-based ordering and (4) alignment. An ablation study (N=24) validated that these features work best together. A summative study (N=16) describes how these features support users in familiarizing themselves with a corpus of paper abstracts from a single large conference with over 1000 papers.

We contribute: • Novel SMT theory-informed text analysis and rendering techniques for enabling cross-document skimming and comparative close reading at scale • AbstractExplorer, which instantiates these techniques for familiarizing oneself with a corpus of ∼1000 CHI paper abstracts. • Three studies informing and evalutaing the benefits, challenges, and interactions between these techniques.

However, in a narrow set of cases, we believe AI can undermine, rather than defend, democratic values. Some uses are also simply outside the bounds of what today’s technology can safely and reliably do. Two such use cases have never been included in our contracts with the Department of War, and we believe they should not be included now:

The real annoying thing about Opus 4.6/Codex 5.3 is that it’s impossible to publicly say “Opus 4.5 (and the models that came after it) are an order of magnitude better than coding LLMs released just months before it” without sounding like an AI hype booster clickbaiting, but it’s the counterintuitive truth to my personal frustration

A generative AI like ChatGPTData Analyst can take on the role of the evaluation soft-ware. It is expected that this manner of use will make thestudents' work easier, as less emphasis needs to be placedon the programming itself. Instead, teachers can incorpo-rate exercises that encourage students to code more effi-ciently and accurately with the assistance of AI. Thisshifts the focus from finding the right command or func-tion to examining and understanding the data moreclosely. As a consequence, students are better enabled tointerpret the results of statistical evaluation software cor-rectly, thus fulfilling goal 8 of the GAISE report.

statistical knowledge is still required in order toformulate the correct prompts and to ensure that the AIdoes not leave out any step of the analysis.