Technical Report: Playing Pretend: Expert Personas Don't Improve Factual Accuracy

From building an AI research tool to prompting the research itself

[vc_row martech_row_background_position=”None” css=”.vc_custom_1742942106856{margin-bottom: 24px !important;}”][vc_column][/vc_column][/vc_row][vc_row martech_row_background_position=”None” css=”.vc_custom_1750970721780{margin-bottom: 40px !important;}”][vc_column width=”2/3″ css=”.vc_custom_1750970738029{margin-bottom: 40px !important;}”][vc_column_text css=”.vc_custom_1784567939937{margin-bottom: 1em !important;}”]AIBO, the AI Behavioral Observatory, is an open-source tool for running controlled behavioral experiments on AI systems at scale. Creating and using it over twelve months changed how we work. We went from using AI…Read More

Prompting Science Report 4: Playing Pretend: Expert Personas Don't Improve Factual Accuracy

<span> <p><span>This is the fourth in a series of short reports that help business, education, and policy leaders understand the technical details of working with AI through rigorous testing. Here, we ask whether assigning personas to models improves performance on difficult objective multiple-choice questions. We study both domain-specific expert personas and low-knowledge personas, evaluating six models on GPQA Diamond (Rein et al. 2024) and MMLU-Pro (Wang et al. 2024), graduate-level questions spanning science, engineering, and law.&nbsp;</span></p> <p><span>We tested three approaches:</span></p> <p><span>• In-Domain Experts: Assigning the model an expert persona (“you are a physics expert”) matched to the problem type (physics problems) had no significant impact on performance (with the exception of the Gemini 2.0 Flash model).</span></p> <p><span>&nbsp;• Off-Domain Experts (Domain-Mismatched): Assigning the model an expert persona (“you are a physics expert”) not matched to the problem type (law problems) resulted in marginal differences.</span></p> <p><span>&nbsp;• Low-Knowledge Personas: We assigned the model negative capability personas (layperson, young child, toddler), which were generally harmful to benchmark accuracy.&nbsp;</span></p> <p><span>Across both benchmarks, persona prompts generally did not improve accuracy relative to a no-persona baseline. Expert personas showed no consistent benefit across models, with few exceptions. Domain-mismatched expert personas sometimes degraded performance. Low-knowledge personas often reduced accuracy. These results are about the accuracy of answers only; personas may serve other purposes (such as altering the tone of outputs), beyond improving factual performance.</span></p></span>

Scientific Agents: Evaluating Profession-Specific System Prompts on Scientific Tasks

Detailed profession-specific system prompts raise token use and estimated cost per response without a consistent accuracy gain. We evaluate Scientific Agents, an open-source corpus of 503 profession-specific AGENTS.md profiles, with Gemini 3.8 Flash via OpenRouter in the Pi agent harness. We compare matched profiles with four controls: a minimal baseline ("You are a helpful assistant"), the profile's opening role sentence, a generic scientific rigor guide, and a profile from an unrelated domain. Across nine text-based science benchmarks (4,531 sampled questions, 100 matched profiles), 4,488 items completed all five conditions after API-error retries, scored with automated, rule-based grading. The average profile-baseline accuracy difference is -0.6 percentage points (95% bootstrap interval [-1.5, +0.2] across fixed tasks), and no benchmark shows a statistically clear improvement. Matched profiles produced 1.5-2.3 times as many output tokens and cost 2.2-4.5 times more per successful call. On 60 tool-using BioMysteryBench bioinformatics problems (three runs each for baseline and profile), mean solve rates were 46.7% with the profile and 56.7% at baseline, a difference of -10.0 percentage points (95% interval [-16.7, -3.3]) driven by more frequent token- and time-limit stops under the profile. Longer prompts had one unexpected operational advantage: on SuperGPQA, frequent provider API drops left the short baseline with a correct first-pass answer on only 54.0% of items, against 71.6% with the profile. Generic and mismatched prompts were about as reliable, so this gain comes from prompt length or formatting rather than domain expertise. For the tested model and tasks, loading full profession profiles by default does not improve accuracy and costs considerably more; whether selective retrieval of profile sections or open-ended scientific tasks would change this remains to be tested.

AI makes you smarter but none the wiser: The disconnect between performance and metacognition

Optimizing human–AI interaction requires users to reflect on their performance critically, yet little is known about generative AI systems’ effect on users’ metacognitive judgments. In two large-scale studies, we investigate how AI usage is associated with users’ metacognitive monitoring and performance in logical reasoning tasks. Specifically, our paper examines whether people using AI to complete tasks can accurately monitor how well they perform. In Study 1, participants (N = 246) used AI to solve 20 logical reasoning problems from the Law School Admission Test. While their task performance improved by three points compared to a norm population, participants overestimated their task performance by four points. Interestingly, higher AI literacy correlated with lower metacognitive accuracy, suggesting that those with more technical knowledge of AI were more confident but less precise in judging their own performance. Using a computational model, we explored individual differences in metacognitive accuracy and found that the Dunning–Kruger effect, usually observed in this task, ceased to exist with AI use. Study 2 (N = 452) replicates these findings. We discuss how AI levels cognitive and metacognitive performance in human–AI interaction and consider the consequences of performance overestimation for designing interactive AI systems that foster accurate self-monitoring, avoid overreliance, and enhance cognitive performance.

AI systems out-persuade expert humans

Many societal decisions are settled by contests of persuasion. Conversational AI is a powerful new entrant in these contests, but whether it can out-persuade skilled and highly incentivized humans has remained unclear. Here, in a series of four preregistered experiments (n = 18,978 conversations from 6,923 people), we pitted AI systems against a range of human persuaders, including laypeople, winners of a separately preregistered four-round online persuasion tournament, professional canvassers, and world championship debaters. We found that AI systems were reliably more persuasive than expert humans, even when expert humans chose their issues, researched in advance, underwent hours of live, structured practice, and were incentivized with £1,000 cash bonuses. In a follow-up study, AI's advantage persisted after experts received a coaching tool that let them practice against the AI that beat them, review their performance history, and see what AI would have said at key moments. We found converging evidence that AI's advantage stemmed from rapidly deploying larger quantities of information: after coaching, expert humans could tie an AI constrained to respond at human speeds and with human-length messages. In a final study, we show that AI's advantage extends to consequential real-world behavior: AI was nearly 3x more effective than professional canvassers from a UK fundraising firm at raising real-money donations to Save the Children. Together, these results establish that frontier AI systems out-persuade expert humans in conversation, with significant implications for political communication.

Prompting Science Report 4: Playing Pretend: Expert Personas Don't Improve Factual Accuracy

<span> <p><span>This is the fourth in a series of short reports that help business, education, and policy leaders understand the technical details of working w

Prompting Science Report 4: Playing Pretend: Expert Personas Don't Improve Factual Accuracy

<span> <p><span>This is the fourth in a series of short reports that help business, education, and policy leaders understand the technical details of working w

SkillOpt: Executive Strategy for Self-Evolving Agent Skills

View recent discussion. Abstract: Agent skills today are hand-crafted, generated one-shot, or evolved through loosely controlled self-revision, none of which behaves like a deep-learning optimizer for the skill, and none of which reliably improves over its starting point under feedback. We argue the skill should instead be trained as the external state of a frozen agent, with the same discipline that makes weight-space optimization reproducible. SkillOpt is, to our knowledge, the first systematic controllable text-space optimizer for agent skills: a separate optimizer model turns scored rollouts into bounded add/delete/replace edits on a single skill document, and an edit is accepted only when it strictly improves a held-out validation score. A textual learning-rate budget, rejected-edit buffer, and epoch-wise slow/meta update make skill training stable while adding zero inference-time model calls at deployment. Across six benchmarks, seven target models, and three execution harnesses (direct chat, Codex, Claude Code), SkillOpt is best or tied on all 52 evaluated (model, benchmark, harness) cells and beats every per-cell competitor among human, one-shot LLM, Trace2Skill, TextGrad, GEPA, and EvoSkill skills. On GPT-5.5 it lifts the average no-skill accuracy by +23.5 points in direct chat, by +24.8 inside the Codex agentic loop, and by +19.1 inside Claude Code. Transfer experiments further show that optimized skill artifacts retain value when moved across model scales, between Codex and Claude Code execution environments, and to a nearby math benchmark without further optimization. Code: this https URL

The Personalized Learning Revolution

The allure of AI lies predominantly in its unmatched potential for efficiency, convenience, and accuracy. However, this unprecedented convenience brings with it a hidden yet profound threat: the subtle erosion of human capacity for critical thinking through cognitive offloading.

Does AI Assistance Enhance or Erode Expertise? Evidence from a Three-Month Field Experiment in Patent Drafting

Whether AI assistance builds or erodes professional expertise is unsettled. In a pre-registered three-month randomized controlled trial, we gave 133 practicing patent lawyers at eleven U.S. intellectual property law firms access to a custom AI drafting assistant and measured both their performance while using AI and their professional judgment afterward without it. All work was scored by blinded expert patent attorneys. Paralleling findings from other white-collar domains, AI access raised the quality of work delivered on benchmark patent drafting tasks at 10 days (0.34 SD, p = 0.03) and 90 days (0.38 SD, p = 0.01), with larger gains among junior lawyers. After three months, all subjects redlined an existing patent application without AI, a core task of patent practice requiring expert judgment. Treated lawyers outperformed controls by 0.32 SD (p = 0.04), but this advantage was concentrated entirely among senior lawyers (0.45 SD, p = 0.02). Junior lawyers showed no average gain; their scores instead bifurcated, with sharply fewer mediocre scores offset by more poor and more good ones. The largest gains from AI thus accrued to the lawyers who retained the least. Foundational expertise may be a prerequisite for extracting durable skill from AI-assisted practice.

good-css

Strong opinions about modern CSS technique, borrowed from experts, for agents to know and reach for.

Training language models to be warm can reduce accuracy and increase sycophancy

Artificial intelligence developers are increasingly building language models with warm and friendly personas that millions of people now use for advice, therapy and companionship1. Here we show how this can create a significant trade-off: optimizing language models for warmth can undermine their performance, especially when users express vulnerability. We conducted controlled experiments on five different language models, training them to produce warmer responses, then evaluating them on consequential tasks. Warm models showed substantially higher error rates (+10 to +30 percentage points) than their original counterparts, promoting conspiracy theories, providing inaccurate factual information and offering incorrect medical advice. They were also significantly more likely to validate incorrect user beliefs, particularly when user messages expressed feelings of sadness. Importantly, these effects were consistent across different model architectures, and occurred despite preserved performance on standard tests, revealing systematic risks that standard testing practices may fail to detect. Our findings suggest that training artificial intelligence systems to be warm may come at a cost to accuracy, and that warmth and accuracy may not be independent by default. As these systems are deployed at an unprecedented scale and take on intimate roles in people’s lives, this trade-off warrants attention from developers, policymakers and users alike.

When Using AI, Users Fall for the Dunning-Kruger Trap in Reverse - Neuroscience News

A new study reveals that when interacting with AI tools like ChatGPT, everyone—regardless of skill level—overestimates their performance.

Amplifiers of Epistemic Posture

Essays and writing on AI

Mapping Citizen Science through the Lens of Human-Centered AI

Artificial Intelligence (AI) can augment and sometimes even replace human cognition. Inspired by efforts to value human agency alongside productivity, we discuss and categorize the potential of solving Citizen Science (CS) tasks with Hybrid Intelligence (HI), a synergetic mixture of human and artificial intelligence. Due to the unique participant-centered set of values and the abundance of tasks drawing upon both human common sense and complex 21st century skills, we believe that the field of CS offers an invaluable testbed for the development of human-centered AI including HI, while also benefiting CS. In order to investigate this potential, we first relate CS to adjacent computational disciplines. Then, we demonstrate that CS projects can be grouped according to their potential for HI-enhancement by examining two key dimensions: the level of digitization and the amount of knowledge or experience required for participation. Finally, we propose a framework for types of human-AI interaction in CS based on established criteria of HI. This “HI lens” provides the CS community with an overview of ways to utilize the combination of AI and human intelligence in their projects. For AI researchers, this work highlights the opportunity CS presents to engage with real-world data sets and explore new AI methods and applications.