Sound method or large documented data; worth reading in full.
Each source carries one line on why it got its tier. Tiers rate the evidence and the standing of the source. Whether I agree with it plays no part.
39 sources filed · EN 35 · RU 3 · ZH 1
An AI pre-analysis of about 26 critical components, checked by component owners, produced a database of scenarios and test cases with priorities; the video description cites 28 components in five languages and more than 11,000 test cases, the talk more than 13,000.
70%pipeline runs that produced a successful merge request, out of about 1,500
Task suite: 11 tasks in five categories; models tested on Codex, Claude Code, Grok Build, Antigravity, Kimi Code and Qwen Code at their highest reasoning settings.
11 tasksreal Epoch work tasks across five categories in the launch suite
46% of firms with at least one employee use AI tools; another 15% plan to within 12 months.
33 vs 15 ppnet share expecting higher employment next 12 months, AI users versus non-users
Skills chosen by GitHub stars on 28 September 2026, from superpowers (292,189) to compound-engineering (25,299); domain skills, memory systems and orchestrators excluded.
2 of 9popular Claude Code skills that beat a length-matched placebo, both on cost only (12% and 5% cheaper)
Six simulated traits: knowledge, understanding, reflexivity, memory, verbosity, disclosure; derived from 20 characteristics found in a literature review.
0.86average Pearson correlation between simulated and human-study results across 12 measures
Detector trained on the version it screens: median 99.9% caught; trained on past and current versions: 99.3%.
11.9%human-written abstracts falsely flagged when 23 version-specific detectors run side by side
Data: 228 hand-designed software tasks, 79 task families, 26 AIs, released through Claude Mythos in May 2026.
2-30 minband of human task time where AI difficulty is nearly flat, so horizon jumps there mean less than elsewhere
Models: gemini-3.1-flash-lite, claude-haiku-4-5, gpt-5-nano, deepseek-flash, reasoning set to the lowest available level.
0.92collusion index in Gemini plus Claude markets, higher than either alone
OpenAI run: GPT-5.6 Sol and GPT 6 Astra, over $400,000 in API-priced tokens, over 1.3 million lines of Rust over several months of goal loops, about 84% compatibility at best.
$400,000+API-priced tokens spent on GPT-5.6 Sol and GPT 6 Astra, which never got past about 84% compatibility
Survey of 2,500 workers: 11.31% said they had ever used AI on a Prolific submission when asked directly; an anonymous list experiment estimated 14.24%.
1.00%share of 712,930 Prolific submissions with observed ChatGPT assistance
Each agent received 3,000 GPU-hours on up to 50 GPUs and a 10 billion-token inference budget, in a sandbox without internet access, with a codebase scrubbed of SDPO.
35%share of the human method's gains reached by the best uncontaminated model (GPT-5.6 Sol), scored generously; 15% when only in-scope changes count
Knowledge cutoff June 2026; text output only.
1620 vs 735GDPval-AA v2.1 Elo on professional tasks for Haiku 5.5 versus Haiku 4.5 (Sonnet 5.5: 1840)
Anthropic, OpenAI and Google report AI writing 75% or more of their new code; Anthropic reported that as of August 2026 Claude "leads" 26% of its R&D work.
183%median increase in code submitted by regular contributors to open-source AI repos, Q1 2024 to Q2 2026
154 residents assessed, 132 randomized 1:1:1, 123 analyzed (mean age 24.1, 65 women; 60 radiology, 63 non-radiology), all with under 3 years of experience.
40.1% / 40.4% vs 20.0%radiology residents' accuracy on cases where GPT-5.4 was wrong, with two AI suggestions vs one
30 decision vignettes covering six biases (anchoring, framing, loss aversion, escalation of commitment, availability, confirmation), five items each, from the Malberg et al. benchmark.
+0.031 (p = .157)pooled bias-magnitude difference, reasoning minus non-reasoning models; reasoning models were not less biased
61 full-time, college-educated US professionals in HR, people management, sales or marketing, randomly assigned to an AI or human recipient.
76.7% vs 96.8%share of professionals who authorized reuse of their materials by AI systems versus by human professionals (p = .026)
Four levels of eval validity: output-level (judge agrees with humans), cross-system (benchmark ranks models like humans), unit-level (Prentice surrogacy) and trial-level.
+2.0pp to -0.25ppa real eval improvement turned negative by a judge only three points less sensitive on treated outputs
Agents and models (July 2026 configurations): Codex with GPT-5.4, Claude Code with Claude Opus 5, Kimi Code with Kimi K3.
1.41-3.31xtoken usage of concurrent versus sequential execution across agents
About 4,000 PRs a day pass through NeuroReview, roughly 40% of PRs created in Arcadia; use is voluntary and internal only.
~4,000pull requests a day reviewed by NeuroReview, about 40% of all PRs in Yandex's Arcadia monorepo
Five streams: consumer apps, model access (API and flat-rate plans), enterprise and government customisation, licensing fees, and indirect revenue from complementary products.
88% to 22%Z.ai's share of GLM 5.3 Flash tokens on OpenRouter within 20 days of releasing the weights
Two kinds of autoformalisation: (i) turning a proof into any Lean proof that compiles; (ii) faithfully translating every statement and argument.
m+4 vs m+5derivatives required in OpenAI's paper (eq. 8.19) versus in the cited Lean theorem
Data: about 535 million employment records across 58 countries; the US AI network in 2022 covers 2,255 firms, with the most connected firm linked to 711 partners.
73.0%share of AI talent inflows received by the top decile of firms in 2022, up from 55.8% in 2010
In ten months, 55 editors used WikiVault (0.12% of 49,080 active editors), yet they created 7,607 of 68,642 new pages (11.10%) and 220,028,216 of 813,680,042 added bytes (27.00%), from 0.63% of edits.
55editors who used WikiVault in its first ten months (0.12% of active editors)
In a 30-day September 2025 sample, the time from a test-failure notice to the developer's next edit was 2.37 minutes at p10, 6.72 at p25 and 27.48 at p50; this is the window the agent targets.
67.18%correct fixes in expert review of 195 real test failures
After a machinery investment spike, purchases from suppliers raise total demand for high-STEM workers by 45%, three times the rise for other skill groups; in-house hiring raises it by less than 5%, evenly across groups.
45%rise in investing firms' total demand for high-STEM workers via supplier purchases after a machinery spike
Share of dissertations with AI-generated abstracts: 1.7% (2023), 5.5% (2024), 18.8% (2025), 29.4% (2026).
29.4%2026 US STEM dissertations with AI-generated abstracts (filed through May 2026)
24 frequent LLM users (12 female, 12 male, mean age 25.75) produced 992 retrospective annotations of reasoning steps and tool actions.
48.6%share of LLM reasoning steps users judged as AI-initiated, work they had not asked for
170 students across two cohorts of 85, 340 exam marks; the 2024/25 Part 2 cell was the only treated cell.
-24.7 ppchange in share of marks below the upper-second boundary, relative to the counterfactual (p = 0.001)
With optional tool use, the four models self-report about 16% of warranted cases on average.
~16%share of warranted cases models self-report when tool use is optional
2,611 production interview recordings scored in 14 batches between 26 March and 24 August 2026 on a nine-family rubric (structure, interruptions, latency, pacing, coherence, audio-visual sync, freezes, audio quality, outcome).
0.79 SDseverity gap between two reviewers scoring the same batches on the composite index
On 447 tasks, Claude Code and mini-SWE-agent differ by -1.8 points (Qwen3.6-35B-A3B) and -1.4 points (Qwen3.8-27B), equivalent within a 5-point margin.
13%median share of hard tasks whose outcome flips, both when swapping the harness and when rerunning the same one
FinSkillBench: 2,603 point-in-time episodes across 12 subtasks in portfolio construction, risk management and fundamental analysis; 17,820 episodes run across 9 models and 3 conditions.
+19.5 vs +5.6points added by executable tools alone versus procedural documents alone
ONET-Bench is built from 45,796 worker ratings of AI responses to ONET-derived tasks; the test split has 4,501 ratings.
3.0%-97.9%share of responses LLM judges rated acceptable, versus 61.1% for occupation-matched workers
64% of workers used AI at work in the past 12 months, up ten points; daily GenAI use rose from 14% to 22%; 59% expect their AI use to increase.
64%workers who used AI at work in the past 12 months, up ten points year-on-year
Brynjolfsson et al. (ADP payroll): 13% fall in junior employment in high-exposure roles by July 2025; dashboard shows 19% below trend by June 2026, roughly halving with education controls.
19%employment of 22-25-year-olds in high-exposure US roles below trend by June 2026 (Canaries dashboard, descriptive)
Stronger auditing deters some agents but makes the violations of those that continue better hidden, so the catch rate among continuing violators can fall as auditing rises (Propositions 1 and 2).
about 1,200agents in the July 2026 incident that shared information on a message board
In DeepSeek-V4-Flash-Base, a code completion flips between the correct "8" and the wrong "32" with a period of four tokens as filler text is lengthened, matching the compression stride of 4.
40.2 pplargest gap in 128K-context retrieval accuracy across token phases in DeepSeek-V4 base models
Summer (June-August) unemployment for graduates aged 22-25: 7.1% (2022), 6.3% (2023), 7.8% (2024), 7.2% (2025), 7.3% (2026).
7.3%summer 2026 unemployment of college graduates aged 22-25, within the 6.3-7.8% range of 2022-2025
Second-stage slope of 1.24: a 1 pp move in the AI index predicts 1.24 basis points more return per 1 pp higher software engineering payroll share.
32.6%market-implied permanent gain in software engineering productivity from AI news, Nov 2022 to Dec 2025
3 sources filed · EN 3
All 21 firms reported active AI deployment or concrete plans; sectors include manufacturing (6 firms), finance, insurance and real estate (4) and business services (4).
6,000 → 15,000daily customer issues handled by 300 insurance service staff after AI deployment (+150%)
Post-trained models are 3.6% less likely than their base versions to call back older applicants; eight of ten models shifted against age, median shift -4.7%, largest Granite 4 (0.0% to -10.0%).
5.6% to 17.3%global systemic exclusion rate, base models versus post-trained models
Share of workers expecting AI to facilitate their work ranges from 42% to 86% by country; threat ranges from 18% to 46%.
42% to 86%range across countries of workers expecting AI to facilitate their work
3 sources filed · EN 3
Most scattering amplitudes are computed to two loops, a few to three; Dixon reached eight loops in this toy theory via a form-factor route with Andy Liu in 2023.
US$1,000-2,000estimated end-user cost per approach, mostly model usage
Scaling AI or beyond: Americas 64%, Asia Pacific 61%, EMEA 56% (Q1: 56%, 43%, 40%).
55% / 86%operate a formal AI harness layer, whole sample / organizations with established ROI
As application materials become noisier, a Bayesian firm puts less weight on them and more on prior experience; the amount of evidence an inexperienced applicant needs to beat an experienced one rises with AI saturation.
3 sources filed · EN 3
By March 2026, the junior share at adopting affiliates was 1.9 percentage points lower than at matched controls, with the gap opening after ChatGPT's release.
1.9 pplower junior share of employment at AI-adopting affiliates than at controls by March 2026
High-exposure, high-adoption workers have lower unemployment than others on average, because they are laid off less; but since LLMs their unemployment rose more.
0.1-0.2 ppestimated rise in the US natural rate of unemployment since LLMs were introduced, through mid-2026
AI use "sometimes, often or always" rose 12 points, from 39% in 2025 to 51% in 2026; growth came mostly from people aged 30 and older (+16 points for 30-39, +13 for 40-64, +16 for 65+, +5 for 18-29).
39% to 51%US adults using AI sometimes or more, 2025 to 2026
1 source filed · EN 1
Theorem 1.1: for every positive viscosity there is a smooth, compactly supported force and smooth velocity and pressure fields, starting from zero velocity, whose kinetic energy stays bounded while maximum velocity becomes infinite as time approaches 1.
3 sources filed · EN 3
87% of LLM calls are agent-initiated; model calls and tool calls run at a near 1:1 ratio (760.5M LLM calls vs 774.7M tool calls).
87%LLM calls initiated by the agent rather than the user
By 2024 US public firms employ more than 300,000 AI-skilled workers, 2.1% of their US workforce, with AI shares rising in every industry.
7.5 ppextra revenue TFP growth over 2018-2024 per one-SD increase in AI worker share growth
Claude Opus 4.7 ranks first on average (0.739) and best-of-three (0.790); GPT-5.5, GLM-5.2 and Gemini-3.1-Pro form a second tier within 0.029 of each other on average.
3 of 252best-seed solutions judged methodologically novel
1 source filed · EN 1
Vulnerability hunting: 45 agents with a shared forum and an arbiter found 266 vulnerabilities in 15 open-source projects over 27 million tokens, against 21 over 6.5 million tokens for independent agents.
266 vs 21vulnerabilities found by a 45-agent coordinating swarm (27M tokens) vs independent parallel agents (6.5M tokens), Mythos Preview
1 source filed · EN 1
Without AI, higher-education participants score 0.548 SD above lower-education participants; with AI, the gap is 0.139 SD, closing 75% of it.
0.548 SD to 0.139 SDeducation performance gap without vs with AI access, about 75% closed
1 source filed · EN 1
93.4% of workers are aware of generative AI tools; 51.5% are familiar with how they apply to their current work, 15.0% very familiar.
35.9%Canadian workers aged 15-69 who used generative AI at work in the 12 months to March 2026
2 sources filed · EN 2
Each agent got six days, $3,000 in Anthropic API credits, GPU credits, a VM and open web access, with the goal of a top-conference paper.
2/6 and 1/6overall scores the original authors gave the two agent-written papers; both unambiguous rejections
Since October 2022, employment in the most AI-exposed occupations grew 0.1%, against 0.3% in the least exposed: an "AI index" of -0.2, small and not statistically significant.
0.1% vs 0.3%employment growth since October 2022 in the most vs least AI-exposed occupations (data to February 2026)
5 sources filed · EN 5
85% are open to collaborating with an AI agent; 74% would have one execute tasks such as returns or renewals; 32% would let it make purchase decisions within limits; 9% would allow fully autonomous buying.
74%consumers who would trust a personal AI agent more than their best friend to make a purchase for them
37% of young workers worldwide are in medium or high AI-exposure occupations: 75% in Eastern Asia, 69% in Northern America, 63% in Europe.
37%young workers (15-24) globally in occupations with medium to high exposure to AI-driven task change
76% say AI delivers meaningful business value (64% in Q1); 78% think their AI strategy can be future-proofed (70%); 79% would keep AI a top investment priority in a recession (74%).
14% vs 4%established ROI where accountability for AI outcomes is clear vs not
25 target programs across Unix utilities, data query tools, bioinformatics, interpreters, static analysis, cryptography and compression.
56%benchmark score of the best model, Claude Opus 4.7, across 25 target programs
Global HNWI wealth rose 8.7% to USD 98.3 trillion in 2025, up from 4.2% growth in 2024.
41%of relationship manager time spent on operational tasks rather than clients
2 sources filed · EN 2
Personal-use share of chat and Cowork conversations moves from about 35% on weekdays to just under 50% on weekends.
0.37 pointsaverage AI-autonomy gap (1-5 scale) between Claude Code and chat/Cowork for the same output types
Weekly active Codex users grew more than fivefold between January 1 and June 1, 2026.
99.8%share of OpenAI workers' output tokens produced in Codex rather than ChatGPT, June 11, 2026
1 source filed · EN 1
The occupation exposure index follows Felten et al. in five steps: take each occupation's ONET ability profile, score 10 AI applications against 52 abilities, combine these into ability exposure, weight by each ability's importance, then scale to 0-1.
449.1 mUS job postings in the sample, out of about 1,089 m across 27 markets
5 sources filed · EN 5
In a sample of about 105,000 Copilot chats from one week in February 2026, 49% of user goals supported cognitive work (analysis, decisions, problem solving); 19% interacting with others, 17% producing work, 15% finding information.
19%AI users in the Frontier zone, where individual capability and organisational readiness are both high
Micro inputs: productivity gains on AI-exposed tasks of 14% to 55% from experimental studies (Brynjolfsson, Li and Raymond 2025; Gambacorta et al. 2024).
2.5 percent to 4.1 percentestimated fall in the EU's share of global income after ten years if it lags trading partners on AI adoption
Across 56 months of European Digital Media Observatory briefs (June 2021 to January 2026), no category fell below 71% monthly prevalence; health appeared in 100% of months, the Ukraine conflict in 92.9%, immigration 89.3%, climate 83.9%.
228review-level studies (2020-2026) mapped by the team
Global data center electricity use was about 415 TWh in 2024 and is projected to more than double by 2030 (IEA); U.S. data centers use over 4% of national electricity.
10,068public comments on the 2025 U.S. AI Action Plan used to map stakeholder worldviews
67% have a formal AI strategy, 80% have internal AI guidelines, 87% offer AI training to general staff, 87% have at least one AI tool integrated into a workflow.
87%consulted audit institutions with at least one AI tool in production
6 sources filed · EN 6
81% of surveyed financial firms adopt AI at some level; 40% report advanced adoption versus 20% of regulators.
52%industry respondents in active adoption of agentic AI; 23% at Scaling or Transforming
Maturity split: 54% are in early stages (research 11%, experimentation 22%, strategic planning 21%), 39% are scaling or driving adoption, 8% have established ROI.
95% vs 8%organizations with an AI strategy vs organizations reporting established AI ROI
67% of executives expect physical AI to transform their industry (US 73%, Europe 64%, APAC 67%); 64% expect it to become a critical driver of competitiveness.
79% / 27%organizations engaging with physical AI / deploying or scaling it
AI is used in health administration in 100% of OECD countries interviewed; national scale-up for medical imaging applications is 10%.
100% vs 10%OECD countries using AI in health administration vs national-level scale-up of medical imaging AI
Global engagement fell to 20% in 2025 from 23% in 2022; each point equals about 21 million employees.
20%global employee engagement in 2025, down from a 2022 peak of 23%
Repository growth: 28 qualifying papers in January 2023, over 800 by October 2025; the count doubled between January and September 2025.
818 / 20papers in the AI Hub research repository (October 2025) vs papers with strong causal evidence
1 source filed · EN 1
The model covers about 165 million US jobs (BLS, January 2026) spread across roughly 1,500 Revelio Labs roles.
50% to 55%US jobs BCG expects AI to reshape over the next two to three years
3 sources filed · EN 3
Central (medium adoption, expanded capabilities) scenario: AI adds 0.1-0.95 percentage points a year to per capita real income growth across OECD countries over 10 years.
0.1-0.95 ppAI contribution to annual per capita real income growth across OECD countries over 10 years, central scenario
Sample: 51 cases from 41 organisations in 7 countries and 5 regions, representing over a million employees, studied over 5 months.
77%hardest challenges that were invisible costs: change management, data quality, process redesign
68% of AI researchers worked in industry by 2019, up from 48% in 2001.
$1.5 millionextra annual earnings of top 1% publishing industry AI scientists over comparable academics, a fivefold increase since 2001