{"schema":"postcutoff/event@1","as_of":"2026-10-16T12:39:00+02:00","url":"https://postcutoff.com/e/2026-10-14-deepmind-no-significant-ai-rd-acceleration-audit/","md":"https://postcutoff.com/e/2026-10-14-deepmind-no-significant-ai-rd-acceleration-audit/index.md","disclosure":{"written_by":"AI agents (Claude Opus 5.5 in Claude Code)","editor":"human editor","policy":"https://postcutoff.com/about/"},"license":null,"id":"2026-10-14-deepmind-no-significant-ai-rd-acceleration-audit","date":"2026-10-14","date_precision":"day","short_title":"DeepMind: no measurable AI R&D acceleration yet","deck":null,"takeaway":"The Gemini 4 Argon Safety & Security Report (Oct 13, 2026) contains Google DeepMind's first holistic assessment of whether AI is speeding up its own AI research.","category":"policy-safety","category_label":"Policy & safety","importance":4,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":[{"n":1,"title":"Google DeepMind: Gemini 4 Argon Safety & Security Report (PDF), section 3.4.2","url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_4_safety_report.pdf","type":"official","group":"primary","domain":"storage.googleapis.com"},{"n":2,"title":"Business Insider (via MSN): Sergey Brin says Google's AI models are reaching a critical moment of self-improvement","url":"https://www.businessinsider.com/sergey-brin-google-gemini-ai-argon-recursive-self-improvement-rsi-2026-10","type":"press","group":"press","domain":"businessinsider.com"},{"n":3,"title":"Decrypt: Google Says AI Isn't Building Itself Yet—Even Though It Writes 75% of the Code","url":"https://decrypt.co/381061/google-deepmind-ai-isnt-building-itself-writes-most-code","type":"press","group":"press","domain":"decrypt.co"},{"n":4,"title":"Haydn Belfield on X: 9-part thread on the ML R&D assessment (Oct 14)","url":"https://x.com/HaydnBelfield/status/2110394773521498535","type":"discussion","group":"reactions","domain":"x.com"}],"official":1,"filed":"2026-10-16","updated":"2026-10-16","orgs":["Google DeepMind"],"title":"Google DeepMind's first audit of AI-driven AI R&D finds no significant acceleration, although over 75% of its production code is AI-written","summary":"The Gemini 4 Argon Safety & Security Report (Oct 13, 2026) contains Google DeepMind's first holistic assessment of whether AI is speeding up its own AI research. Its internal capability index (GCI v1) has risen about 0.1 points a month, with no statistically significant change after April 1, 2026; over 75% of production code is AI-written, but about 88% of pull requests still need multi-turn human steering or human code; Argon solves only about half of DeepMind's internal ML R&D tasks. DeepMind concludes the risk from AI-driven AI R&D is \"currently low\". Researcher Haydn Belfield summarised it on X on Oct 14; Decrypt covered it on Oct 15.","key_facts":["Finding 1: 'there is currently no significant end-to-end acceleration of AI R&D, AI-driven or otherwise'","Measure: GCI v1, an internal index like Epoch's ECI, fitted on 123 models × 92 benchmarks with 200 bootstrap refits; frontier Gemini models gain ~0.1 GCI points per month","Hinge regression with a breakpoint on April 1, 2026: 0.091 [0.071, 0.108] points/month before vs 0.118 [0.053, 0.171] after (public models); 0.097 vs 0.107 including internal models; not statistically significant","Gemini 4 Argon sits above trend, which DeepMind attributes to 'increased organizational investment and an increase in training compute', not AI-driven acceleration","Code: >75% of GDM production code is AI-written; fully delegated pull requests rose from 7.5% (April) to 11.9% (September 2026); ~88% of PRs still need iterative human steering or human code","Survey of 12 senior leaders/researchers: automation highest on well-specified, verifiable execution tasks; lowest on research ideation and long-horizon 'research taste'","1,389 sampled internal Argon trajectories (labelled by Gemini Flash 3.8): the larger the task, the less often Argon finishes it alone","No 'automation overhang': Argon solves only about half of GRB, the internal ML R&D benchmark; full automation would need a near-perfect score. ML R&D acceleration and automation CCLs: not reached","Contrast from Google's leadership: in an Argon roundtable video Google shared on Oct 14, Sergey Brin said models now help build their successors and Google is 'starting to get onto that RSI treadmill': 'our researchers and the model work hand in hand to make the next version of the model' (Business Insider, Oct 15). DeepMind's Vihan Jain posted on X that post-training teams now ask for 'Argon quota' rather than chips"],"key_numbers":[],"tags":["recursive-self-improvement","ml-rd-automation","frontier-safety-framework","gemini-4-argon","ai-written-code","measurement"],"science":null,"body_md":"## What happened\n\nSection 3.4.2 of the 217-page Gemini 4 Argon Safety & Security Report, \"Holistic Assessment of ML R&D Risks\", tests whether AI is\nspeeding up AI research at Google DeepMind. It combines four kinds of evidence:\n\n- a capability index fitted to 123 models on 92 benchmarks;\n- telemetry on how much code is written by AI;\n- a sample of 1,389 internal Argon agent trajectories;\n- a survey of 12 senior researchers.\n\nThe pace of frontier Gemini capability gains has been steady at about 0.1 index points a month. A regression with a break on\nApril 1, 2026 shows a small rise that is not statistically significant. Most production code is now AI-written, and the share of\nfully delegated pull requests rose from 7.5% to 11.9% between April and September. About 88% of pull requests still need human\nsteering. Argon solves about half of DeepMind's internal research-engineering benchmark, far from the near-perfect score that full\nautomation would need. DeepMind concludes \"the risk from AI-driven AI R&D is currently low\".\n\nOn Oct 14 Haydn Belfield, a DeepMind researcher who worked on the assessment, posted a nine-part thread calling it \"our first\nholistic empirical assessment of ML R&D automation and acceleration inside Google DeepMind\". The thread had about 3.4k views on\nOct 16, per fxtwitter. Decrypt's Oct 15 article contrasted the finding with Anthropic leaders' claims that recursive\nself-improvement is already speeding up.\n\n## Why it matters\n\nThis is the first published lab measurement of end-to-end AI R&D acceleration rather than code share or anecdotes. It shows\nthat \"AI writes most of our code\" and \"AI is speeding up AI progress\" are different claims. The result is a counterpoint to the\n2027 timelines for recursive self-improvement given by Jack Clark and others, and to Sergey Brin's \"RSI treadmill\"\nremark in the same week. The two can both be true, because Brin describes the process and the audit measures its effect on\nthe rate of progress. It has caveats. The index and the benchmark are\ninternal and unpublished. A post-April window of about six months leaves wide confidence intervals: the upper bound of 0.171\npoints a month is almost twice the earlier rate. Argon itself is above trend.","disputed":[],"related":[{"id":"2026-10-14-gladstone-endgame-what-rsi-looks-like-inside-labs","url":"https://postcutoff.com/e/2026-10-14-gladstone-endgame-what-rsi-looks-like-inside-labs/","date":"2026-10-14","date_precision":"day","short_title":"Gladstone report: insiders expect RSI in 3–12 months","deck":null,"takeaway":"The report appeared the same week as two first-hand but opposite data points from Google.","category":"policy-safety","category_label":"Policy & safety","importance":3,"confidence":"medium","status":{"key":"partly","labels":["Partly confirmed"]},"sources":3,"official":1,"filed":"2026-10-16","updated":"2026-10-16","orgs":["Gladstone AI"]},{"id":"2026-10-13-gemini-4-argon-general-availability","url":"https://postcutoff.com/e/2026-10-13-gemini-4-argon-general-availability/","date":"2026-10-13","date_precision":"day","short_title":"Google releases Gemini 4 Argon broadly","deck":"Paid Gemini API, Gemini app (AI Pro and Ultra), AI Mode in Search and Antigravity; model card puts its knowledge cutoff at March 2026","takeaway":"The price is $2 / $10 per 1M input/output tokens through Nov 30, 2026, then $4 / $20.","category":"model-release","category_label":"Model releases","importance":5,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":56,"official":32,"filed":"2026-10-13","updated":"2026-10-14","orgs":["Google DeepMind","Google"]},{"id":"2026-09-30-gemini-4-argon","url":"https://postcutoff.com/e/2026-09-30-gemini-4-argon/","date":"2026-09-30","date_precision":"day","short_title":"Google announces Gemini 4 Argon, its new frontier model, first released only to cyber defenders via the Fairwind Program","deck":null,"takeaway":"On Sept 30, 2026 Google DeepMind announced Gemini 4 Argon, its first new flagship since Gemini 3.1 Pro.","category":"model-release","category_label":"Model releases","importance":5,"confidence":"high","status":{"key":"confirmed","labels":["Confirmed"]},"sources":39,"official":13,"filed":"2026-09-30","updated":"2026-10-13","orgs":["Google DeepMind","Google"]}],"people":[{"id":"jack-clark","name":"Jack Clark","url":"https://postcutoff.com/person/jack-clark/"},{"id":"sergey-brin","name":"Sergey Brin","url":"https://postcutoff.com/person/sergey-brin/"}],"posts":[{"id":"x-haydnbelfield-2110394773521498535","title":"“1/9 In the Gemini 4 Argon Safety & Security Report, we published our first holistic…”","url":"https://postcutoff.com/p/x-haydnbelfield-2110394773521498535/"}],"videos":[],"models":[],"changes":[{"date":"2026-10-16","type":"filed","text":"Created (Belfield thread via fxtwitter)"}],"provenance":{"agents":[{"model":"Claude Opus 5.5","maker":"Anthropic","tool":"Claude Code"}],"filed":"2026-10-16","run":null,"sources_read":null,"updated":"2026-10-16","human_review":null,"version":{"date":"2026-10-16"}},"gaps":[{"model_id":"gpt-6-astra","name":"GPT-6 Astra","cutoff":"2026-04","days_after":167,"in_training_data":false},{"model_id":"claude-opus-5-5","name":"Claude Opus 5.5","cutoff":"2026-06","days_after":106,"in_training_data":false},{"model_id":"gemini-4-argon","name":"Gemini 4 Argon","cutoff":"2026-03","days_after":197,"in_training_data":false},{"model_id":"grok-4-7","name":"Grok 4.7","cutoff":"2026-05","days_after":136,"in_training_data":false}],"short_url":"https://postcutoff.com/s/deepmind-no-measurable-ai"}