ai r&d-5

Chris @ChrisGPT

2026-09-26 — saved image

Chris (verified) @ChrisGPT · 1h
Anthropic's own AI R&D-5 threshold is basically the point where AI causes
a dramatic acceleration in effective scaling. Using their historical estimate
of a 35x in effective scaling per year, compressing two years of progress
into one works out to roughly 35² = 1,000x effective compute.


And by the way, Opus 5.5 is already at 55.8% on CoBench 2.1, while
Anthropic says roughly 85% would correspond to a model capable of fully
substituting for its research staff on that benchmark. So it's only about 29
percentage points below their own researcher substitution threshold.

Anthropic has quantified the point where AI starts materially accelerating
the process that builds the next AI. Just like OpenAI both of these labs are
probably within one year of fully automating R&D it is my opinion the last
great barrier is continuous learning and long term memory.

[attached image: excerpt from an Anthropic document, white card, with yellow highlight over the second half of the AI R&D-5 paragraph]
Autonomous AI Research and Development:

    AI R&D-4: The ability to fully automate the work of an entry-level, remote-only Researcher at
    Anthropic.

    AI R&D-5: The ability to cause dramatic acceleration in the rate of effective scaling. [highlight begins] Specifically, this
    would be the case if we observed or projected an increase in the effective training compute of the
    world's most capable model that, over the course of a year, was equivalent to two years of the average
    rate of progress during the period of early 2018 to early 2024. [highlight ends] We roughly estimate that the 2018-2024
    average scaleup was around 35x per year, so this would imply an actual or projected one-year scaleup
    of 35² = ~1000x.[footnote 22]

ai r&d-5anthropicautomated ai researcheffective computeopus 5.5cobenchrsprecursive self-improvement