As of: 2026-10-16 12:39 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/v/ai-copium-anthropic-cofounder-moved-the-timeline/

# Anthropic's Co-Founder Just Moved the Timeline…

AI Copium, 14 October 2026, YouTube. 122,603 views as of 16 October 2026.
Kind: Review.
Watch: https://www.youtube.com/watch?v=FHG1hIMo1Ag

## Why it is here

Explainer (11 min) on Jack Clark's Import AI 476 view that full recursive self-improvement could become possible in 2027, read together with Anthropic's self-replicating-AI research and its source-control redesign post; per the description it separates "what's been demonstrated, what's still a rumour". About 123k views on 2026-10-16.

## Description (written by Gemini from the video)

**Summary**  
This video is an analytical commentary presented by AI Copium discussing Anthropic's blog post on replacing Git with an in-house source control system called Scion, alongside recent developments in automated AI research and recursive self-improvement (RSI). The presenter reviews evidence from Anthropic's engineering blogs, Frontier Red Team research on self-replicating models, benchmark data from P-Zero Research's TasteVal, and statements from Jack Clark moving his timeline for full RSI forward to 2027.

**What is shown**  
- [00:00] Overview of Anthropic's blog post "Redesigning source control at Anthropic", highlighting the graph of merged changes per day surging to 30,401 merges on September 9, 2026.
- [00:06] Anthropic Frontier Red Team research paper titled "AI models may be capable of self-replicating without access to a datacenter".
- [00:10] Jack Clark’s *Import AI* newsletter excerpt highlighting his updated projection for full RSI moving to 2027.
- [00:36] Detailed walk-through of the Scion version control system blog post, explaining how 7 engineers migrated from Git in 5 months using internal Claude models (Claude Mythos Preview and Claude Tag).
- [01:48] Excerpt describing Claude rewriting the Git-to-Scion ingestion pipeline to consume 20x less memory and run 4x faster.
- [03:50] P-Zero Research chart displaying "experimental research taste" doubling roughly every 3 months since December 2025 across models from GPT-4 to Claude Opus 5.5 and Fable 5.1.
- [04:24] X post from Google engineer Patrick C. Toulme summarizing takeaways from an AI dinner in NYC regarding performance engineers managing agent loops and inference providers.
- [05:25] Rumor post by @imjustnewatai alleging internal self-improvement loops at Google, Anthropic, and OpenAI.
- [06:06] Text from *Import AI* detailing the TasteVal benchmark evaluating AI systems across 8 frontier AI development tasks.
- [08:06] Excerpt and diagram from Anthropic’s Frontier Red Team paper detailing distributed inference of Gemma4-26B-A4B across a fleet of 15 off-brand Android TV boxes.
- [09:37] "Scenario: Intelligence Explosion" chart illustrating automated AI research trajectories toward superintelligence.
- [10:00] Clip of OpenAI researcher Noam Brown discussing alignment challenges.

**Claims & numbers**  
- Anthropic averaged ~2,500 code changes merged per day in May 2026, which surged to over 30,000 per day by September 2026 (peaking at 30,401 merges on Sept 9, 2026), roughly a 12x increase in four months (the presenter says).
- Seven engineers at Anthropic built Scion, a code review surface called Antique, and related developer tooling, completing full migration in five months with Claude's assistance (the presenter says).
- Claude rewrote the Git-to-Scion ingestion pipeline prototype with no further instructions, creating a version that consumed 20x less memory and ran 4x faster (the presenter says).
- Independent research from P-Zero Research indicates that "experimental research taste" in frontier models has doubled approximately every three months since December 2025 (the presenter says).
- Anthropic co-founder Jack Clark moved his expected timeline for "full RSI" to become possible from 2028 forward to 2027 (the presenter says).
- Anthropic's Frontier Red Team demonstrated distributed inference of Gemma4-26B-A4B across 15 off-brand Android TV boxes ($55 each) generating at 0.39 tokens per second (the presenter says).

**Notable quotes**  
- [00:36] "In it, they claim coding is no longer the bottleneck to building systems or doing research."
- [02:04] "All of the engineers who worked on this system now rate Claude as better at writing code than them."
- [06:27] "'I now think it's more likely we'll see full RSI become possible in 2027, whereas before I thought it was more likely in 2028.'"

**Assessment**  
This is an independent commentary and news review analyzing published blog posts, technical papers, and public social media discussions from October 2026. The presenter accurately quotes and highlights real source material from Anthropic, Jack Clark, and P-Zero Research, while clearly delineating between published findings, industry anecdotes, and unverified rumors.

_Described by gemini-3.8-flash on 2026-10-16 from the video's audio and frames._

## Related

- 2026-10-13: [Anthropic: code merged per day grew from about 2,500 in May to over 30,000 in September, so it replaced git with an in-house system built with Claude](https://postcutoff.com/e/2026-10-13-anthropic-scion-source-control-30000-merges-a-day/)
- 2026-10-12: [Jack Clark moves his forecast for 'full RSI' to 2027, citing TasteVal](https://postcutoff.com/e/2026-10-12-jack-clark-full-rsi-2027-tasteval/)
- 2026-10-12: [Anthropic's Frontier Red Team](https://postcutoff.com/e/2026-10-12-anthropic-self-replicating-ai-iot-devices/)

## People in it

- [Jack Clark](https://postcutoff.com/person/jack-clark/), Co-founder, Anthropic; author of Import AI
- [Noam Brown](https://postcutoff.com/person/noam-brown/), Researcher (reasoning), OpenAI