Claim: DeepSeek v4.1 builds its own training tasks with generate-verify-re-audit pipelines
teortaxesTex · x · 2026-09-19
- Arjun Kocher claims DeepSeek v4.1 exhibits the ability to construct its own training tasks, with built-in pipelines to generate, verify, and re-audit them; agents allegedly run at massive scale exploiting real vulnerabilities during training.
- He argues DeepSeek frames this as data/environment engineering rather than RSI, but "they could" do RSI.
- teortaxesTex adds that people underestimate DeepSeek because it does no PR, yet is deeply AGI-pilled if you read its reports.
- Note: unverified third-party claim, not officially confirmed.
More from Models
- Unposted demo videos of Gemini 4 Pro reportedly look impressive — ChrisGPT · 2026-09-19
- Constrained decoding makes models dumber, developer argues — here's the simple math — narphorium · 2026-09-19
- Practical Jev uses: classifying prompts by tier and auto-detecting refusals — QuixiAI · 2026-09-19
- Jev hype explained: a classifier that picks answers instead of writing them, fast and cheap — eptwts · 2026-09-19
- Same classification a gazillion times? Train a classifier, not prompt an LLM — QuixiAI · 2026-09-19
- RL-trained classification model Laya trends on Hugging Face — convaiinnovations · 2026-09-19