DeepSeek targets 'infinite context' and test-time parametric continual learning
teortaxesTex · x · 2026-09-13
Blogger teortaxesTex reads DeepSeek's roadmap: since V3 the goal has been 'infinite context' — KV per token reduced to 890 bytes, near-linear costs at 1M context, extendable to millions via RLM and harness tricks. He also surfaces a mission statement from DeepSeek's Shengding Hu: 'Not RSI, not harness-level evolving, straight shot to test-time parametric continual learning (similar to SSI, guessed),' with a call for eligible collaborators.
Related event: DeepSeek Researcher Reveals Path to Test-Time Parametric Continual Learning(2 posts)→
More from Models
- ARC Prize 2026 closes in 3 days as Yi-Chia Chen tops ARC-AGI-3 at 75.56% — GregKamradt · 2026-10-31
- Fable 5.5 Max stuns with gorgeous SVG of a pelican riding a bicycle, headed to Claude Code — ChrisGPT · 2026-10-31
- OpenAI reportedly to train on users' Daybreak sessions with no opt-out — dyn___ · 2026-10-31
- NVIDIA Nemotron 3.5 scales LLM self-verification, +9-10 points on agentic benchmarks — sarahcat21 · 2026-10-31
- Mistral Large 4 scores 94% on 1,273 USMLE-style questions but flips 44 at temp 0 — MaziyarPanahi · 2026-10-31
- Researchers find 5 personality structures in LLM weight space, Big Five-like — luke_drago_ · 2026-10-31