Antirez vs mlpowered: do LLM hidden states encode plans, not just next tokens?
rickasaurus · x · 2026-11-06
antirez cautions that an LLM's hidden state only carries what's needed for the next token — it's not an encoder. mlpowered counters that this is a misconception: hidden states predict all future tokens, with representations corresponding to plans, citing Anthropic's circuit-tracing work. The accompanying graphic shows the J-lens reading out intermediate reasoning steps, mental math, bug anticipation, protein recognition, and prompt-injection awareness inside Claude.
More from Models
- Google Ultra subscriber blocked from Antigravity by region detection — lxfater · 2026-11-06
- Palisade's Jeff Ladish: skip the 40 benchmarks, ask if agents finish real work — FinanceYF5 · 2026-11-06
- AA Index 31: 594B total / 17B active params, Apache 2.0 open source — AIFlow_ML · 2026-11-06
- Moonshot's Kimi K 3.1 Pro surfaces: 2.8T-param MoE, weights on Hugging Face — thursdai_pod · 2026-11-06
- Tencent Hunyuan Hy4 hits limits on day one, WorkBuddy rushes 10x compute expansion — TencentHunyuan · 2026-11-06
- NaceAI open-sources Drex decision models, hitting SOTA 67.52 on Decision Index — nischay_twt · 2026-11-06