Astra's rumored latent CoT is hard to monitor; researcher proposes auxiliary decoder
beffjezos · x · 2026-09-02
A user flagged that Google's Astra appears to use neuralese / looped transformers, meaning it no longer has a fully monitorable chain of thought like earlier models — bad news for CoT oversight. Interpretability researcher beffjezos offered mitigation ideas:
- Train an auxiliary decoder model to translate latent representations (latent CoT) into readable form;
- Run it as a side path on GPUs via prefill batching, keeping overhead manageable;
- Monitoring can be periodic rather than constant, which is fine in practice.
More from Models
- MOSS SFT vs Flux 3 Klein vs SD 3.0: Same-Prompt Comparison Exposes SD3's Age — ZootAllures9111 · 2026-10-18
- Quick take: GPT-6 matches Opus 5.5 at coding, far ahead at computer use — _vztu · 2026-10-18
- Open models now above Sol-level, outpacing expectations as community awaits V4.1 Pro — yacineMTB · 2026-10-18
- Codex took 3 days, Opus 5.5 nailed it in an afternoon — banned user's video protest — dotey · 2026-10-18
- Yacine MTB: Users will remember how Anthropic treated them once intelligence proliferates — yacineMTB · 2026-10-18
- Mistral's Timestral forecasting models start at 5M params and run on plain CPUs — lmoroney · 2026-10-18