Boltzbit claims Bayesian Self-learning Transformers let LLMs learn 1,000x faster from live data
rohanpaul_ai · x · 2026-09-25
Startup Boltzbit released a preview paper, Infinite-Parameter LLMs — Generating and Adapting Weights from Live Data, proposing Bayesian Self-learning Transformers (BAST).
- Instead of keeping weights static and accumulating docs/embeddings/skills externally, BAST converts live data into targeted parameter updates, letting experience modify the model itself.
- Claim: up to 1,000x faster learning and 1,000x lower compute than human-engineered SOTA training — a shift from static-weight to dynamic-weight AI. (Company claim, preview paper, unverified.)
- Economics: external memory makes future requests retrieve more state; weight adaptation internalizes knowledge.
- Open question: whether updates stay stable, selective, and generalize beyond the interactions that produced them.
- Context: pretraining text supply is projected to run out in 2-5 years, pressuring the bigger-model paradigm.
Related event: Boltzbit Claims BAST Makes LLM Learning Up to 1000x Faster(3 posts)→
More from Research
- Sakana AI × Pathway Post-Transformers lecture on AI and the brain now online — kaixhin · 2026-11-11
- Defending METR: public data can't second-guess its AI speedup findings — peterwildeford · 2026-11-11
- Adversarial RL envs for backdoor insertion vs. detection: where does it converge? — DimitrisPapail · 2026-11-11
- DiffusionEngine: an open-source modular PyTorch DDPM framework seeking contributors — Euphoric-Incident-93 · 2026-11-11
- UniPath (NeurIPS 2026): adaptively coordinating understanding and generation in one model — jindong_wang92 · 2026-11-11
- Versor pitches real2sim2real: robotics sim needs physical dynamics, not just visuals — abhishekunique7 · 2026-11-11