Lampinen: math may not convey more meaning, but it makes things more precise
AndrewLampinen · x · 2026-09-29
In a discussion with Grady Booch and others, Andrew Lampinen argues that at least some kinds of math fall within the scope of natural language distributions. He would not assume math conveys more meaning, though in practice it likely does—at minimum it makes expressions more precise. The thread builds on his team's paper on passive learning of causal strategies in language models, engaging the B & K octopus thought experiment.
Related event: Lampinen Debates Booch on Whether Math Exceeds Natural Language(2 posts)→
More from Research
- Bayesian framework from DeepMind-linked researchers scores LLM consciousness between 1% and 80% — mhutter42 · 2026-09-29
- BioNeMo team boosts Mixtral-8x7B training throughput 2.21x vs HF BF16 baseline — AllThingsApx · 2026-09-29
- CheXanatomy generates chest X-ray anatomy segmentation via next-token prediction, accepted at MICCAI 2026 — denisparra · 2026-09-29
- Relic turns multi-agent collaboration failures into executable org protocols, +5.7pp delivery — ulab-ai · 2026-09-29
- Nereus: adaptive parallelism boosts 8B PPO throughput up to 7.27x over OpenRLHF — Songlin Jiang · 2026-09-29
- G²PTQ: generalized gradient compensation improves LLM post-training quantization — Ruikang Liu · 2026-09-29