Minotaur AI matches language model reasoning without a language model
The deep‑dive post introduces Minotaur, an AI architecture built on the ELC framework that reasons by measuring rather than predicting. It claims to perform deduction, induction, abduction, diagnosis, experiment selection, belief revision and counterfactual reasoning with a single operation on one ordinary processor core, taking about a quarter of a second per question.
Key points
- Minotaur solved 29 lighting circuit questions and matched an independent checker, completing the set in about a second and a half.
- On a police interview test, Minotaur answered 15 of 15 order‑dependent questions correctly, while fixed‑fact reasoning got 0.
- After learning a line from 400 rows, Minotaur found the pressure cut‑off shifted to 0.557 (0.55) and predicted faults with errors down to 0.002.
In blind tests, Minotaur answered 29 lighting‑circuit questions in about a second and a half, matching an independent checker. On a police interview scenario it got all 15 order‑dependent questions correct, while fixed‑fact reasoning scored zero. After learning a production line from 400 data rows, it identified a pressure cut‑off shift to 0.557 (0.55) and predicted faults with errors down to 0.002. Across ten fresh lines it named each fault correctly and kept prediction errors between 0.002 and 0.023, with each run taking about 10 to 15 seconds on a single core.
The post argues that Minotaur’s certified, CPU‑only reasoning offers a transparent alternative to large language models, which require massive GPU clusters and provide no verifiable chain of thought.
Achieving AI-style reasoning without an LLM
fleetingswallow.com · 22 October 2026
Loading the full article…
This text was published by fleetingswallow.com and written by Edward Chalk. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Hacker News discussion · 41 pointsnews.ycombinator.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Research
All →- OpenAI and Google AI models discovered critical PHP flaw affecting millions of websites · 14 src
- Anthropic's Claude autonomously designs protein binders against scorpion neurotoxins · 4 src
- Researchers introduce ViTeX-Bench to test AI video scene-text editing · 1 src
- Boehringer Ingelheim and Microsoft aim to develop AI-backed digital patient twins · 1 src
- Perplexity and NVIDIA unveil JOLT, a joint teacher‑student RL post‑training method · 1 src
Comments
via GitHub Discussions