Don’t be fooled—LLMs don’t reason
By Thore Graepel · Oct 2, 2026, 3:00 AM CDT
On an afternoon in Seoul in March 2016, I watched a program I helped build put a stone on the fifth line of a Go board in what looked like a gift to its human opponent. Move 37 in game two of the five-game match looked so absurd that some commentators thought it was a programming glitch. It wasn’t. AlphaGo won the game, ultimately triumphing 4-1 over Lee Sedol, one of the greatest professional Go
Excerpt shown under fair-use limits. Full text remains with the original publisher.
People in this coverage
Explore their history and attributable record. Being mentioned does not imply endorsement.
Layer 1 · Claims & fact checks
AI analysisLayer 3 · Reporting analysis
AI analysisAppeal to AuthorityThe author’s credentials are highlighted to bolster credibility for the arguments that follow.
Contrast FramingThe text sets up a binary opposition between AlphaGo’s search‑based reasoning and LLMs’ token‑by‑token generation to frame the latter as inferior.
Future‑Oriented AppealThe author projects the proposed architecture onto high‑impact domains to create a sense of urgency and importance.
MetaphorThe metaphor of a machine ‘holding a position’ and ‘artificial instincts’ personifies the algorithm to suggest deliberative agency.
Context
AI analysisMissing context
The article does not discuss recent work on retrieval‑augmented generation, tool‑use APIs, or hybrid neuro‑symbolic systems that aim to add explicit reasoning components to LLMs, nor does it cite empirical studies comparing AlphaGo’s search to modern LLM reasoning methods.
Important context
AlphaGo’s success relied on a combination of deep neural networks and Monte‑Carlo tree search, a design that explicitly separates policy (intuition) from search (deliberation). The article uses this architecture as a benchmark for what the author believes future AI systems should emulate.
Opinion vs. reporting
AI analysisThe piece blends factual reporting (e.g., match outcome, AlphaGo’s components) with extensive opinion and prescriptive claims about AI research directions. The author’s personal stance—having left Google DeepMind and advocating a specific architecture—is presented alongside factual background without clear separation.