The concept behind the show
AI reasoning, explained
Chain of Thought takes its name from how modern AI gets hard things right: by reasoning through them, step by step, before answering. This is the builder's map of that idea — the vocabulary, the trade-offs, and the conversations where the people building it work out where it's heading.
Ask a language model for the answer to a multi-step problem and it often jumps straight to a wrong one. The breakthrough was simple: ask it to show its work first. Chain-of-thought prompting — "think step by step" — makes a model build its answer on visible intermediate reasoning, and on math, logic, and planning that single change reliably lifts accuracy. It came out of a 2022 paper and quietly reshaped how everyone prompts.
From there the field moved the reasoning inside the model. Reasoning models are trained to do extended step-by-step thinking by default — exploring, checking, and revising before they commit — so you no longer have to coax it out with a prompt. The lever underneath them is test-time compute: keep the model fixed and let it spend more effort while answering. For years the way to a better answer was a bigger model; this is the other axis, and DeepSeek's breakthrough made it concrete — strong reasoning from heavier inference rather than a larger base model.
None of this makes a model honest. A fluent, confident chain of reasoning can still land on a wrong answer — an AI hallucination — because the model is still generating what's plausible, just over more steps. So reasoning is a dial, not a default: it buys accuracy on genuinely hard problems at a real cost in latency and tokens, and it's overkill for simple ones. Knowing when to spend it is the actual skill.
The vocabulary
- Chain-of-Thought Prompting Chain-of-thought prompting asks a model to work through its reasoning step by step before giving a final answer. Spelling out the intermediate steps measurably improves accuracy on multi-step problems like math, logic, and planning.
- Reasoning Models Reasoning models are LLMs trained to do extended step-by-step thinking before they answer, spending more compute at inference to work through hard problems. They trade latency and cost for accuracy on math, code, and multi-step logic.
- Test-Time Compute Test-time compute is the processing a model spends while answering, rather than during training. Letting a model 'think' longer at answer time — exploring and checking more before it commits — raises accuracy on hard problems without retraining the model.
- AI Hallucination An AI hallucination is when a model states something false or fabricated as if it were fact — a confident answer with no grounding in its training data or the sources it was given. It's a property of how language models generate text, not an occasional bug.
From the conversations
Where guests building this work through reasoning, with full transcripts.
- How DeepSeek Changed the AI Race Overnight Atindriyo Sanyal, Galileo
- Why LLMs Are Plausibility Engines, Not Truth Engines | Dan Klein Dan Klein, Scaled Cognition
- Hallucinations Are a Data Architecture Problem | Sudhir Hasbe, Neo4j Sudhir Hasbe, Neo4j
- Agent Memory: The Last Battleground in the AI Stack | Richmond Alake, Oracle Richmond Alake, Oracle
- The Making of Gemini 2.0: DeepMind's Approach to AI Development and Deployment | Logan Kilpatrick Logan Kilpatrick, Google DeepMind
- The Critical Infrastructure Behind the AI Boom | Cisco CPO Jeetu Patel Jeetu Patel, Cisco
Common questions
- What is chain of thought in AI?
- Chain of thought is a model working through a problem step by step before giving a final answer. Spelling out the intermediate reasoning measurably improves accuracy on multi-step problems like math, logic, and planning.
- What is the difference between chain-of-thought prompting and a reasoning model?
- Chain-of-thought prompting coaxes step-by-step reasoning out of an ordinary model by asking it to "think step by step." A reasoning model is trained to do that extended thinking internally by default, spending more compute at answer time rather than needing to be prompted for it.
- Does reasoning make AI more accurate?
- On hard, multi-step problems, yes — letting a model reason before answering reliably raises accuracy. The trade-off is cost and latency, and a confident-looking chain of reasoning can still reach a wrong answer, so reasoning improves results without guaranteeing them.
Chain of Thought is where builders reason through what's changing in AI, hosted by Conor Bronsdon. New conversations every week.