reasoning
In the AI context, the model's capacity to spend additional compute time deliberating through a problem step-by-step to reach a higher-quality answer.
As defined in et_982
Other definitions
In the context of AI, the ability to build a model of the world, apply rules or patterns to derive new conclusions from existing facts, and solve novel problems—fundamentally different from pattern matching or text prediction.
Authors of "THE AI CON" discussing the problematic hype about AI & what those working tech should do
Where it comes up · 10
- Sam Altman on The Future of AI | Ep. 137 claims
- AI's Trillion-Dollar Opportunity: Sequoia AI Ascent 2025 Keynote3 claims
- Mark Zuckerberg — Llama 3, $10B models, Caesar Augustus, & 1 GW datacenters2 claims
- Authors of "THE AI CON" discussing the problematic hype about AI & what those working tech should do2 claims
- Capital, Compute & AI Scaling1 claim
- Inside the Trillion-Dollar AI Buildout1 claim
- John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI1 claim
- Mark Zuckerberg — AI will write most Meta code in 18 months1 claim
and 2 more
Most important claims · 10 of 20Search all claims for “reasoning” →
- We have cracked reasoning in AI models - they can now perform the kind of domain-specific reasoning expected of PhD-level experts in their respective fieldsSam Altman on The Future of AI | Ep. 13
- AI models have cracked reasoning—they can now perform the kind of reasoning in specific domains that a PhD in that field would be expected to do, such as winning top scores on world's hardest math competitions or solving problems an expert PhD would tackle.Sam Altman on The Future of AI | Ep. 13
- Pre-training scaling is slowing down because 9-10 orders of magnitude of progress since AlexNet means most low-hanging fruit has been picked, requiring research breakthroughs like reasoning, synthetic data, tool use, and AI scaffolding to continue intelligence scaling.AI's Trillion-Dollar Opportunity: Sequoia AI Ascent 2025 Keynote
- Reasoning models that consume more test-time/inference-time compute to provide more intelligence are a compelling paradigm for math and coding, but for consumer products latency and intelligence-per-cost matter more — people won't wait half a minute for an answer when a good answer in half a second is a great tradeoff.Mark Zuckerberg — AI will write most Meta code in 18 months
- Psychological reasoning is not formal logic but the generation of reasons for what we think and feel, driven by a desire to be considered trustworthy to our peers; reasons must be 'plausible' in the sense that our most trusted social network would judge them reasonable, which is why beliefs are calibrated to social acceptance rather than to truth.David McRaney on How Minds Change
- There is no presupposition of a path from language modeling to reasoning; reasoning requires building a model of the world with rules and patterns for inferring conclusions, which is fundamentally different from predicting which word is likely to come next.Authors of "THE AI CON" discussing the problematic hype about AI & what those working tech should do
- Training models on coding makes them more rigorous and better at reasoning across many domains, even for users who never ask coding questions—so Meta heavily emphasized code in Llama-3 training despite WhatsApp users rarely coding.Mark Zuckerberg — Llama 3, $10B models, Caesar Augustus, & 1 GW datacenters
- Large language models like Gemini, ChatGPT, and Claude are built on absolutely enormous, poorly curated and undisclosed training sets, making it impossible to reason about the kinds of biases that should be mitigated for particular use cases.Authors of "THE AI CON" discussing the problematic hype about AI & what those working tech should do
- Reasoning—tasks requiring test-time, step-by-step computation or deduction—is best achieved by combining inference-time computation (the model thinking step by step while deployed) with practice at training time, rather than relying on either approach alone.John Schulman (OpenAI Cofounder) — Reasoning, RLHF, & plan for 2027 AGI
- Society is not experiencing a commensurate shift despite ChatGPT being deployed at scale and achieving PhD-level reasoning, suggesting that capability breakthroughs do not automatically translate to societal transformation—the Turing test passed without fanfare.Sam Altman on The Future of AI | Ep. 13