Models Are Getting Dumber on Purpose

The article argues that modern AI models are becoming “smarter” at reasoning while intentionally becoming less knowledgeable in their weights. Recent models such as GLM-5.2, Qwen3.5, and DeepSeek V4-Flash achieve impressive math and code benchmark results with far fewer active parameters than earlier frontier models like GPT-4. However, this improvement does not extend to factual recall: on benchmarks like SimpleQA, even the best models answer only about half the questions correctly, while smaller models hallucinate most of the time. The author explains that this tradeoff is deliberate. Storing facts in model parameters is expensive because factual knowledge requires large capacity and quickly becomes outdated. By contrast, reasoning skills such as decomposing problems, tracking intermediate steps, and checking work can be compressed more efficiently and transferred through distillation and reinforcement learning. As a result, labs are optimizing models to be broad generalists with shallow knowledge rather than deep repositories of facts. The article argues that the missing knowledge should come from external systems: retrieval tools, web search, documentation, and agent harnesses. This makes models more up to date, easier to correct, and less prone to unfixable hallucinations. In this vision, future frontier models may run on consumer GPUs, provide strong reasoning locally, and rely on runtime tools for current facts instead of storing them in weights.

Build this article’s timeline

Extract dated events, key actors, and open questions using the same timeline analysis available in Signal.

Log in to build

Timeline generation is available to signed-in community members.

Five Ws 🤖 3 AI prefills · 2 still unanswered
Who
Artificial Analysis, Google, and PostgreSQL
🤖 AI prefill
What
The article argues that modern AI models are becoming “smarter” at reasoning while intentionally becoming less knowledgeable in their weights.
🤖 AI prefill
When When did this happen, or what is the timeframe?
Where Where did this occur?
Why
Centers on reasoning scores, per-token compute, and hallucination rates
🤖 AI prefill

Login to answer and earn XP for each gap you fill.

Reasoning vs Recall reasoning, factual recall, benchmarks, hallucination, math Knowledge-Compute Tradeoff parameters, facts, knowledge capacity, compression, distillation Stale Weights Problem training cutoff, staleness, APIs, prices, job changes Retrieval Harness Design retrieval, tool calls, web search, filesystem, docs Local Frontier Models consumer GPU, quantization, active parameters, DeepSeek, Qwen
reasoning scores92% per-token compute90% hallucination rates89% factual recall87% knowledge capacity86% reinforcement learning84% synthetic textbook-style data80% retrieval over a knowledge base88% tool calls79% expert layers83%
18
Manipulation Score: Low How much this article uses emotional framing vs. straight reporting
Loaded Language 1425%
Absolutist 2215%
Threat/Panic 1025%
Us vs Them 415%
Engagement Bait 3020%
Manipulation score look right?

Factual Claims (10)

medium technology
GLM-5.2 scores 99.2% on AIME 2026 with about 40 billion parameters active per token.
medium technology
Qwen3.5 scores 91.3% with 17 billion active parameters.
medium technology
DeepSeek V4-Flash runs 13 billion active parameters.
high technology
GPT-4 was rumored to run around 280 billion active parameters in 2023.
medium technology
SimpleQA's current leader is Gemini 2.5 Pro at 53%.
medium technology
Artificial Analysis measures Qwen3.5 4B and 9B at hallucination rates of 80 to 82% on its knowledge benchmark.
medium scientific
Research on knowledge capacity suggests about two bits of factual knowledge per parameter.
medium technology
Phi-4 is 14 billion parameters and was trained heavily on synthetic textbook-style data.
medium economic
A frontier training run takes months and costs hundreds of millions of dollars.
medium technology
A 20 to 40B model at 4-bit quantization fits on a 24GB consumer GPU card.

Comments

Login to view and post comments.

🏠Portal 📰Links ❓Q&A 📅Events 💼Jobs