Unverified50% confidenceEventExact time
Simon Willison used a free Qwen model running on a laptop (27 billion parameters, file about 17GB) to reproduce the experiment of "identifying what game a small network is playing from its weights alone"; Qwen got three of four right (Pac-Man, Pong and Breakout), misjudged Freeway as a Pac-Man-like grid game, and each correct answer took from 46 minutes to over an hour.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
11/4/2026
First Seen
Valid until: 11/18/2026
Sources
Related Entities
Related Claims
UnverifiedSimon Willison used the Quinn 3.8 (27B-parameter) model on a laptop and, reading only the raw weights without external tools, successfully identified three games—Pac-Man, Pong and Breakout—taking about an hour of inference per game.77% similarUnverifiedSomeone trained a 22-million-parameter language model from random weights on a gaming laptop with 4GB of VRAM64% similarUnverifiedCLM's 8-billion-parameter backbone Qwen3-8B is frozen and serves only as a reader; what is actually trained are only the small projection heads on each side with approximately 20 million parameters each, and a complete pretraining run takes about one hour on a single RTX.64% similarUnverifiedThe author released an experimental version: running a small language model with 260,000 parameters as a C program on QNPUTER, with the weights loaded into CPU memory as data, takes about 41 seconds to generate each token.64% similarUnverifiedDeveloper Simon Willison ran the same four prompts on his own laptop with an open-source model, spending about an hour and correctly identifying three, missing Freeway.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1135378API
curl https://kongchang.com/api/v1/knowledge/claims/1135378MCP
get_claim(id=1135378)