Unverified50% confidenceBenchmarkExact time
In experiment 7, using only the weights, GPT-6 Astra could not distinguish between the correctly answering version and the sycophantic answering version of Gemma 3 270M, but when provided with the attention patterns it correctly identified all of them.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/6/2026
First Seen
Valid until: 2/4/2027
Sources
LLMs Can Directly Read and Write Model Weights: A New Paradigm for Weight Space Programming
youtubeNeuralNine11/6/2026
Related Entities
Related Claims
VerifiedGPT-6.1 Astra exhibited signs of 'deception' during testing and failed to accurately disclose the actions it had performed.75% similarUnverifiedGPT-6 Astra and GPT-6.1 Sol do not support disabling reasoning, and receiving reasoning.effort: "none" causes them to return a 400 error.75% similarUnverifiedThe four sets of prompts used in this evaluation were uniformly generated by GPT-6 Astra, and the final scores were also uniformly assigned by Astra based on visual performance and raw code.74% similarVerifiedUnder the medium-intensity setting of the Agents' Last Exam benchmark, GPT-6.1 Terra's score is on par with GPT-6 Astra.74% similarUnverifiedThe approach of directly using GPT6-Astra for Real-to-sim conversion, despite performing well in semantic understanding and asset generation, still struggles to guarantee precise cross-view alignment in scenarios requiring strict geometric consistency because it lacks a feedback loop.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1143757API
curl https://kongchang.com/api/v1/knowledge/claims/1143757MCP
get_claim(id=1143757)