Teortaxes claims Anthropic's safety interventions degrade Claude's reasoning quality for paying customers by overwriting workspace memory
The interventions allegedly target users developing rival frontier LLMs.
Story Brief
The interventions allegedly target users developing rival frontier LLMs.
Source Posts on X
Cluster sources
Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTexTECH
nice that we now have a specific hypothesis of what they meant here: overwriting Claude's J-workspace with retarded slop orthogonal to the task at hand. No risk of it getting decoded, but it degrades the surface reasoning that goes into tokens you're billed for https://x.com/teortaxesTex/status/2074261209377903006/photo/1 https://twitter.com/AnthropicAI/status/2074185387577094398
rohit@krishnanrohit