CIRCLE WITH A DOT

"Our analysis shows that current LLMs are unreliable delegates: they introduce sparse but severe errors that silently corrupt documents.""Our large-scale experiment with 19 LLMs reveals that current models degrade documents during delegation: even frontier models (Gemini, Claude, GPT) corrupt an average of 25% of document content by the end of long workflows, with other models failing more severely."https://arxiv.org/abs/2604.15597#ai #llm #tech #science #gpt #claude #gemini

Yesterday’s side‑by‑side release of #GPT‑5.5 and #deepseekv4 is interesting, but not for the usual “is this a leap?” debate.What stands out is that DeepSeek continues to operate near this tier at all, given the hardware and compute constraints they’re clearly optimizing against. That’s not luck; it’s a sustained signal about where leverage actually lives and the biggest battle in the industry. 🧵1/4