Post #10669

AI researchers have been working on getting gen-ai systems working without human intervention for longer. Just 8 months ago the longest a system could run unsupervised was 8 minutes. This week a leading team of researchers announced that they had had an LLM doing agentic devlopment unsupervised successfully for EIGHT CONTINUOUS WEEKS, The magic behind this breakthrough is profoundly simple; they simply run the model at a much lower speed than before so it still only has to write 8 lines of code (the maximum such a system can produce safely under optimal conditions), but now it does it much more slowly. This technique is easily replicated and is expected to allow all models to spend longer on tasks from this moment on.