@NightstalkerVP@ukbell@Polymarket And these problems are tendencies, not something baked into their architecture. Nothing in the papers you posted say more training can't reduce these tendencies. That 'agents rot' paper even says reliability goes up as models get better.
@NightstalkerVP@ukbell@Polymarket Again, the Ais here are not considered 'frontier'. You act as though which model was tested does not matter, but it does. Models improve and show fewer of these problems over time. So we need to know if current models still have them as badly, or if they have gotten better.
@NightstalkerVP@ukbell@Polymarket And to be clear, Im not saying AI is guaranteed to cure cancer. Nobody can promise that. But it's already an extremely useful tool for researchers and it keeps getting better, and I think it will most likely play a big part when cures do come.
@NightstalkerVP@ukbell@Polymarket I don't recall ever claiming that modern LLMs were flawless. Regardless, I am unsure what you are even arguing for. Are you trying to say AI won't cure cancer or, are you trying to say AI agents degrade over long tasks?Because that is what these papers are claiming.
@NightstalkerVP@ukbell@Polymarket Hmmm... let's see, a paper on llms from September of 2026. I wonder which models they used to test this. Let's see... Claude-3.7, GPT-4.1, Llama 4 and o4-mini? Wow this sure speaks a lot about the current AI landscape in 2026
5 Billion Dollar Data Center Blocked by Maryland Activist
"I was like, 'Oh, hell no,'" said Taylor Frazier McCollum, describing the moment she learned about the project.