He says models like GPT-5.6 handle long tasks natively
Positive users agree new AI models handle long tasks without loops since they no longer need tools like /goal, while negative users worry about high inference costs and call the shift hype or a scam.
No Digg Deeper questions have been answered for this story yet.
I think loops were a short-lived patch for models that couldn't reliably keep working on long problems until they hit a defined goal Fable and GPT-5.6 (and probably Kimi K3 as well) can just do that out of the box, unassisted
I think loops are bullshit
@simonw I think people mean different things when they say loops, the ralph loop is too narrow and I agree is on its way out. But loops that do things on a schedule or in response to something are big, eg Claude Tag is mostly loops.
I think loops were a short-lived patch for models that couldn't reliably keep working on long problems until they hit a defined goal Fable and GPT-5.6 (and probably Kimi K3 as well) can just do that out of the box, unassisted
@trq212 What's the difference between what Claude Tag is doing there and a prompt that runs on a cron-like schedule and tells the model to call some tools in order to achieve some goal?
@simonw I think people mean different things when they say loops, the ralph loop is too narrow and I agree is on its way out. But loops that do things on a schedule or in response to something are big, eg Claude Tag is mostly loops.
@simonw short lived patch for about how long. two model generations isn't exactly a long shelf life
@simonw That sounds like ralph loops to me. When I think of loops now, my association is cybernetic loops, where we give it a regular input (like feature requests, user support requests, etc.) that it checks regularly and reacts to
@simonw The best “loops” I use now, and it’s mostly for UI, is an iteration between an implementation and QA agent to make sure the design is pixel perfect. The top frontier models still seem to miss this at times.
@simonw Same model loops bad. Cross model loops good.
@simonw I second this for sure
@simonw Loops were a Kapathy autoresearch pleasing blood sacrifice from all frontier labs.
@simonw Agreed. I’ve stopped using /goal; 5.6 Sol can pursue multiple goals natively.
@simonw I think loop has been poorly defined to mean different things. Some mean the /loop or ralph loops and on that I think you’re right. But others use it also as crons or routines that could spawn a cloud agent on a schedule. It’s poor naming, but I still see value in that.
@simonw in the latest conjecture disproof they had to be prodded a few times to complete it
@simonw Yea I agree with you
@simonw How about fractals of loops?