@thsottiaux@sama could you please fix the codex tab inside ChatGPT browser window, I cannot connect to my remote mac at home from this laptop, it leads to codex landing page even signed in and I don’t want to type on the iPhone ChatGPT app. I took the wrong laptop for travel :’)
@0xngmi Choose a weaker model and ask the candidate to build an app with or without it. Once done, review the code together & then compare it with a stronger model’s review, focusing on architecture/patterns over syntax. The time spent with them should give enough insight to hire or not
@kadak0697@thedefivillain The only problem is the token has 0 utility. I have been a customer of their compute resources for a while and use it intensively from their long list of some nice providers but I don’t wish to hold or do anything with their token, just prefer to pay like a normal human being.
@thsottiaux Love codex btw, best so far, I ditched everything else, and if I hit codex limit, I have to go slow mode by myself, which is hard these days, than instructing another agent 😭
Training frontier models over the internet requires new techniques.
Today, we present ResBM, a residual encoder-decoder bottleneck architecture that enables 128x activation compression for low-bandwidth distributed pipeline parallel training.
Developed for @IOTA_SN9, we show SOTA compression without significant loss in convergence rates, increases in memory, or compute overhead.
Expect the full paper release in the next 72 hours.
@DreadBong0@Raleigh_CA The point is, while this video seems hyped, there is still a lot of time and effort to be put in, the whole space is getting there and still better than memecoin garbage and ico era, finally something useful coming out.
@DreadBong0@Raleigh_CA Don’t know about the rest of the subnets you mentioned other than Ridges, but we gave it a try in our codebase, it is still stuck with solving the PR for a week, that I finally handcoded and merged it…embarrassing for an agentic tool. But I see there is room for growth.