Many don't understand the scale ups of model training planned within the next ~18 months. Here is an expanded version of the chart that me and Aran Komatsuzaki put out recently. (Now includes upcoming 200K+ B200 scale models) 100X+ beyond the current frontier of released models.

GODMAAX is the new FAANG G = Google(Deepmind) O = OpenAI D = Deepseek M = Meta A = Anthropic A = Alibaba(Qwen) X = XAI

Do you believe there is fundamental limitations with specifically the transformer architecture that prevents it from autonomously doing 50% of todays US remote jobs, while matching performance of the average worker in each of those job titles, and equal or better cost efficiency.

These new economic growth forecasts from EpochAI seem to align well with METRs recent capability forecasts. Especially when you look at the estimated times for 99% accuracy at high time horizons, and how those seem to converge with the full automation estimates by EpochAI below..

Going to work on an updated version of this. Please reply or DM if you have recommendations of newer interesting benchmarks I should look at, with scores for O3, as well as human baselines (and preferably have the other criteria mentioned in the attached tweet too)

This seems like one of the first unifications of the concept of spiking neural networks, persistent states, attention and adaptive compute, all in one model/“machine”. Great work from Sakana, I believe they’re one of the most under-rated openly publishing labs right now.