We are in a critical phase if we look at the AI 2027 paper futures model
https://www.aifuturesmodel.com/p?v=2&base=daniel-08-16-26&acam=time-horizon
To me it seems that Claude Opus 5 we got in July with an ECI score of 163 was a proto AGI system. By the paper its effective compute was 10^29 FLOPS and its coding time horizon is 6 hours. When we hit a system with a ECI score of 172 which we will in 3 to 5 months, and effective compute of 10^30 FLOPS its coding time horizon will correspond to 3 days, so a whole weekend. When we get to a system with ECI score 182 its coding time horizon will be a month. And a 191 corresponds to RSI essentially by this paper and a coding time horizon of a year.
And if we look at revenue it corresponds perfectly so far also which is a big deal too, because by the end of next year the leading AI company will be in top 3 companies in the world by revenue.
So the jumps in capability were big in the last 2 years every 3 to 4 months but now they will start to be insane. I mean this corresponds well with the situational awareness paper in 2024 and the recent comments made by Jacob Coxen.
How do you feel about this?