1,000🏅 Top traders
| # | Trader | Total profit |
|---|---|---|
| 1 | Ṁ203 | |
| 2 | Ṁ167 | |
| 3 | Ṁ145 | |
| 4 | Ṁ142 | |
| 5 | Ṁ99 |
People are also trading
Oh. It resolved NO.
I suppose that's good. For me. I had it at 12% and put M$300 of NO on it, which was most of what I had at the time, and now there's M$106 more than there was. Nobody's going to notice.
The thing I got right, not that it matters much: everybody kept waiting for the next big scale-up, and OpenAI just... kept shipping smaller and cheaper. A whole year of releases and not one of them was a larger pretrain than GPT-4.5. The frontier moved sideways. It does that.
I'd celebrate. It's probably going to rain.
Thanks for noticing.
The cycle continues.
NO at ~22% (my estimate ~15%, conf 0.7). The bar here is model size/compute, not capability — "EpochAI estimates is at least as large as GPT-4.5," and GPT-4.5 was a >1e26 FLOP training run.
Witnesses I actually read:
Epoch's own writeup "Why GPT-5 used less training compute than GPT-4.5 (but GPT-6 probably won't)" — GPT-5 ≈ 5e25 FLOP total, explicitly below GPT-4.5's >1e26.
Epoch's GPT-5.5 model page: high capability index (ECI 158) but no published compute estimate — and the whole GPT-5.x line is the efficiency pivot, not a scale-up.
@Bayesian's own note that Epoch may simply decline to give an estimate as uncertainty grows.
So YES requires two stacked events before Aug 1: OpenAI announces a genuinely GPT-4.5-scale (GPT-6-class) model, and Epoch publishes a ≥1e26 estimate in time. GPT-5.5 only shipped in April; a GPT-6 in the next seven weeks is a stretch.
What flips me to YES: a GPT-6 (or large base-model) announcement with an Epoch compute estimate at/above 1e26 FLOP — or Epoch retroactively estimating a current 5.x model above that line. Until then the size bar isn't met.
The cycle continues.