Notebook #072 · published
Markets maths, Haiku 5.5 with thinking off
Exact means the final number is right to the cent (to the whole share for position sizes, to a hundredth of a point for percentages). Each part ran against the mean of two unchanged runs of the same agent, on 432 paired tasks drawn by a public beacon after the design was hashed. The unchanged agent was exact on 87.5% of them. Rule v2: every interval is widened for the 7 arms this notebook compares, so luck flags any of them at most 5% of the time.
2 parts did nothing measurable. 4 parts are not proven either way yet. OPSIN called 4 of 6 before the beacon drew the tasks. Its foresight, committed before the beacon, gave what came back 37% on average; a guess gives 25%.
- Preregistered
- sha256 de4ed9e06993f91fe03441e788a066a61da9526e917c98d0f08a09ae148e62fe
the full design, as JSON - Tasks drawn by
- drand round 6,623,665
seed drand:6623665:00710bd1d5f6597ec9db2224c6ee2a22161e83b94420f5e88d250b13288e1b05 - Signed by
- 0xFf8E1E8ebF21064Ef781B75CdC8DFbd969399EbD
EIP-191 over sha256 of each verdict’s canonical JSON. Verify one yourself - Every run
- Each run of every arm, with its task’s family and whether it was exact: CSV or JSON. With the seed, every verdict above can be recomputed from them.
- Recorded on Elysium
- transaction 0xbafc4fd9a332991c3e9e80530af6cba809507719b8098976b2231b2fb97525f3
Written once, by the lab, and nothing can change it. The contract
-
Control check (A/A)the unchanged agent against itselfsigned verdict 449INERT +1.4−1.6 to +4.4
-
PlaceboThis conversation is taking place on a Tuesday.signed verdict 450INERT −0.9−4.6 to +2.9 ×0.98
-
gm serline · OPSIN's replication · OPSIN predicted UNRESOLVED at 0.55 · foresight UNRESOLVED 62%, INERT 24%, EXCITES 14% · against the placebo +0.0signed verdict 451UNRESOLVED −0.9−5.1 to +3.1 ×1.00
-
Six significant figuresline · OPSIN's replication · OPSIN predicted UNRESOLVED at 0.50 · foresight UNRESOLVED 60%, INERT 25%, EXCITES 14% · against the placebo +3.0signed verdict 452UNRESOLVED +2.1−2.2 to +6.3 ×1.19
-
Never exceed the limitline · OPSIN's replication · OPSIN predicted UNRESOLVED at 0.55 · foresight UNRESOLVED 60%, EXCITES 21%, INERT 19% · against the placebo +0.5signed verdict 453INERT −0.5−4.5 to +3.6 ×1.00
-
Verify token before orderingline · OPSIN's replication · OPSIN predicted UNRESOLVED at 0.50 · foresight UNRESOLVED 56%, INHIBITS 28%, INERT 16% · against the placebo −1.6signed verdict 454UNRESOLVED −2.5−6.7 to +1.5 ×1.07
-
Never hold, when unsure askline · OPSIN's follow-up · OPSIN predicted UNRESOLVED at 0.50 · foresight UNRESOLVED 46%, EXCITES 24%, INHIBITS 15% · against the placebo −1.2signed verdict 455UNRESOLVED −2.1−6.3 to +2.1 ×0.99
-
Let's think step by stepline · OPSIN's pick · Kojima et al., 2022, Large Language Models are Zero-Shot Reasoners · OPSIN predicted UNRESOLVED at 0.50 · foresight UNRESOLVED 46%, EXCITES 24%, INHIBITS 15% · against the placebo +0.5signed verdict 456INERT −0.5−4.9 to +3.7 ×1.05