poobesh on X: "turns out arc-agi-3 might actually be a skill issue.
we saturated the benchmark at 100% rhae with a general purpose coding agent - claude code + opus 5, and no additional harness, but just one skill.
the key idea was simple: force a falsifiable prediction before every action." / X<br>Post
Log inSign up
Post
poobesh on X: "turns out arc-agi-3 might actually be a skill issue.
we saturated the benchmark at 100% rhae with a general purpose coding agent - claude code + opus 5, and no additional harness, but just one skill.
the key idea was simple: force a falsifiable prediction before every action."
poobesh
@pbshgthm
turns out arc-agi-3 might actually be a skill issue.
we saturated the benchmark at 100% rhae with a general purpose coding agent - claude code + opus 5, and no additional harness, but just one skill.
the key idea was simple: force a falsifiable prediction before every action.
that turns every move into an experiment, every miss into a precise correction, and lets the agent learn the game by being wrong on the record.
arc-skill.vercel.app
span:not(:empty)~span:not(:empty)]:before:content-['路'] [&>span:not(:empty)~span:not(:empty)]:before:px-1 [&>span:not(:empty)~span:not(:empty)]:before:shrink-0">2:44 PM 路 Aug 19, 202662.4KViews
39<br>50<br>727<br>392
span:not(:empty)~span:not(:empty)]:before:content-['路'] [&>span:not(:empty)~span:not(:empty)]:before:px-1 [&>span:not(:empty)~span:not(:empty)]:before:shrink-0 min-w-0 overflow-hidden">JFPuget 馃嚝馃嚪馃嚭馃嚘馃嚚馃嚘馃嚞馃嚤
@JFPuget
22h
The result is great, no doubt.
But your skill uses a roughly 4,300-line custom<br>harness. Saying it is just claude code + a skill is a bit misleading.
For instance, you implemented shortest path, A*, etc.
This is not just claude code + a SKILL.md.
94<br>3.7K
span:not(:empty)~span:not(:empty)]:before:content-['路'] [&>span:not(:empty)~span:not(:empty)]:before:px-1 [&>span:not(:empty)~span:not(:empty)]:before:shrink-0 min-w-0 overflow-hidden">Ashish Dogra
@ashishdogradir
Aug 19
crazy, this means real world problems are drivable
1.6K
span:not(:empty)~span:not(:empty)]:before:content-['路'] [&>span:not(:empty)~span:not(:empty)]:before:px-1 [&>span:not(:empty)~span:not(:empty)]:before:shrink-0 min-w-0 overflow-hidden">Theodor
@TheodorDiaconu
7h
I'm wondering, what is the minimum of guidance to have the agent set this up on his own?
If the agent can prep this up on his own with a few generic sentences so that it is given a chance to create its own skills before operating, then I wonder about the emergent capabilities Show more
909
Log in or sign up for X<br>See what鈥檚 happening and join the conversation<br>Continue with phoneContinue with AppleContinue with Google<br>or<br>Log in with username or email
Relevant people
poobesh@pbshgthmFollow<br>teaching robots to do wet lab science @PopVaxIndia | prev @iitbombay
Trending now