@JaredSleeper Eh there's another framing that is closer to true. I'm not spinning up some marginal project that would hit my limit now to preserve optionality for a potentially more valuable project I might come up with tomorrow
Rounding out the All-Gather lineup: @corbtt (CoreWeave), @maxrumpf (https://t.co/ju8NIiY7I9), and @alxnails (RadixArk) will lead our chalk talks.
See you at the Exploratorium tomorrow.
https://t.co/fKVepxDdv6
i believe this is true.
if a rogue employee says "add a backdoor that can never, ever be detected" it'll be very hard to detect.
unless you have the intent.
The obvious levers in my mind are:
1) more rigor in defining scaling laws. Lots of experiments currently on 100B models could be on 0.5B models if we had better scaling laws
2) better early-stopping decisions
3) there's actually still a lot of difference in idea quality between ideas that all seem "strong" to a top-human-level researcher
I have a solution to opus 5.5 over-refusals I'm happy with!
1. Use bb
2. opus 5.5 is main agent
3. when it refuses (eg. pasting API keys between systems), tell it to have sol 6.1 do it
4. Opus is fine responsibility-laundering thru sol, sol is fine doing the dirty work!
@patrickc I have a story that I sincerely believe captures a positive, recognizable and realistic vision of what the near-post-ASI world might look like. I'd love your take! https://t.co/E83cSjBHMQ
I have a solution to opus 5.5 over-refusals I'm happy with!
1. Use bb
2. opus 5.5 is main agent
3. when it refuses (eg. pasting API keys between systems), tell it to have sol 6.1 do it
4. Opus is fine responsibility-laundering thru sol, sol is fine doing the dirty work!
Back to 6.1-sol for most life automation tasks since opus 5.5 keeps getting hobbled by guardrails like "I can't copy/paste your known traveler number into the airline booking form since it's a government ID"
@deredleritt3r He committed to that as a goal... but the game theoretic reasons why everyone was worried about RSI->ASI years ago still seem to apply? Is there reason to believe he will be able to honor that commitment, assuming he sincerely wants to (which I'm willing to grant)?