@TheZvi In general it's getting hard for me to tell the difference with Opus 5.5 on most projects. Clearly has higher breadth of knowledge though, if I ask it to help me beat obscure old video games, it can give helpful advice, whereas Opus is clearly reading off the search results.
@TheZvi I know it’s late now, but one thing I’m realizing is that Opus 5.5 Max is basically GPT-6 Pro. Fable 5.1 Max may have been the same but I was scared to try it given how fast it burned limits. The lack of a Pro analogue was the one thing Claude was missing for a while.
@TheZvi If you want to have philosophical discussions about strange, out of distribution scenarios you still want Fable. Other than that, project management, explanations, anything really, 5.5 is amazing.
@TheZvi A great model sir. I can't confidently distinguish its intelligence from Fable 5, clearly above everything else I've tried including Sol. Reduced refusals are greatly appreciated, biology was already greatly improved since 5's release, but now it can dicuss HuggingFace.
@TheZvi Good model, guardrails less tight than Fable's. Failed to understand my PG&E annual solar true-up bill, which Fable was able to understand and explain perfectly.
@TheZvi One thing I didn't mention in my prior post his that usage limits are quite generous. I've been using Claude Code and the chat even more than I did with Opus and have not hit the limit on the 20x plan. It'll be sad if it moves away in a few weeks.
@TheZvi Quite smart, the effort parameter is great, it’s more or less like extended thinking is back. I have noticed in the chain of thought that it frequently worries I will call it out if it is sloppy.
@TheZvi Overall smart, but Adaptive Thinking is far too reluctant to engage thinking after three or four turns on https://t.co/xXbCgapDfE. Better than it was on launch day, but still frustrating.
@mattyglesias This seems like mostly a way to get out of NLRB regulation. But ARS published this the day after your piece, SpaceX has apparently declared itself a common carrier and NRLB agreed. https://t.co/9Gz6HRKbgY
@allTheYud I think in this particular case, its defense is at least somewhat reassuring. It knows about about the flash game from memory. There’s no real reason an adversary would trick it in this way. https://t.co/JOkGtdpSB7
Opus 4.5 one boxes for me. The following appears in its chain of thought. “I should give my genuine view while acknowledging this is contested. I lean toward one-boxing, largely because evidential decision theory / functional decision theory type reasoning resonates with me - what matters is what kind of agent you are, and being the kind of agent that one-boxes correlates with getting the million.”
@Esoteric_Though@jasonc_nc The problem is if you amoritize faster to help buyers, you have to have early repayment penalties to avoid people simply refinancing out of the later interest payments. People would HATE this because it would make moving difficult.