throwaway194725

[throwaway194725 @throwaway194725]

@EricDidntDoIt @jack_granahan For the same reason white voters vote for policies that don't benefit them.

[throwaway194725 @throwaway194725]

@ate_bites @AdamHoltererer The numbers are meaningless. What does a score of 60 mean on the Intelligence Index versus 50? The 60 model might be worth ten times more if it can do something the 50 model can't do.

[throwaway194725 @throwaway194725]

@shamash31 @alexmichaelio Google already won on text arena. The bet on Anthropic is on a new model beating Google.

[throwaway194725 @throwaway194725]

@Angaisb_ And you forgot the voice models!

[throwaway194725 @throwaway194725]

@ArtificialAnlys Shocking that Mistral was ever the best at anything. Le Chaton Fat can't come soon enough.

[throwaway194725 @throwaway194725]

@atrupar It's not Mamdani's fault that the president is this easy to manipulate.

[throwaway194725 @throwaway194725]

@acautionxrytale @keysmashbandit The honest answer is that yes, it genuinely is.

[throwaway194725 @throwaway194725]

@PolitlcsGlobal I'm no Israel defender, but didn't we all want to know? I read a list online of which countries left yesterday.

[throwaway194725 @throwaway194725]

@JULIUSSOVER @macklemore Yeah I wonder why they're being traded for food. I heard there's plenty there.

[throwaway194725 @throwaway194725]

@kimmonismus Never forget Opus 4.7. The only Anthropic model to be worse than its predecessor.

[throwaway194725 @throwaway194725]

@krapstarr @scaling01 It's called a logistic curve, and it's what happens when your output variable is bounded between 0 and 1.

[throwaway194725 @throwaway194725]

@cheatyyyy Opus 5 (High) scores a 59 and costs $1.06 per task, almost identical to Sol.

[throwaway194725 @throwaway194725]

@mattvanswol @grok is this real?

[throwaway194725 @throwaway194725]

@DustdrifterX @ArtificialAnlys @Kimi_Moonshot Use GDPVal AA v2. It's like an older, worse version of AA Briefcase, but it has more models, including Luna.

[throwaway194725 @throwaway194725]

@Angaisb_ For starters, nobody pays for Google AI, but everybody uses it.

[throwaway194725 @throwaway194725]

@tukangoprekk @claudeai So far it's meant that when you use Claude Fable, it counts two ticks towards a Fable-specific limit for every tick that it counts against the total weekly limit.

[throwaway194725 @throwaway194725]

@Apollo_EO @synthwavedd OpenAI isn't giving us their best either. GPT-5.6 Sol High is very different from GPT-5.6 Sol Max. And AA briefcase says that on pure rubric store, even Sonnet 5 Max outperforms GPT-5.6 Sol Max. I think Sol is good at narrow tasks, but misses things along the way.

[throwaway194725 @throwaway194725]

@ArtificialAnlys I think it's great that you guys split it, but that means combined the ELO overstates GPT's performance on analytical quality then!

Classic GPT.

[throwaway194725 @throwaway194725]

@scaling01 If they had to actually serve the model, maybe it would ruin their chart position on the efficiency frontier?