@tszzl Agreed. This is monopolistic gatekeeping on a civilizational scale. Even if we take this on good faith, the negative externalities ensure a handful of well-funded actors control this technology indefinitely.
Dario give me back my legions!
After four Claude Fable / Mythos prompts trying to crack a simple cryptographic puzzle that Gemini 3 Pro solved in November on its first try.
And it failed each time after guardrails kicked in.
@ahmadaccino This benchmark was only introduced 23 hours ago out of nowhere, giving Opus 4.8 a score more than 2x that of GPT 5.5. It's not known or proven, but included in the Fable blog like it's already a standard.
Anyone else think this is sus?
FrontierCode, a suspicious coding benchmark that no one knew about a day ago (and had Opus 4.8 beat GPT 5.5 by 2x), extensively tested Fable / Mythos, and Anthropic put it eyelevel with SWE-Bench Pro as if it's already a standard.
Something shady is going on.
FrontierCode was introduced 22 hours ago:
Claude Opus 4.8 scores ~2x higher than GPT 5.5 (already sus).
Fable / Mythos already has it listed front and center as the second agentic coding benchmark.
Anyone else think this is benchmaking / benchmaxxing bullshit?
FrontierCode was introduced 22 hours ago:
Claude Opus 4.8 scores ~2x higher than GPT 5.5 (already sus).
Fable / Mythos already has it listed front and center as the second agentic coding benchmark.
Anyone else think this is benchmaking / benchmaxxing bullshit?
@ai_sentience It's a holdover from millennia of organized religion:
We're god's children/made in the likeness of god (we're special)
We have dominion over all other forms of life (license to dominate the world, superior)
We're the center of the universe (universe made to accommodate us)
@daniel_mac8 Two months from now: loop engineering is dead. Meta-looping is the new loop engineering.
We're just moving one decimal point, one layer of abstraction at a time.
@scaling01 I don't buy that Opus 4.8 is 3x over Opus 4.7, or 2x GPT 5.5.
That doesn't line up with anyone's experience, unlike with DeepSWE.
It's either methodologically flawed (it's a new benchmark) or nothing but 4.8 is remotely good at the languages other benchmarks don't test.
@Hitchslap1 Schopenhauer touches on this. People's unintentional demonstration of their intelligence in front of fools makes fools feel worse about themselves, so the fools distance themselves from the intelligent and talk of differences in intelligence becomes culturally taboo.
Todd McFarlane, and by extension, Gen X, is right. AI is a tool, not a competitor. And if you see it as a competitor, you'll be replaced by someone who sees it as a tool.
@daniel_mac8 Two months from now: loop engineering is dead. Meta-looping is the new loop engineering.
We're just moving one decimal point, one layer of abstraction at a time.
@bindureddy AI can't give governments the political will to implement renewables at scale, nor stand up to corruption from the fossil fuel lobbies, or the military-industrial complex.