• mindbleach@sh.itjust.works
    link
    fedilink
    arrow-up
    1
    arrow-down
    1
    ·
    1 day ago

    ‘But DeepSeek…’ No, my sweet idiot child. DeepSeek is not OpenAI, and OpenAI’s latest models only get more expensive as time drags on. GPT-4.5 costs $75 per million input tokens, and $150 per million output tokens. And at the risk of repeating myself, OpenAI is effectively the generative AI industry — at least, for the world outside China.

    Ed’s recent months in a nutshell. Rightly pointing out one company is fucked - aaand then ignoring that other companies exist, which will not be fucked in the same way. ‘Yeah yeah yeah R1 costs a hundred times less even if DeekSeek’s lying their asses off. But OpenAI is the big boy now, so when they finally run out of money, the last four years will unhappen.’

    Meanwhile, the software is out there for the having, and it does things we otherwise can’t. Not everything Sam Altman wishes it did… eppur si muove. It will in fact generate images roughly as described. It will try to answer questions, even if the questions are nonsense. Yesterday someone linked to an imaginary radio station, where every song is written and let’s-say-recorded in the minutes before you hear it. All of that runs locally at a level higher than last year’s hypermegapeptoscale cloud guff.

    These options would sound magical if described circa 2019. Straight-up sci-fi business. We’d talk about cramming it into video games, like we all did when it first emerged for real. Who cares if this bard gives wrong directions? That shit’s why people like Morrowind. Of course it doesn’t guarantee correctness; it’s autocomplete that’s powerful enough to work on sound. What kind of dipshit would confuse that for an oracle?

    There is nothing else after generative AI. There are no other hypergrowth markets left in tech.

    Jesus.

    Ed. My guy. All of this has been for two model types. There’s “which word comes next?” and “which parts are noise?” Those stupid gimmicks can write code and render video. They’re completely the wrong questions, and still function. Finding better questions should drive decades of holy-shit advances. If we don’t stumble into solving consciousness, we’ll see whole new abilities we assumed would require consciousness.

    Will any of them be a trillion-dollar market that deserves to replace everything? No, but neither is this one, so shut up.