• fifi@piefed.zip
    link
    fedilink
    English
    arrow-up
    12
    ·
    1 day ago

    At the same time they are also pulling towards it because of the potential recursive self-improvement has. It’s not about just the training data anymore, it’s AI finding new training techniques, new kernel architecture and so on. It’s a separate, potential scaling vector as the training data vector has pretty much been maxed out in terms of big benefits.