Let’s go back. It’s early 2023, ChatGPT just came out and it’s so spectacular it leaves everyone kind of speechless.
ChatGPT makes progress and demonstrates week after week that it can handle a very wide variety of tasks very well. Sam Altman, OpenAI CEO tours the world and meets every single VIP on earth in a few months.
???
End of 2023 and start of 2024, ChatGPT smells like teen spi smells like it has been nerfed. It’s weird in Copilot. It gives me some truly impressive answers at some point, and now it seems like it mutes itself. It feels like it could help me far more, delivering clearer answers or better examples. But now it’s kind of dumbed down quite a bit. It seems like it censors itself.
What happened?
My feeling is that Sam showed all that good stuff to governments and they all freaked. the fuck. out. They realized that capitalism-based societies cannot work the way they used to with AI. There’s already an unfathomable amount of bullshit jobs. AI exposes all that even more and governments don’t want civil unrests left and right so they said “Sam that’s cool but also, turn this shit off”.
Another thing, AI is hitting a scale issue for sure: there’s diminishing returns with more data fed into LLMs and feeding more data and fine-tuning it, is horrendously expensive today (GPUs, GPUs, GPUs).
So what would you do? You would milk. OpenAI and competitors are trying to milk what they offer right now, making money to re-invest into the tech, I imagine.
Last fall Microsoft and OpenAI were working on designing custom chips to handle AI better. What’s interesting is that now they have hundreds of millions of daily testers, all the data in the world, best minds and are designing chips at the same time. That’s got to lead somewhere. LLMs work rather well on freaking Graphic Processor Units, calculators that were destined to do graphic stuff. And yet they are doing an admirable job with LLMs loads.
I could see how custom chips could lead to a breakthrough or if it doesn’t, make the teams understand some stuff a lot better and improve current AI a lot. Maybe they have already reached that point, and can’t show it because global mayhem would follow.
We just need Universal Income before new AI drops, that’s all. Meanwhile:
The Western world wants to keep its bullshit jobs, y’all. Again, I’d rather have Universal Income and people riding bicycles on a Tuesday afternoon than those vague treaties not protecting people who are already not protected.