Comment by netdevphoenix
8 hours ago
> f investors were taking seriously the idea that this could be "it" with AI enabling even full automation — not superhuman AI, not even a fast learner, just AI that can fully automate everything we've currently got and not be limited to the subset of desk jobs that LLMs can do OK — it would allow the economy to double in size in whatever wall-clock time period it takes for the AI to gather enough training data by simply observing human workers doing the things the AI has not yet learned to do.
This assumes that the economy and the governments will happily suffer the electricity and financial demands that Sam and co would give and the world would accept. Even if Sam's words were true, it is still a leap of faith because there is no data backing those words.
My take in late 2023 was that GPT-5 would either further push transformers into the "there is still a real chance transformers might be IT" or will become the unquestionable sign that transformers have peaked as a general technology. My take in 2024 was that the AI bubble was going to be noticed. My take in 2025 is that AI companies (other than the big ones like OpenAI) are going to struggle getting funding.
> This assumes that the economy and the governments will happily suffer the electricity and financial demands that Sam and co would give and the world would accept. Even if Sam's words were true, it is still a leap of faith because there is no data backing those words.
Most economic growth since electricity became economically relevant assumes that, the only thing that changes are the names of who (supplies and) demands the energy.
> Even if Sam's words were true, it is still a leap of faith because there is no data backing those words.
Ignore Sam's, assume mine in the hypotheticals of my comment: even something slower than Musk's ongoing failure to make FSD is still a massive increase over the status quo that justifies the price, yet it can do so in a way that's still not going to feel like a radical shift to live through.
> My take in 2025 is that AI companies (other than the big ones like OpenAI) are going to struggle getting funding.
Mine is that there's definitely a bubble overall, but not necessarily in any single player: who wins the pop and who disappears is unclear, but whoever does survive can easily reduce spending when there's no competition and they're consequently not having to put 1/3rd of their compute budget into the next marginal improvement needed to beat whoever else just beat them.