Hacker News new | past | comments | ask | show | jobs | submit login

I think what's silly about "shepherding responsible AGI" is this is basically math, it's not some genie that can be kept hidden or behind some Manhattan Project level of effort. Pandora's box is open, and the best we can do is make sure it's not locked up behind some corporation or gov't.



I mean, that's clearly not really true, there's a huge "means of production" aspect to this which comes down to being able to afford the datastructure infrastructure.

The cost of the computing machinery and the energy costs to run it are actually massive.


Yup it's quite literally the world's most expensive parrot. (Mind you, a plain old parrot is not cheap either. But OpenAI is a whole other order of magnitude.)


Parrots may live 50 years. H100s probably won’t last half that long.


Parrots are very smart animals that understand at least some of the words they learn. I wish people would give llms more credit.


Sure but I meant the costs are feasible for many companies, hence competition. That was very different from the barriers to nuclear weapons development.


Are you sure this is the case? Tens of billions of dollars invested, yet a whole year later no one has a model that even comes close to GPT-3.5 - let alone GPT-4 Turbo.


> yet a whole year later no one has a model that even comes close to GPT-3.5 - let alone GPT-4 Turbo

Is that true and settled? I only have my anecdotal experience, but in that it is not clear that GPT-3.5 is better than Google's bard for example.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: