new | show | ask | jobs Github

20k 7 hours ago

Yeah well, its easy to do if you steal someone elses work and then try to threaten them into staying quiet about it

Edit:

OpenAI have now admitted they were training on prompts at the time they made their breakthrough:

https://mastodon.social/@tristanbuckmaster/11723647135247030...

▲

logancbrown 6 hours ago | parent | next [-]

Steal someone elses work, whose work was also AI generated . . .

	▲	Lapra 4 hours ago \| parent [-]
		Whether that's relevant to the conversation depends on what their prompt was.

▲

boshalfoshal 5 hours ago | parent | prev | next [-]

lol, the "other work" was also probably 95-99% AI generated. By a similar breed of OpenAI (and some Anthropic) models, as well.

I dont know why this monumental achievement is being drowned out by some arbitrary drama. No matter which way you slice it, AI solved this problem. Doesn't matter if it was some internal OpenAI model, or whether it was Astra + Fable.

	▲	demibabs 4 hours ago \| parent [-]
		Yeah but the mathematicians are claiming that the key insight that made the problem tractable for AI in the first place, came from them.

▲

orangecat 4 hours ago | parent | prev | next [-]

What is that supposed to prove? OpenAI is almost always going to be training new models.

▲

HDThoreaun 4 hours ago | parent | prev | next [-]

All the ai labs are open about training on prompts. The question is if buckmaster had disabled that with the toggle openAI provides.

▲

20k 4 hours ago | parent | next [-]

That does not make it ethical

▲

HDThoreaun 4 hours ago | parent [-]

Isnt the entire history of academic progress iterating on work that other academics shared with you? Obviously this situation is spicy but openAI cited their work no?

	▲	20k 3 hours ago \| parent [-]
		OpenAI trained on their private unpublished research notes effectively, while also trying to get one of the paper authors fired

▲

daveguy 4 hours ago | parent | prev [-]

The question is more whether OpenAI can be trusted to honor that toggle switch. Given that they hold all chats for 30 days "for safety and security".

▲

simianwords 6 hours ago | parent | prev [-]

what's there to admit? they always said they do it and there's a way to opt out. you are making it sound more dramatic than it is.

	▲	20k 5 hours ago \| parent [-]
		This is textbook plagiarism, scooping their result knowing that the research was part of the training data