Artificial Intelligence and the Ever-Receding Horizon of the Future
Data & Society Director of Research @jennaburrell says tech leaders want to distract us from the enormous wealth and power they stand to gain from the rise of AI.
https://t.co/vAusDl0995
@Noahpinion "Destroy humanity" but not in the sense of a sudden extinction event, which is what the phrase is assumed to mean without context. Most concerned about slow bleed -- over-reliance eroding incentives to think, create, and to talk to one another; misinformation eroding trust, etc.
i don't know who needs to hear this but RhymeZone's thesaurus has a METER FILTER!!!
this is life-changing. I literally spent the last few months on-and-off trying to code something that would let me quickly look up rhymes by stress pattern, but this is SO MUCH BETTER.
@chrmanning@StevenZapata@StanfordHAI Purely text-based models would seem to have a similar problem, though the authors of the appropriated work may not call themselves artists.
@Noahpinion Are there examples of when you have felt forced to change the words you use, other than by professional editors? Who did the forcing, and how was the force applied?
@Miles_Brundage "Incentive issues" captures it for me. And consequent enfeeblement. Would write more but i can't fit all the nuance into a tweet. I am but a humble rhyming dictionary.
@Miles_Brundage OpenAI's actions suggest that it has already decided that the answer to this question is close to the "forge ahead rapidly" end, despite the lack of regulatory brakes you mention. If you have visibility into that decision, can you say what its framework for deciding that was?
@mrb4prez@pmddomingos Cynical take: Companies can get away with taking labor from millions of creative people to build a service that replaces them, since no individual creator will notice or cause a fuss. Whether they have any reason to create in the future they're building is another question.
@mrb4prez@pmddomingos Unclear, since they haven't deigned to tell us whose content they've taken. But ChatGPT's ability to generate rhyming poetry is probably not due to RhymeZone or any one website -- no doubt, it's an emergent capability that owes more to all the lyrics sites out there.
@timhwang Have you seen evidence that "move to LLMs" is really happening, i.e. that LLM interaction is actually replacing keyword search? (No doubt they are getting a lot of usage, but the numbers coming out of the search engine companies suggest that it's mostly been additive.)
Bark Text-to-Audio Model
Full Text Input: "Why was six afraid of seven?"
Ignore Bark's "I'm done with this input" token and tell Bark to just keep generating more audio anyway.
✨An example of citing sources for similar code, as well as licenses for that code.
I *really* appreciate this nod to open-source authors, and OSS community projects. Not to mention, it's always cool to see that someone has already implemented a function / tool for you! 😁🙌🏻
Which is it's own problem. ChatGPT may *seem* easier, but it's so much less transparent and reliable and prone to including bad info—Google at least includes citations and links. Now it's a choice of a dazzling but often wrong AI or a turgid ad-stuffed search engine. Not great!
@Miles_Brundage@namitchadha I appreciate that this is more detail than is in the technical report. Given the impact of the service I think it's vital that OpenAI be fully transparent about the training set, subject to audit, and prioritize source attribution within the service itself.