Post AT2uSDCXrQdTKICvwG by hackerfriendly@mas.to
(DIR) More posts by hackerfriendly@mas.to
(DIR) Post #AT2qSKjRSjxWriOgxk by TedUnderwood@sigmoid.social
2023-02-25T17:30:52Z
0 likes, 0 repeats
I'm not very interested in the content of #OpenAI 's reflections on the nebulous concept of "AGI." But I am interested to know whether they're putting this out to prepare us for an actual paper about GPT-4 next week. https://openai.com/blog/planning-for-agi-and-beyond/
(DIR) Post #AT2qaHEeLZGCKymYue by TedUnderwood@sigmoid.social
2023-02-25T17:32:18Z
0 likes, 0 repeats
Everyone's whispering that it's a mixture of experts and the new maximum context length is 32K. But I want to knowwwwww.
(DIR) Post #AT2uSDCXrQdTKICvwG by hackerfriendly@mas.to
2023-02-25T18:15:40Z
0 likes, 0 repeats
@TedUnderwood Oy, the hubris of this piece is _killing me_. Society cannot be harmed by "AGI", but it absolutely can be harmed by people behind the corporate shield hosting misbehaving computers on the Internet.Release it already and we'll see what happens. API or GTFO.
(DIR) Post #AT3pZX4g4AEyNN9uWO by Adverb@sigmoid.social
2023-02-26T04:55:39Z
0 likes, 0 repeats
@TedUnderwood honestly feels to me like a weird relation to Bing going so weirdly. Like "we really should probably slow down here" while their partner company accelerates/provokes a LLM-race with Google lolMaybe also relates to GPT-4; hoping it's exciting, even if mostly scaling!
(DIR) Post #AT4J92YzscNSqMclbU by TedUnderwood@sigmoid.social
2023-02-26T10:27:00Z
0 likes, 0 repeats
@Adverb I see how it could be that. There’s a lot in the post about releasing and iterating, and gradualness, which seems to be justifying a “Release it and then fix it” approach. It’s funny how OpenAI is always invoking xrisk to justify its latest decision to release or not release something. But the rationale is different each time.