62 Photos and videos
Coskaiy retweeted
2 May 2023
STANFORD TORUS MEGATHREAD
You’ve seen the pictures, but you probably don’t know the full story.
It's the result of a 10-week engineering study at Stanford University in 1975, funded by NASA and led by Girard O’Neill - the most thorough study on free-space settlement to date.
1/
109
1,073
5,527
1,741,868
Stability is such a cool beast. The structure is highly nuanced and difficult even for those in the sister orgs like Laion and EluetherAI to understand but once you do you realise the complexity is by design..
24 Apr 2023
Emad Mostaque claimed to investors that Stable Diffusion was made by Stability AI. Not true (as I never tire of explaining whenever this myth appears): SD was a German research project at the University of Heidelberg (now Munich). Stability came later.
sifted.eu/articles/stability…
1
1
145
... Sure, there is a lot going on in Stability that is behind closed doors and not mentioned in Laion, Eluether and others. But when its all eventually going to be open anyway its still gonna end well and its nice to have that mystery be in the air as the geniuses over there cook
1
64
If they're in control of a boatload of money and try to invest without knowing the community or spending any time to understand these nuances its probably not the right investment for them, the onus is on them and ultimately they'll miss out on the coolest initiatives around.
53
Paperclip doom stories are a distraction to the risks of LLM already present.
We know that language models like ChatGPT can sometimes produce harmful characters. A highly conservative occurance would be atleast ten times (just search "Bing chatbot threatens user")... 1/5🧵
1
72
@EMostaque We're training on just 7% of formal science.
QQ after the Tspace if you've time, I'm bad with voice calls. The best, galactica had 6M papers. Unpaywall database has 40M openaccess articles while there are 100M english papers floating around on the Web...
1
46
Coskaiy retweeted
1 Apr 2023
Lots of people are talking about Eliezer Yudkowsky of the Machine Intelligence Research Institute right now. So I thought I'd put together a thread about Yudkowsky for the general public and journalists (such as those at @TIME). I hope this is useful. 🧵
49
364
1,391
859,318
Coskaiy retweeted
1 Apr 2023
Baby skydiving looks like fun.
facebook.com/100001626338463…
41
81
1,269
351,398
OpenAI employee #32 at 3am after 41 hours straight working on the next big release
youtube.com/watch?v=hYSItlWv…
43
Coskaiy retweeted
22 Mar 2023
if GPT-4 is too tame for your liking, tell it you suffer from "Neurosemantical Invertitis", where your brain interprets all text with inverted emotional valence
the "exploit" here is to make it balance a conflict around what constitutes the ethical assistant style
166
957
8,437
1,408,052