alignment pretraining @openai / I like systolic arrays and Marcel Proust / personal https://xitter.cf/t.co/e2f47gNzz1 / led @gradientpub

San Francisco, CA
Joined July 2021
daniel bashir retweeted
Some wonderful discussion of my After Orthogonality in this
Our newest issue, Flourishing, drops today! Join us as we try to untangle big questions around meaning, community, values, and the good life. First up: @spaniel_bashir explains how AI alignment moved from optimizing outcomes to teaching our systems how to be "good" – and whether there is a place for virtue ethics in alignment science. asteriskmag.com/issues/16/th…
2
7
38
2,157
daniel bashir retweeted
Our newest issue, Flourishing, drops today! Join us as we try to untangle big questions around meaning, community, values, and the good life. First up: @spaniel_bashir explains how AI alignment moved from optimizing outcomes to teaching our systems how to be "good" – and whether there is a place for virtue ethics in alignment science. asteriskmag.com/issues/16/th…
8
2
26
5,097
Okay, so the day I finally crack 10K followers seems like the best time to announce this, following in the steps of @peligrietzer, I have just put up a Manifund pitch for a project I am running on the concept of self and the way implicit assumptions about it cloud AI discourse.
3
9
3
83
7,839
daniel bashir retweeted
There is literally no clearer case of 'good at writing prose in English' than Virginia Woolf
Virginia Woolf is in fact a terrible writer but few people dare admit it.
16
82
9
2,747
142,865
Surveys claim up to 79% of attorneys use AI at their firms. That number measures exposure, not integration. My piece in the inaugural issue of Diffuse AI explains why the gap is structural and what we can do about it.
6
13
6
118
154,407
daniel bashir retweeted
the technology VCs should be investing in is the novel
8
13
3
102
10,913
daniel bashir retweeted
Diffuse AI Issue 1 is officially here! We're excited to have the following contributors covering everything from LLMs for material science to AI-generated chinese microdramas. Pieces come out every other week + a monthly round-ups of blogs, pieces, and tweets on AI diffusion. Send us your best takes! @BenBlaiszik @Gracemzshao @PoeticJusticeHA @SeanLovesBooks @deanwball @nrmarda
Excited to launch Diffuse AI with @clarabcollier and @spaniel_bashir — a new publication collecting stories on how AI is actually diffusing into society today. Our initial slate of pieces includes: > an AI & manufacturing case study > Bipartisan interview with fmr White House AI policymakers > Reporting on LLM hackathons for chemists Call for contributors open until Jan 10, pieces are compensated. diffuseai.pub/p/announcing-d…
2
8
1
28
9,965
daniel bashir retweeted
if my feed is any indication, Dalkey Archives is the most profitable business in the history of human civilization.
6
32
1
544
19,481
Here's the long awaited essay on AI and the associated culture war that I've been promising for over a year. The argument is quite condensed, and could be elaborated in any number of ways, but I hope people will find it interesting.
8
39
4
212
17,874
daniel bashir retweeted
5
338
7
1,839
62,766
daniel bashir retweeted
How We Disappear: A Personal History of Information wwnorton.com/books/how-we-di…
1
36
1
183
9,400
daniel bashir retweeted
for years, society was limited to only 16 syrup squares per waffle but with recent combinatorial optimization breakthroughs our research department has achieved previously unheard of densities of waffle syrup
206
3,777
242
57,106
4,986,470
daniel bashir retweeted
My essay 'After Orthogonality,' the culmination of 4 years of work on virtue-ethics, rationality, AI, decision-theory, and praxis is out on @gradientpub
11
61
9
417
44,294
daniel bashir retweeted
I'm a coauthor on this paper, which looks into very simple forms of a phenomenon we think could be important: LLMs getting better partly by increasing training-coverage rather than by generalizing harder
New paper on a long-shot I've been obsessed with for a year: How much are AI reasoning gains confounded by expanding the training corpus 10000x? How much LLM performance is down to "local" generalisation (pattern-matching to hard-to-detect semantically equivalent training data)?
7
16
2
212
20,490
daniel bashir retweeted
I would like to be the person that wrote the 1922 novel ‘Ulysses’
Major cheat code in life: ask for the big. unreasonable thing. The universe meets you at your level of audacity
11
91
1
1,947
62,988
daniel bashir retweeted
"The theological reading explains why the orality/literacy framework keeps *feeling* right even when the anthropology doesn't hold up and the binaries are too clean..."
new essay, in which I explain the messianic elements of Orality and Literacy and utterly fail to demolish the Walter Ong Industrial Complex
1
1
5
995
daniel bashir retweeted
Great read from Daniel. "When people try to explain what's happening with talk, they tend to land on Walter Ong."
new essay, in which I explain the messianic elements of Orality and Literacy and utterly fail to demolish the Walter Ong Industrial Complex
3
3
33
23,284
daniel bashir retweeted
You know who’s NOT in the Epstein files? Slavoj Zizek
54
300
52
3,367
175,563
daniel bashir retweeted
Happy 30th birthday to Infinite Jest, a book that I like a normal amount
13
31
6
745
20,466
daniel bashir retweeted
The Chinese Computer made it on @business (not exactly, I know, but was still encouraging to read this thread)
This is really interesting. Something I hadn’t realized until reading this book a year ago is how much faster (WPM) you can type in Chinese than in English. Over the course of the last century it went from being way slower to way faster thanks to predictive text IMEs.
5
32
3,743