nabeelqu Profile Banner
Nabeel S. Qureshi Profile
Nabeel S. Qureshi

@nabeelqu

Followers
29K
Following
34K
Media
967
Statuses
8K

make yourself proud

nyc
Joined November 2010
Don't wanna be here? Send us removal request.
@nabeelqu
Nabeel S. Qureshi
9 hours
(Actually, the essay goes way deeper and richer than that. But that’s not for Twitter…).
2
0
14
@nabeelqu
Nabeel S. Qureshi
9 hours
The point of his essay is that myths and stories compress large swathes of context, and eventually once you learn them even just one word from that story can 'activate' the entire circuit in your brain. This is why Plato, in the Republic, says the ideal education is Greek myths
Tweet media one
5
3
64
@nabeelqu
Nabeel S. Qureshi
9 hours
All the Grok stuff reminds me of this throwaway remark of Ted Hughes's that the words "Hitler" and "crucifixion" contain much of the compressed consciousness of our civilization:
Tweet media one
5
11
182
@nabeelqu
Nabeel S. Qureshi
1 day
Freebird Books in Brooklyn, btw. Great spot.
2
0
13
@nabeelqu
Nabeel S. Qureshi
1 day
Today’s used bookstore haul
Tweet media one
8
0
77
@nabeelqu
Nabeel S. Qureshi
9 days
Ok, a few reflections on the book:. 1. qntm defines antimemes as self-erasing information, but this book has a different (but related) definition of the concept: antimemes are (a) high-impact and (b) low transmissibility. Roughly, they are "important secrets". 2. The low.
@nabeelqu
Nabeel S. Qureshi
10 days
Reading @nayafia’s latest. Incredibly good.
Tweet media one
43
119
1K
@nabeelqu
Nabeel S. Qureshi
10 days
Reading @nayafia’s latest. Incredibly good.
Tweet media one
3
5
133
@nabeelqu
Nabeel S. Qureshi
11 days
Wait, you guys actually believed Claude wrote all that code? That was all just Soham.
5
2
178
@nabeelqu
Nabeel S. Qureshi
20 days
RT @nostalgebraist: the more i think about that "agentic misalignment" research, the more frustrated i get. it is deeply, *offensively* un….
0
34
0
@nabeelqu
Nabeel S. Qureshi
28 days
Bobby Fischer’s diet: “two litres of fresh milk, plus two steaks medium rare”
Tweet media one
8
1
60
@nabeelqu
Nabeel S. Qureshi
28 days
I often worry about this with o3 -- it's evolved a bunch of tactics for producing text that looks extremely impressive and correct, but isn't. Basically, there are many more ways to reward hack than to be correct, so in many cases it just reward hacks.
4
0
65
@nabeelqu
Nabeel S. Qureshi
28 days
Terence Tao on LLM mathematicians: "the AI-generated proofs, they look superficially flawless. the [RL] has actually trained them to produce text that *looks like* what is correct. the errors are often really subtle and then when you spot them, they're really stupid. "
Tweet media one
@vitrupo
vitrupo
29 days
Terence Tao says today's AIs pass the eye test -- but fail miserably on the smell test. They generate proofs that look flawless. But the mistakes are subtle, and strangely inhuman. “There's a metaphorical mathematical smell. it's not clear how to get AI to duplicate that.”
9
19
194
@nabeelqu
Nabeel S. Qureshi
28 days
Link to the post
0
0
2
@nabeelqu
Nabeel S. Qureshi
28 days
He also has the same definition of slop as me, i.e. the absence of care:
Tweet media one
0
0
23
@nabeelqu
Nabeel S. Qureshi
28 days
Interestingly, Venkat disagrees with me here, and thinks the right line to draw is just slop vs. not-slop. Probably a better take actually.
Tweet media one
2
0
18
@nabeelqu
Nabeel S. Qureshi
28 days
I like what Venkatesh Rao did here: he made a section of his newsletter that was clearly marked as AI-assisted. Only example I'm aware of at the moment.
Tweet media one
1
1
35
@nabeelqu
Nabeel S. Qureshi
28 days
Anyway, good thing I like reading old books, because this problem only gets worse from here and the only place that'll guarantee human provenance will be old books.
2
1
38
@nabeelqu
Nabeel S. Qureshi
28 days
With a newsletter I think part of the 'thing' you are buying is that the author actually took care over the words, so I think this annoyed me more than, say, if it were a news article or a research report.
2
0
34
@nabeelqu
Nabeel S. Qureshi
28 days
Maybe this is a boomer opinion, I don't know. But the newsletter purports to be the author's actual words, so I feel about the same here as I would if it turns out the author was paying a ghostwriter.
3
0
36
@nabeelqu
Nabeel S. Qureshi
28 days
I was reading a newsletter this morning and realized a few paragraphs in that the author had written substantial parts of it with o3. I don't think there's anything "bad" about this, but IMO writers who use verbatim AI-written paragraphs should disclose the AI as a co-author.
12
8
239