Joined July 2019
- Tweets 13,212
- Following 193
- Followers 182
- Likes 15,906
119 Photos and videos
Pinned Tweet
19 Jul 2022
Years ago I participated in "fight apathy" on the delusion we would all sit in a bar discussing nerdy things like corporate integration.
Now no ones apathetic and everyone is discussing stupider things. I regret my participation.
2
1
15
“Claude 4.8 please do this”
“can’t do it sorry”
“Why not???”
“The code as is exists today doesn’t to that”
“SO ****ing change it!”
Louis CK meme but Dario pls work things out with the admin I want Fable back pls
2
1
67
(Claude 4.8, once again, made inferences about how an API response worked from how I was processing it, which is infuriating because it had the API spec available, refused to read it, and I was asking to change how the response was processed)
21
Jun 14
Even as Opus got smarter and smatter, you can kinda tell when someone is relying on intellectual crutches? Less syncopathic, but aggressive adherence to the status quo
Fable 5, it felt to me, was in its own class, not just “smart” but also “intellectual”. The bans a shame.
52
Jun 10
Evergreen: whenever the discourse on here is like “the economy is fine but everyone’s pissed off what a mystery” and then rambling about mass delusion and ‘the media’ the answer to that is to reject the premise, the economy is not fine
20
“The other is listed as 6’2”, and is certainly shorter”
Damn NBA announcers throwing shade before the game begins
27
May 31
HS, reunion, met my old math teacher who I did research with:
“Hi Aladdin, how are you? So nice to see you! What are you doing nowadays”
“Hey Mrs. <>, I’m doing well! Im at a hedge fund right now”
“Oh that’s great” <glares>
1
42
May 28
It is absolutely essential that tech companies are able to write blog posts and especially post mortems without months long approval processes. Lawyers, who job it is to manage legal risk, should understand the risk of not publishing is to destroy the culture of the org.
May 27
Sometimes I think those 3 years in law school were a waste. Then there are times I interview lawyers for our team and can effectively screen for an understanding that there are some things more important than not getting sued.
1
1
137
May 28
It is also essential that tech companies be *honest* in those post mortems, and if any legal risk comes of it so be it, find a solution that doesn’t involve withholding info or being dishonest by omission.
1
28
May 28
Due to a product being popular, demand for the product explodes, bringing the price up. People don’t like paying high prices, so we can predict the price will go down.
Econ 101 question: explain why this is wrong
106
May 27
It was mentioned when ATT and “privacy restrictions” were enacted that they didn’t safeguard privacy in a meaningful sense people care about, and would lead to stuff like this. The model doesn’t work if people don’t pay for it
May 26
I think the worst thing about the contemporary Internet is that nearly nearly every serious source of news is at least partially paywalled—and in any case, all of the major social media platforms now shadowban such links.
1
44
May 27
Google propped up the publisher model for *years* and for their efforts they got sued over and over with all the antitrust idiots on this website cheering them on so some retrospective is in order perhaps
1
21
May 27
“We’re going after Meta and Google”
“No, your going after meta in the short term, in the long term they will restrict links and you are killing the model you depend on and they are supporting”
“Why are you defending Meta”
It’s tiresome
16
May 27
No matter what happens to the job market with this economy and AI thing and the layoffs the Tony Brothers will still have a job somehow.
May 27
I guess the refs decided this game was over
2
184
May 26
“Due to AI, we are conducting layoffs. We are also taking admin access away from developers in dev, now you need to get approvals from X and Y teams. Also we fired the guy nominally in charge of approvals”
Is this the new lean efficient company I’ve been hearing about?
23
May 25
As impressed as I am about AI capabilities and I’m generally an optimist about AI capabilities, I do think there is something here. You can sort of get it to explore out-of-distribution areas with randomness but … the smartest models still aren’t great at it?
May 25
The other theory says there is something more fundamental missing: AI systems are worse at “creativity” more broadly; perhaps humans have better heuristics for out-of-distribution search.
1
68
May 25
“Creativity” requires a deliberate exploration of states that aren’t in high probability spaces; and if you try to “force” LLMs down those roads it very quickly loses any coherence whatsoever
1
20
May 25
Anyway, probably cope! And the problem is easily solved by just trying something millions of times eventually you will hit a low probability space. But … idk, again it’s just the sense I get
16
May 20
I understand, based on LLMs work, why this happens, but … Claude just found a bug in a data processing script based on the filter being incorrect. Correct! But when I investigated, it never had access to the table schema, it just guessed and got it right!
That’s weird!
3
341
May 20
Not defending this, but in general imo there are too many citations in papers and not everything needs to be a literature review.
If every paper cited only 5-6 other papers and the rest was common knowledge, you wouldn’t have this problem.
May 17
Have you read all the sources you ever cited? During my PhD we, along with dozens of other papers, cited a paper that I later found did not contain the result for which it was commonly cited. I should be banned I guess.
1
1
134