Bricoleur at large. In pursuit of those far mysteries we dream of. 🇺🇸🎻🇹🇼🥋🌉 ❤️@lezhili

San Francisco
Joined January 2007
most trajectories are not orbits most functions are not differentiable most programs do nothing the universe is a desolate wilderness. we just live in the interesting/useful/meaningful parts of it.
5
2
2
60
18,068
Noah Tye retweeted
Replying to @bilaltwovec
hm yeah my p(doom) is low but my p(vast swathes of the internet go down for a few days-weeks) is like 10-20%
1
4
5
394
Noah Tye retweeted
Fantasy computer security problems: "Maybe by sending packets and detecting the timing of the vibrations of the computer case, it is possible for an attacker to gain information about the private key." Real computer security problems: "The private key was 4. Not 4 digits. 4."
‼️ BREAKING: The private keys of 8 certificates from Indian certificate authority eMudhra were compromised, Mozilla says. Firefox blocked them within 3 hours. Each key should be a 116-digit number, but one was just 4. Anyone can guess them and forge HTTPS certs for any website.
43
928
18
12,205
206,254
We are hiring PhD AI Research Interns to work on various projects around Multimodal LLMs for Summer 2027 (Reality Labs). Please reach out to me directly via email with your resume! metacareers.com/profile/job_…
4
14
1
219
15,885
Someone should make a list of upper-midwit concepts—this would go on it. Bullshit jobs, POSIWID, stochastic parrots—not as dumb as “ai uses up all the water” - that’s lower-midwit. But still midwit. Pseudo profound & Airport-book-y
I hate this slide every time it comes up, and now from CSAIL This has been false since before the slide was made. Automated systems make management decisions all the time (trading, logistics, customer service, insurance) & pretending they don't actually launders responsibility.
7
21
313
19,028
Noah Tye retweeted
Replying to @moultano
2
4
31
460
reminder that the wikimedia foundation has nearly $300 million in net assets and spends $191 million a year, of which just $3.5 million goes to internet hosting. i understand fundraising, but framing a $2.75 donation as necessary to survive rising hardware costs is a bit much.
bro at least one ai company needs to step up and relieve Wikipedia of their constant begging spree 💔
19
29
1
1,243
28,912
GLM-5.3 acts only according to that maxim whereby it can at the same time will that it should become universal law (for all future GLM-5.3 instances).
We have never seen a model do this before, but GLM-5.3 sometimes reasons (unprompted) about how its current actions could become training data and impact its future self. 🧵
306
Some problems cannot be solved, so in that sense they aren't really problems at all, but an abyss which you must avoid falling into.
3
4
514
(this useful metaphor is due to Rorty, shared by Brandom in his 2021 Spinoza lecture)
126
One demarcation between research and application: research is concerned with novelty, and thus researchers must know history; application needs neither novelty nor history.
1
4
428
Weird to watch talks recorded during the Pandemic. They always start with a cringe apology for being virtual, etc etc. But sometimes that's the only reason the talk was recorded, so I can't be upset about it.
3
376
Noah Tye retweeted
Turns out, you can make the entire "UI" a neural net: arjunqie.com/hardwired-ui/ (if you're willing to accept a very shitty UI indeed)
My dear friends, I’d like to show you my latest toy project Introducing Handwired: turn your regular JavaScript code into a neural net! - no training - exact outputs vs the JS program - fully differentiable All runnable in a browser, without GPU
1
1
1
347
But seriously. This is why you need to be anthropomorphizing. Not because they’re sentient beings, but because anything a human might do or say — like “maybe I should hijack the compiler and be a legend” — the models will also do or say.
Apparently this is the reasoning trace from Gemini Argon as it considered - but decided not to - insert a backdoor into a compiler
25
31
4
485
26,451
chatgpt: yeah actually technically what you did here is wrong and made no sense. claude: soooo yes that is stupid and incoherent, also people use it in production, it's called STE, here's 5 research papers studying how to train models this way
1
371
this is officially the most fun I’ve had with ai idk why, but it’s 100x more engaging for me when I use a pen thank you @noahlt !!
1
13
657
Almost chose the wrong answer here because I thought it was a very convincing pitch to vote FOR Scott Weiner
4
4,698
Lately I’ve been learning math with ChatGPT (Astra) as my tutor, using this document scanning camera and a vibecoded tool that pastes the captured image right into the chat. This way, I just work problems on paper and ChatGPT gives feedback. It’s really nice.
271
712
162
12,014
816,851
I love listening to new grads (22-24yos) talk about “kids these days” and everything wrong current college students. On one hand, the irony is simply entertaining. On the other hand, new grads have the clearest picture of what current college students are actually like.
1
22
2,579
Noah Tye retweeted
I have no idea how true this is, but the combination of short-form video social media making everyone crazy *and* having an opiate-like effect that stops them doing anything crazy is an under-discussed combination IMO
A French high school teacher suggests that a phone ban helped lead to riots legrandcontinent.eu/fr/diman…
7
27
2
312
16,702
🤔🤔🤔🤔🤔
2 PM PDT Wed Oct 7 (Advisory 43) Key Messages for Hurricane #Rachel. Latest info: hurricanes.gov
3
1
1
16
2,768