thomasahle

22 ms·

61.
by thomasahle 6mo ago
Good musicians care about music theory / “first principles” as much as good writers care about language theory / grammar.
62.
by thomasahle 6mo ago
I don't know anyway using these models everyday who think they are hitting a ceiling. If anything there's a plateau between each model release.
63.
by thomasahle 6mo ago
It's hard to train models in the open. All the big players are using lots of "dodgy" training data. Like books, video, code, destinations. If you did that in the open, the lawyers would shut you down.
64.
by thomasahle 7mo ago
Sorry, I said that wrong. Estrin's doesn't reduce the number of multiplications.
65.
by thomasahle 7mo ago
Did you try polynomial preprocessing methods, like Knuth's and Estrin's methods? https://en.wikipedia.org/wiki/Polynomial_evaluation#Evaluati... they let you compute polynomials with half the multiplications
66.
by thomasahle 8mo ago
We scaled on "virtually all RL tasks and environments we could conceive." - apparently, they didn't conceive of pelican SVG RL. I've long thought multi-modal LLMs should be strong enough to do RL for TikZ and SVG generat
67.
by thomasahle 8mo ago
To encourage female participation and representation. Most people think it would be good for chess long-term to have a larger female player base.
68.
by thomasahle 8mo ago
Most players actually peak in strength around age 35 [1]. But Carlsen has been number one for more time than any player for him, safe Kasparov [2]: - Kasparov 255 months at number 1 - Carlsen 188 - Karpov 102 - Fischer 54 Bonus nuance: Carl
69.
by thomasahle 8mo ago
On the topic of lecture notes, I can really recommend Scott Aaron's Quantum Information lecture notes: https://www.scottaaronson.com/qclec.pdf
70.
by thomasahle 8mo ago
Maybe they just checked with a compiler and got the same code?
71.
by thomasahle 8mo ago
> This matters because (1) the world cannot be modeled anywhere close to completely with language alone LLMs being "Language Models" means they model language, it doesn't mean they "model the world with language"
72.
by thomasahle 8mo ago
There's a graveyard of 100s of papers with "approximate near linear time attention." They always hope the speed increase makes up for the lower quality, but it never does. The quadratic time seems inherent to the problem. Ind
73.
by thomasahle 9mo ago
I tried Prism, but it's actually a lot more work than just using claude code. The latter allows you to "vibe code" your paper with no manual interaction, while Prism actually requires you review every change. I actually think
74.
by thomasahle 9mo ago
> And also plagiarism, when you claim authorship of it. I don't actually mind putting Claude as a co-author on my github commits. But for papers there are usually so many tools involved. It would be crowded to include each of Claude
75.
by thomasahle 9mo ago
Textual is cook, but it's maintained by a single guy, and the roadmap hasn't been updated since 2023, https://textual.textualize.io/roadmap/
76.
by thomasahle 9mo ago
I'm always surprised that Python doesn't have as good TUI libraries as Javascript or Rust. With the amount of CLI tooling written in Python, you'd think it had better libraries than any other language.
77.
by thomasahle 9mo ago
What about replacing > Haskell provides indexable arrays, which may be thought of as functions whose domains are isomorphic to contiguous subsets of the integers. with > Haskell provides indexable arrays, which are functions on the do
78.
by thomasahle 9mo ago
I'd rather see a programing language optimized for "few tokens". Something like toon, but for code.
79.
by thomasahle 9mo ago
That's amazing :D
80.
by thomasahle 9mo ago
Sure, but then 50% reliability just becomes a matter of whether you can make a strong enough verifier.
81.
by thomasahle 9mo ago
It took Andrew Wiles 7 years of intense work to solve Fermat's Last Theorem. The METR institute predicts that the length of tasks AI agents can complete doubles every 7 months. We should expect it to take until 2033 before AI solves Cl
82.
by thomasahle 9mo ago
Not being a democracy does not give other countries a right to invade you.
83.
Show HN: Make your own Danish Julehjerter (Braided hearts)(juleflet.dk)
3 points by thomasahle 10mo ago|0 comments
84.
by thomasahle 10mo ago
ACR — Automatic Content Recognition: tech in some smart TVs/apps that identifies what’s on-screen (often via audio/video “fingerprints”) and can report viewing data back to vendors/partners. VPPA — Video Privacy Protection Ac
85.
by thomasahle 10mo ago
I assume they are just advocating rolling back the recent roll of legalization over the past few years.
86.
by thomasahle 10mo ago
> I’m wondering if some kids are saying they need this stuff to justify the condition or to play up the sympathy, to make the condition their personality. In a highly competitive environment like Stanford, isn't it more likely that
87.
by thomasahle 10mo ago
Lots of things don't go away, like socioeconomic factors, intelligence differences, not having been tutored in childhood, but we don't accommodate for that.
88.
by thomasahle 10mo ago
I'm not sure we really know how much of learning is memorization. As we memorize more stuff, we find patterns to compress it and memorize more efficiently.
89.
by thomasahle 11mo ago
A bunch of flat-earthers went to Antarctica to see if the midnight sun was real. Turns out it was. Jeran from Behind The Curve was one of the ones to flip, and since then, he's been making videos on how the earth is actually round. He
90.
by thomasahle 11mo ago
From my understanding, Google put online the largest RL cluster in the world not so long ago. It's not surprising they do really well on things that are "easy" to RL, like math or SimpleQA

More ›