vessenes

17 ms·

331.
by vessenes 6mo ago
… and the stock has not dropped 39%, in fact it’s trading about where it was a year ago. Shareholders have been content to let Cohen add to the balance sheet, adjust operations and make a large move. This is one such move. And GME is up 5+%
332.
by vessenes 6mo ago
There is, literally, nothing fishy about this offer. It’s a cash and stock offer from a public company to public company shareholders. We could call the financial or shareholder benefits to ebay dubious (I don’t hold any opinion about this)
333.
by vessenes 6mo ago
Since roughly 1995 or so we've been in a world where quality tooling was provided by on the order of 1,000s of developers, mostly open source. GNU, Xorg, Apache, emacs, nginx, and so on. Or you could opt in to the Microsoft ecosystem.
334.
by vessenes 6mo ago
I learned it in the math context - a sphere eversion is a 3 dimensional process that ends with the inside of the sphere becoming the outside.
335.
by vessenes 6mo ago
If you want to know authenticity, you submit to https://pestcontroloffice.com/auth1.asp -- they verify / deny, and that is the final word.
336.
by vessenes 6mo ago
I had the same reaction. We're headed into a period where you can shape your tools exactly as you like them; artisanal rather than factory-created workshops, essentially. I think the instinct that APIs, validation layers, and so on tak
337.
by vessenes 6mo ago
Yeah I agree — I don’t think I said what I’ve experienced very well — there is a different sort of state you can get in during a large multiagent session. But to me it’s more manic, not ‘smooth’ or peaceful. On a good day hand coding I will
338.
by vessenes 6mo ago
We’re a long way from figuring out how to get flow state with agents. Maybe some sort of less stonerish upper-based master of the universe overseeing 100s of agents manic state. I think there’d be a lot of demand from long time engineers th
339.
by vessenes 6mo ago
This is cool. For practitioners, What’s the current state of the art for free form multi picture to splat? The last time I looked at it the pipeline was pretty janky and included a few separate steps.
340.
by vessenes 6mo ago
Don't apologize. Keep writing and trying things. Ignore the haters and non-curious, listen to the (even if salty) interested. There's a fair amount of talk right now about the value being in the verification layer -- once there&#x
341.
by vessenes 6mo ago
You’re going to love that newfangled M1 chip.
342.
by vessenes 6mo ago
They is doing a lot of work in your sentence. Almost the entire employee population signed a public letter of support with names attached in the middle of the drama. More accurate to say the board I think.
343.
by vessenes 6mo ago
Have you tried the Gemma 4 series, out of curiosity? I haven’t run a local model in a while, but the benchmarks look good. I’d take a free local tool-use model if it was relatively consistent.
344.
by vessenes 6mo ago
I think you mean inference compute? I believe all expert weights are updated in each backward pass during MoE training. The first benefit was getting a sort of structured pruning of weights through the mechanism of expert selection so that
345.
by vessenes 6mo ago
When I was a kid, the Unix greybeards had lists of shell and C quirks ready to go when there was trouble. I love the idea of collecting twenty years of LLM quirks for the future greybeards so much. “Hmm, that vibes vintage 2023 sycophancy —
346.
by vessenes 6mo ago
As always, rooting for these guys — model and national diversity is great. This looks like a solid foundation to build on; hopefully the 3.6/3.7 will dial in more gains. It looks like maybe from the computer use benchmarks that their v
347.
by vessenes 6mo ago
I appreciate the work! And I hear you; writing is a lot of work, especially if you don’t need the benefits of synthesis it gives the writer. I’ve cadged some rules somewhere for de-LLMing the text; if I can find them I’ll post a follow up l
348.
by vessenes 6mo ago
Amazing comment. As a non-hardware guy, I read, “well, duh, for a 20yr practitioner dealing with the intricacies of specific FPGA series, all this makes tons of sense”.
349.
by vessenes 6mo ago
Yeah I find this current LLM voice very tiring to read; I get enough of it day-to-day wrangling claude and others. I don’t think ‘writing’ this took very much work though, it was probably a “read the research logs, and write a blog post wit
350.
by vessenes 6mo ago
https://www.slashgear.com/1769684/ford-model-t-top-speed/ Says that Frank Kulick hit 108mph in 1912 in a modified Model T. (Yeesh). So I’d say we should rate the Edison / 125 comment as at the very least poss
351.
by vessenes 6mo ago
In the original claude degradation followup email Boris mentioned they are upping the percentage of engineers required to use the public version of claude code. I have no idea what percentage this is, or how much of a punishment it is consi
352.
by vessenes 6mo ago
Man, I hated last year’s clanker-tone, and I hate this one’s too by now. I don’t want to read the word load-bearing ever again. The reminder that there’s structure and content and they’re different is a good one. Even a small legal docu
353.
by vessenes 6mo ago
It’s a particular sort of bug that’s harder to detect because … internal Anthropic engineers don’t apply these prompts to themselves, and in fact have access to ‘helpful only’ models that also do not have additional limitations RL’ed in. (O
354.
by vessenes 6mo ago
Interesting questions (and responses). Nota bene - The 1927 Bugatti Type 35 had top speeds ca 125. So, there were cars that fast pre-1930. I have no idea if Edison made, repped, or had anything to do with one such car, though.
355.
by vessenes 6mo ago
Nearly 100%. Let me reframe this for you — “If we find a team substantially closer to AGI than us, we would seek to merge with them.“
356.
by vessenes 6mo ago
Time for a beer obviously. :) I think sora backs my point actually - it didn’t pay and they killed it. As openAI aims at becoming the Facebook of AI (working model that has a permanent free tier) they are trying stuff out, and ultimately cu
357.
by vessenes 6mo ago
Don’t forget Yomif Kejelcha who finished in 1:59:41, a world record up until 11 seconds prior. Amazing.
358.
by vessenes 6mo ago
My experience with models that can reach high 90%-ile benchmark rates on tests is that often that last few percentage is arguable, vague, and often experts would disagree. You could try it yourself by training an MNIST classifier and seeing
359.
by vessenes 6mo ago
Tao mentions that the conventional approach for this problem seems to be a dead-end, but it’s apparently a super ‘obvious’ first step. This seems very hopeful to me — in that we now have a new approach line to evaluate / assess for rel
360.
by vessenes 6mo ago
…and this problem was in fact sourced directly from that list!

More ›