vessenes

12 ms·

211.
by vessenes 4mo ago
Nice to see more literature on HN recently -- Infinite Jest came up yesterday to my delight. Pale Fire is not my favorite Nabokov novel, largely because it's so successful at getting you in the head of someone who just fully and comple
212.
by vessenes 4mo ago
"This is what makes the load-bearing case work," Man I hate that AI writing tic. I appreciate the instincts for sharing the workflow. It's still very difficult to get AI to put an info dense description together though, we te
213.
by vessenes 4mo ago
I guess you could just construct a wav file from the shell and then play it. Agreed doing it all on device sounds challenging.
214.
by vessenes 4mo ago
Having a guaranteed audio channel makes this so much cooler for exploits -- you can exfiltrate over audio!! I love it. I wonder how many of these were sold. I also imagine based on Creative's response (this is fine) that many other dev
215.
by vessenes 4mo ago
Nice. That latency is tough though. On that topic, I wrote an equivalent "magic mirror" type tool that lets you write to the mirror/oracle/VLM on the remarkable and get an answer back. A few possible solutions I explored
216.
by vessenes 4mo ago
I vibe coded hn10k earlier this year. You could choose to see pages with comments only started by 1k+, 10k+ or 100k+ karma contributors. I'm too lazy to keep it up, but I found 1k and 10k both to be better experiences than "vanill
217.
by vessenes 4mo ago
* Gemini 2.5 Pro (no outside resources), and * NotebookLM (not versioned -- with added legal resources). NotebookLM was considered slightly better than 2.5 Pro by the evaluators.
218.
by vessenes 4mo ago
I read IJ in a two (three?) day binge shortly after it was published and never returned. Although I will say I really liked it. Since you're an experienced reader, I'll ask a question bugging me since the 1990s -- my intuition pos
219.
by vessenes 5mo ago
Yeah that layer looping is super interesting. Also, it’s memory friendly with the right inference harness.
220.
by vessenes 5mo ago
This is a super low quality comment. I try to put my current thoughts out here largely because high quality comments refine my thinking. What about the original conversation seems correct / incorrect / interestingly correct or inc
221.
by vessenes 5mo ago
Very Chomsky friendly. Interesting paper, thank you!
222.
by vessenes 5mo ago
Ah, it’s a good time to check in with gwern on our conversation about oAI vs Anthropic: https://news.ycombinator.com/item?id=40816755 and our predictions (ca two years ago). Upshot - poetry expertise does not seem to be the
223.
by vessenes 5mo ago
Since there’s only one inference provider it could be a recycling/ad experiment. The similar usage between trial and paid periods would be explained by this as well.
224.
by vessenes 5mo ago
You have to think about why open models are behind. Exfiltration is a big part of it. So you could change the Nash equilibrium by increasing your security, or other multilateral approaches.
225.
by vessenes 5mo ago
Interesting. Just checked some numbers. So Coreweave has like $3bn in operating cashflow last quarter. Your point is that after we are done with all the capex/interest rate deductions/etc and look at a steady state business for a
226.
by vessenes 5mo ago
Both can be true. A small number dropping off could be a big boost for DDG.
227.
by vessenes 5mo ago
I'm not the original commenter, but I do think this is true, so I'll bite. Datacenters process data, but they do it in a particular location, and therefore are subject to local and national laws. It would be folly for a government
228.
by vessenes 5mo ago
Why don't they pay much in taxes? hyperscalers are pretty profitable.
229.
by vessenes 5mo ago
In any batch inference environment that includes experts, expert routing may vary depending on what else is in the batch. For one thing.
230.
by vessenes 5mo ago
Really enjoyed this, and learning a little bit of combinatorics at the same time. As danbruc mentions below we also would really like our networks to only ever split into sets such that there is at most one set which could include a leader;
231.
by vessenes 5mo ago
> They are supposed to struggle due to chip sanctions As Jensen has been pointing out for almost a year now, these sanctions were ineffective and probably had the opposite effect of the desired goal. The history is fairly long, but an in
232.
by vessenes 5mo ago
Boy I'd like to read a compact non-LLM version of the key concepts here. The signal ratio is very low, and crafted with weird LLM-isms throughout, and very hard to parse. I've been experimenting with mapping a zettelkasten system
233.
by vessenes 5mo ago
One thing that's been interesting to me over the last few years is charting the edge of my coding laziness. As a coder, I'm lazy about boilerplate code -- I hate writing it, I hate maintaining it, etc. And so I design and architec
234.
by vessenes 5mo ago
Have you tried his suggested workflow? I think it's a useful workflow, and if I hadn't found a workflow like this already would appreciate the pointer. I guess he could write a code harness to do this, or gin one up really quickly
235.
by vessenes 5mo ago
I have a very similar workflow, and experience similar temperaments from the agents. I also find anecdotally that they are moderately competitive - you get very different attention from them when you say "competitor X wrote this - plea
236.
by vessenes 5mo ago
For many years my go-to plan has been to stay one point release behind apple's releases, especially the .0 releases -- but, times change. Last night I pushed the button for 26.5, thinking about the Glasswing/Mythos reporting. Seem
237.
by vessenes 5mo ago
Yes
238.
by vessenes 5mo ago
? Starship capacity is 150 (reusable) to 250-300 (non-reusable). Saturn V non-reusable capacity was like 140 metric tons or so. Saturn V cost in today’s dollars roughly $1.5bn per launch. Current Falcon heavy launch from SpaceX is in the $2
239.
by vessenes 5mo ago
I can’t tell if I love or hate the idea of this. On the hate side, ComfyUI is just so, so difficult to use on a normal size screen with a trackpad. It’s designed for someone with a 34” gamer monitor and a mouse with like six buttons, and I
240.
by vessenes 5mo ago
> They get taxed on that! The funds don't get taxed on unrealized gains. Nor do the pensioners. They do get taxed on spendable income they get out of the fund's investments, just like the other owners of the company. > [Shou

More ›