Y
HN Search
Hacker News Search
new
|
comments
|
top
|
jobs
vessenes
searching PlanetScale…
1.
▲
2.
▲
3.
▲
4.
▲
5.
▲
6.
▲
42 ms
·
871.
▲
by
vessenes
1y ago
it might or might not ship. There is an IMMENSE market for $1-2/hr labor worldwide. Even very low skill labor.
872.
▲
by
vessenes
1y ago
I sort of love it; it feels like the equivalent of humans humming when stressed. "Just keep calm, write a song about lowering voltage in my quest to dock...Just keep calm..."
873.
▲
by
vessenes
1y ago
It can be tiring - and I don’t do it all the time. The reason to travel at all for work is fundamentally relational; I find if I am happy with the relationships the travel on balance is a net positive: it lets me work on what I want to and
874.
▲
by
vessenes
1y ago
To be rejected for a job and then pursue a criminal complaint against the person who rejected you has a lot to do with things like rule of law and corruption.
875.
▲
by
vessenes
1y ago
To be fair I had like 300 flight miles July-September, but us PE guys occasionally do work for a living! I currently have deals in Japan and UK and family in the west coast of the US so it’s a realistic number.
876.
▲
by
vessenes
1y ago
Interesting. Upshot - bill sent to Claude, Claude generated questions, human in the loop to negotiate and summarize. Ultimately they suggested a number to the hospital, the hospital chiseled them a few grand, and they settled. Not mentioned
877.
▲
by
vessenes
1y ago
Interesting. I have roughly 50,000 flight miles on the new AirPods Pro 3, and while I do prefer them to the 2s, I too have some annoyances with the left AirPod. In my case it never feels like it fits my ear canal quite right, to the point t
878.
▲
by
vessenes
1y ago
I’ll be interested to see benchmarks. My expectation is that accuracy will take a hit on mid or longer context prompts: I’d bet that the heavy use of JSON in fine tuning will end up impacting quality of a more terse (less reasoning space) n
879.
▲
by
vessenes
1y ago
That might be true, but CZ is a good candidate for a pardon. Did you know before Gensler went after him at the SEC he asked CZ for a job and was rejected?
880.
▲
by
vessenes
1y ago
Hard no. Communications is a human right. I’d say routing communications as a private company is a privilege that can be extended or denied, but this perspective is poison IMO.
881.
▲
by
vessenes
1y ago
You may be dismayed to learn how America got to the moon the first time around as well, then.
882.
▲
by
vessenes
1y ago
Pre-nerf the 4o voice model had a wide range of expressivity, and it would match affect (still tries to do this) and idiolect of listeners if asked. Nowadays there's a list of accents that are considered "hate-ish" and a list
883.
▲
by
vessenes
1y ago
Arthur, question from your GitHub + essay: In GitHub you show stats that say a "cache hit" is 200ms and a miss is 1-2s (LLM call). I don't think I understand how you get a cache hit off a novel tweet. My understanding is that
884.
▲
by
vessenes
1y ago
If the picture offends you, just wait until you read the essay.
885.
▲
by
vessenes
1y ago
I think it's a subtle request to spell it properly capitalized (at least in English): Ai Weiwei
886.
▲
by
vessenes
1y ago
rg is a first for me in that it's a CLI tool that an LLM taught me about -- it's a go-to tool for Claude and codex, and since I got most of my bash skills pre-dotcom-one-boom I'm historically just a grep user. Anyway I'm
887.
▲
by
vessenes
1y ago
Definitely recommend you do a few days of market research before you start the robot design
888.
▲
by
vessenes
1y ago
It does. It's just your worldview is pretty black and white. All of these teams ultimately answer to executives that report to a board, that board is guided by a Chief Legal Officer that engages with regulators among other groups. Some
889.
▲
by
vessenes
1y ago
I'm on the ag robot bandwagon lately. I'd like to see some fruit picking and weeding olympic challenges. Maybe slug removal.
890.
▲
by
vessenes
1y ago
How many diapers have you changed, out of curiosity? I've changed maybe 5,000 diapers (4 kids non-primary caregiver), and I feel confident that a shower head + hair dryer is not going to be safe or in fact work at all in many circumsta
891.
▲
by
vessenes
1y ago
Thanks for the summary. I noticed from the announcement you trained on parallel tool calling to save on serial round tripping. This is awesome. Most LLM coding is so slow that you're permanently out of flow state, and in 'manager&
892.
▲
by
vessenes
1y ago
I mean.. they work within the legal frameworks of very large corporations with nation state engagement. It's not like they're autonomous anonymous DAOs
893.
▲
by
vessenes
1y ago
Well you might be pleased to know that there are large safety teams working at all frontier model companies worried about the same thing! You could even apply if you have related skills.
894.
▲
by
vessenes
1y ago
I thought that too until I turned on the window mode on the new iPadOS. Blech. I gave it a month; I really just didn't want it, even though I thought I did. I would like a full screen shell that's got a reasonable unix variant und
895.
▲
by
vessenes
1y ago
For the haters here - Apple sells roughly 2x as many iPads as Macs (including MacBooks). Roughly as many iPad Pros as laptops. I've cycled through using an iPad Pro as my main device on and off over the years - particularly the cellula
896.
▲
by
vessenes
1y ago
I like sniping - but I could make a product call here to support the messaging - when it's running outside diffusion models and LLMs (as per the press release) we could call that AI. Agreed that they should at least have mentioned Ap
897.
▲
by
vessenes
1y ago
Liquid Glass feels rushed to me. Tons of UI annoyances especially on iPhone - it's suddenly many clicks to get to prior calls for instance, a core way I call people. I'm imagining it will get ironed out over the next two years.
898.
▲
by
vessenes
1y ago
Ooh, this is cool. The Humane was a cool form factor, and I always thought that hand laser projection thing looked awesome. Upshot is the author is a ninja and is building an open assistant platform on the pin, which first requires that the
899.
▲
by
vessenes
1y ago
Simon, I had to run "brew install git-lfs && cd nano-chat && git lfs install && git lfs pull" and then it worked. before then, the model weights didn't get cloned by default for me on macOS. % uv run
900.
▲
by
vessenes
1y ago
This is the spec for a training node. The inference requires 80GB of VRAM, so significantly less compute.
More ›