josh-sematic

5 ms·

31.
by josh-sematic 6mo ago
I’ve always been fond of describing unexplained program behaviors as gremlins. In this case the gremlin was goblins!
32.
by josh-sematic 6mo ago
I wonder how much money OpenAI has spent lobbying for alternative economic models research (or better—spent on such research themselves). Then compare that to how much they’ve spent lobbying to enable deploying this all as fast as possible.
33.
by josh-sematic 6mo ago
This is what the replication crisis looks like in tangible effect. I feel that consensus around major questions should never form without totally independent confirmation of experimental results. Even if not due to intentional fraud, you ha
34.
by josh-sematic 6mo ago
There’s a version of this I could support: - pre-specified age gates baked into the protocol (perhaps just 13 and 18). - account admins on a device get to specify which bracket is associated with the account - an api that allows sites to qu
35.
by josh-sematic 7mo ago
Mythos aside, frontier LLMs can already be used to find exploits at faster pace than humans alone. Whether that knowledge gets used to patch them or exploit them is dependent on the user. Cybersecurity has always been an arms race and LLMs
36.
by josh-sematic 7mo ago
Must be nice to be in a position to sell both disease and cure.
37.
by josh-sematic 7mo ago
Yes it’s LTS but the point is that the LTS system has overlapping support so you can wait on an older LTS for a bit before upgrading to a newer one. And it’s somewhat prudent to do so if you value stability highly, because often a few new i
38.
by josh-sematic 7mo ago
I’m sure you can get fairly close at design time but then need to tune it (I think “regulate” is the term of art) to get it just right before sending the watch off.
39.
How a Mechanical Watch Works(ciechanow.ski)
4 points by josh-sematic 7mo ago|3 comments
40.
by josh-sematic 9mo ago
This is yet one more indication to me that the winds have shifted with regards to the utility of the “agent” paradigm of coding with an LLM. With all the talk around Opus 4.5 I decided to finally make the jump there myself and haven’t yet b
41.
by josh-sematic 9mo ago
> My estimate is that within 2 to 3 years, the lowest cost way to generate AI compute will be in space. I have never been so tempted to join Kalshi
42.
by josh-sematic 9mo ago
Strange to me that they don’t mention HuggingFace, which I think of as a pretty leading player in open{source|weight|data} AI.
43.
by josh-sematic 10mo ago
This is the main factor that makes me leery to try going the full self host route.
44.
by josh-sematic 10mo ago
> long enough to become lost and/or irrelevant For the vast majority of things yes. But thankfully the cream of the crop that does stand the test of time eventually makes it to the public.
45.
by josh-sematic 10mo ago
Can you look at any arbitrary program and tell if it halts without running it indefinitely? If so, you should explain how and collect your Nobel. Telling everybody whether the Collatz conjecture is correct is a good warm up. If not, you c
46.
by josh-sematic 11mo ago
I hadn’t heard of that until today. Wild, it seems some people report genuinely feeling deeply in love with the personas they’ve crafted for their chatbots. It seems like an incredibly precarious position to be in to have a deep relationshi
47.
by josh-sematic 11mo ago
I think some of the point is what you can’t do with it rather than what you can. It’s an intentionally very restrictive protocol.
48.
Elitzur–Vaidman Bomb Tester(en.wikipedia.org)
3 points by josh-sematic 11mo ago|0 comments
49.
by josh-sematic 11mo ago
Clicked the link hoping it would be a new attempt at something like the iPhone mini. Sadly disappointed!
50.
by josh-sematic 1y ago
This unlocked some good memories that have been sitting latent for probably 15 years or so.
51.
by josh-sematic 1y ago
Self-reported data is subjective, but when the very thing you are studying is the self-reported subjective experiences of people then it is actually the only data you should care about. Yes “I don’t feel depressed anymore” is a subjective s
52.
by josh-sematic 1y ago
The idea is that if something would have been a compile-time error (ex: using a method that doesn’t exist), but you don’t see that compile error because you don’t have a compiler, the error is still there . It’s just that you won’t see it
53.
by josh-sematic 1y ago
…and instead you get runtime bugs which is somehow better?
54.
by josh-sematic 1y ago
Yeah the file picker can’t do moves or deletes. Personally that’s a rare enough part of my workflow that I don’t mind doing it via the terminal.
55.
by josh-sematic 1y ago
Regarding open files, <space>+b opens up a picker for the files in the buffer. There is also now a file explorer under <space>+e (little e for starting at workspace root, capital E for starting in the buffer’s directory).
56.
The Tunguska Event(en.wikipedia.org)
4 points by josh-sematic 1y ago|0 comments
57.
by josh-sematic 1y ago
Setting aside whether or not I think it should be fair use, you’re only going to be training a new foundation model these days if you have billions of dollars to spend on the endeavor anyway. Nobody is training Llama 5 in their garage.
58.
by josh-sematic 1y ago
8 works x $3k/work is not a bad payout!
59.
by josh-sematic 1y ago
Somewhat ironic that we have the pleasure of reading Wordsworth’s thoughts for free thanks to the fact that Project Gutenberg is able to rely on the copyright being expired (given that Wordsworth argues copyright should last indefinitely).
60.
by josh-sematic 1y ago
At that point you’re probably spending more money blocking the scrapers than you would spend just letting them through.

More ›