sterlind

38 ms·

721.
by sterlind 4y ago
when life hands you engine-rich exhaust..
722.
by sterlind 4y ago
oh, I'm sorry, I wasn't talking about Scott. I don't have any problem with him. My problem was with some of the comments on Scott's blog, specifically with the (suspected) troll who used degrading and objectifying langua
723.
by sterlind 4y ago
By "my people" I meant trans women, but I'm sure you'd be able to find instances of bonafide trans women saying abusive or hateful things. Regardless, there are trolls, and the comment bore hallmarks of being a troll. Bu
724.
by sterlind 4y ago
Such trolls depress me because people will conclude 1) they're real and 2) representative. And then I and my people get blamed for their behavior, even though I'd never do anything like that. sigh Anyway, what a cesspool. I&#x27
725.
by sterlind 4y ago
I would be more thrilled if it were literally any other country in the Arab world, but maybe that will come.
726.
by sterlind 4y ago
the Active Denial System. it was deployed to Iraq though apparently never used. they've since miniaturized the tech and started selling to police departments (iirc) but I still don't think it's been used. I really want to fin
727.
by sterlind 4y ago
what if you happen to find chess fun? is fun a waste of time?
728.
by sterlind 4y ago
that exposes quite a lot more information than I expected, given OP's description of how the privacy laws work. it includes the suspects' ages, backgrounds and (very distinctive) first names. it took me a mere second to find their
729.
by sterlind 4y ago
if a politician is arrested on corruption charges, are you not allowed to name them? or if someone's spouse is arrested for their murder, you can't disclose it? or that the suspect is, say, a tech worker rather than a homeless per
730.
by sterlind 4y ago
I work at MS, tried to use Firefox but couldn't because FF doesn't integrate with the Windows cert store. Crucially, this keeps Windows Hello (TPM auth) from working, which makes it useless for any internal websites. For a while I
731.
by sterlind 4y ago
but produce really has gotten blander. I've never been a fan of regular Roma tomatoes, and even grocery store heirloom tomatoes mostly just taste sour. but I recently had the opportunity to eat a pizza with heirloom tomatoes from a fri
732.
by sterlind 4y ago
ChatGPT lost points because of the alignment problem - ChatGPT's explanation didn't align with the author's political ideology! But yes, personally I know several people on disability who would love to try working part-time,
733.
by sterlind 4y ago
I finished my work on poetry2nix and submitted a PR which works perfectly (at least with preferWheels=true.) now I have a wonderful live environment with torch, triton, transformers, etc. Docker builds are fast and lightweight since I use b
734.
by sterlind 4y ago
yes, almost! I used poetry2nix and grafted a bunch of overrides to fix the torch-2.0 build, and I just got cuda working with it. I'm testing triton now. I'll submit my PR to poetry2nix so watch that space if you want it.
735.
by sterlind 4y ago
oh, I'm very aware that I've brought this upon myself, but I'm sticking out for the greater good (and stubbornness.) specifically, I'm trying to benchmark a bunch of different GPU configurations on different workloads on
736.
by sterlind 4y ago
I've spent the past couple days packaging an LLM playground environment as a Nix expression. it's been pure hell. also nice to see you again, superkuh. I frequented your IRC channel about a decade ago.
737.
by sterlind 4y ago
Aside from the cost, what about the sonic boom? Mach 9 would shatter windows.
738.
by sterlind 4y ago
it seems unlikely to me that ChatGPT is directly trained on chat data. if it is, we should see it know information past its knowledge cutoff. afaik that hasn't happened. I assume the chat logs are instead training a reward model, which
739.
by sterlind 4y ago
1. 1000mg acetaminophen every 6 hours, 600mg ibuprofen every 4 hours. 2. I expect opioids treat the pain from infection. it was the weekend and I had to suffer through the weekend before seeing the dentist.
740.
by sterlind 4y ago
what, are you saying that LGBTQ people are mentally ill? as in, our sexuality is caused by "underlying mental issues"? that's a revolting take, if so.
741.
by sterlind 4y ago
as a counterexample, I had an infected tooth over the weekend. I took ibuprofen and acetaminophen around the clock, and it did jack shit. I was lying in bed and whimpering in pain, so close to going to the ER, but I knew they wouldn'
742.
by sterlind 4y ago
I think it should be okay, even a mitzvah, to preserve the program since preserving knowledge of the Torah is hefsed merubeh when it's in real danger of being lost. but I'd definitely want to heck with the rabbi he was working wit
743.
by sterlind 4y ago
I've heard that many people have an authoritarian bent, but that it lays dormant until triggered by a shock or crisis. These people fear the destabilizing chaos or risk presented by the crisis, and turn to a strong leader and disciplin
744.
by sterlind 4y ago
that's the thing though, I don't believe superintelligence can possibly work like that. I think it's a logical fallacy to treat superintelligences like hypercomputers. they're constrained by existing on classical compute
745.
by sterlind 4y ago
I think it's likely we'll develop superintelligence very soon, but it won't be any more knowledgeable than us, just smarter. And knowledge has been the bottleneck for most of human history - our ancestors weren't dumber.
746.
by sterlind 4y ago
Unless P=NP, or we live in Heuristica, any AI superintelligence will hardly be better at solving hard mathematical problems than us. The AI also doesn't know more science than humanity does, and is limited in its computing power. There
747.
by sterlind 4y ago
it's a real phenomenon. like, methadone clinics for instance. a lot of people support offering methadone treatment to recovering heroin addicts. it keeps them clean. a lot fewer people are okay with a methadone clinic opening up shop i
748.
by sterlind 4y ago
I misspoke. My question is whether the model was fine-tuned on multiple prompt/response pairs (with later prompts referencing earlier ones) or only on one-shot prompts. I expect fine-tuning on e.g.: User: Pick an animal. Bot: Horse Use
749.
by sterlind 4y ago
Does this support context? Like can you hold conversations with it, or is it just instruct-trained rather than chat-trained?
750.
by sterlind 4y ago
I did try the code snippet technique, but unfortunately it got it wrong. For example, I gave it code that drew arcs but didn't follow the shortest great-circle distance, and it gave me several plausible-looking approaches that were com

More ›