We’re probably seeing demand destruction. I’m curious if the thin-client thing that a lot of these companies seem to want will actually take off. I’ve also been noticing a lot of “cyberdeck” and retro handheld stuff lately; I wonder if that’s related.
Mastodon has some weird etiquette I don’t quite understand. Maybe it’s because I’ve never really used “microblogging” or much social media outside old forums. Most people seem to use it for self-promotion. If you respectfully disagree on some part of someone’s post or just try to be helpful and add more context, sometimes the OP gets upset (I’ve heard the term “reply-guy” mentioned on there before, and also seen “experts” pulling the “do you know who you are talking to” card when others try to add context). There is little discussion and interaction from what I’ve seen; mostly just posts without meaningful replies.
Politically, Lemmy seems to lean left-ish. Mastodon leans liberal, but there’s more variety than Lemmy (I guess since interaction is low). I haven’t used Bsky, but I saw when Stephen King announced he voted for Platner he got dogpiled (dunno if that means Bsky leans establishment Dem that would prefer zionist and corporate Janet Mills or what).
Unlikely. The US executes people when there’s so much new evidence after sentencing pointing toward them not being guilty that the original prosecutors plead for them to not execute. The Dem establishment’s complaints against ICE are mostly just procedural; they just argue for things like more transparency for these detention centers, not their abolishment.
“Small,” non-generative, BERT-like models, would probably be more appropriate.
Anthropic does sentiment analysis with regex in Claude Code though ¯\(ツ)/¯
Tesla FSD is not safer than real drivers. All robotaxis are level 4 at most, so it’s an “apples vs oranges” thing. It’s like saying trains are safer than real drivers. By all accounts, the Tesla robotaxis are the worst that exist, and they’re barely being used where they’ve been rolled out.
I will give credit that Tesla forced other car manufacturers to innovate, but they’re outclassed in every way now.
I only tried it a few times in the web chat interface. Seemed worse than Sonnet in that environment. Probably needs more reinforcement tuning for chat. Guessing it was tuned more for Claude Code.
I don’t. But, I don’t think posts about oral sex are typically the top post on the equivalent of “All” on other social media, as this post was. I could be wrong. I routinely see sexual topics on All here (sorted by Active). Nothing wrong with it, just making an observation.
Lemmy is horniest social media I’ve ever used.
Is it people deleting specific posts, or people deleting their accounts? I do find it kind of weird that when you delete your account, it deletes all the data you contributed. Maybe it’s the right choice though, IDK.
14 H100 GPUs per user that they serve
Not per user, but probably decent rough estimate to that per vibecoding dev that is continually running agents 8+ hours/day. Some people’s “workflows” involve running multiple parallel agents sometimes or even a significant portion of the time (using the git worktree feature), so I think that’s probably a decent rough estimate. I imagine the limit would be serving 10 of these types of “devs.” Of course, there’s batching and stuff that can be done, but I think it still slows everybody else down near linearly. H100s aren’t the only accelerators used for inference; I just chose it as an example. Google has ~5 million H100 equivalent accelerators, Microsoft has 3.5 million, and Amazon has 2.5 million (https://www.networkworld.com/article/4156949/google-owns-the-most-ai-compute-and-it-built-it-its-way.html).
Probably more expensive than the subsidized costs. Hmm…
H100 GPUs cost $25k, and have 80GB of RAM. Kimi k2.6 has 1.1T parameters. Assuming 8 bit quantization, would need 14 GPUs to run a single agent at a time (I’m not sure the cloud models use quantization; it could be double). So, $350k per vibecoding dev on GPUs alone. Life expectancy is ~4 years, so ~90k/year amortized. This is ignoring the significant electrical/HVAC cost of handling 10KW of electricity and heat per vibecoding dev (and tons of other costs as well).
I’m sure the search engines log the IP, and with enough searches could be correlated back to you fairly easily even if you host it on a VPS? Unless it does some kind of mitigation I’m unaware of.
Qwant is French. Not a privacy search engine and looks like it features AI now. Kagi is really good, but US based and expensive. I think most of Europe does intelligence sharing with the US (“14 eyes” or something like that)?
AI companies offer HIPPA compliant services. They have zero data retention policies (supposedly).
It’s possible to do some server side AI stuff with decent privacy guarantees. Apple utilizes homomorphic encryption to do stuff like landmark ID in photos, which is neat.
I’ve heard the excuse that they’re private property so they can’t. They’re not hard to remove though. They’re usually just attached to poles with pipe clamps.
The trend I see are the Mac Minis with a lot of unified memory. These are typically very well off people though. Prices for even old GPUs like 3090s are ridiculous now. I don’t think connecting 2 machines over Ethernet would work well, but putting 2 GPUs in a single machine does.
I think they know it’s a somewhat viable option and is part of the reason they’re doing the hardware cartel/circlejerk thing.
Voron Micron or Salad Fork kit.
I thought enshitification was the specific process of “platforms” gaining a large market share, then exploiting both the buyers and sellers that use the platform to jack up profits/extract more rent.
I’ve heard “loops” will burn a lot of tokens. Haven’t tried it myself. A person could also spool up multiple loops to work on multiple branches at the same time.