vyr

i'm so sorry. i went through that with one of my own recently. please make sure she knows she is loved.

a countdown clock telling you that you should talk to the model again before your streak expires? that's the kind of UX i'd expect from an F2P mobile game or an abandoned shopping cart nag notification

GPT-OSS 120B (really 117B-A5.1B) is a lot bigger. better comparison would be to 20B (21B-A3.6B).

have worked closely with customer support teams, can confirm that the goal of any technical improvements that go in front of CS agents is to reduce ticket volume, and thus costs. of course they measure retention and satisfaction but ticket volume is always the big one. chatbots were big for this long before LLMs existed.

a fun side effect is that CS is also an early warning system for companies, so when you make it harder to get through to a human, you start throwing out info on your users' pain points. of course this only matters if people have a choice about whether to use your product, so that's gotta be an upside for insurance companies, etc.

i was hoping this was going to be AppCode rising from the grave but nah it's just more rebranded versions of VSCode. nothing new here

because it's not faster than the Ryzen 395's GPU. power efficiency doesn't matter as much as TTFT for desktop users, especially when they're tasking their AMD box as a dedicated inference machine.

some older pre-395 AMD articles suggested it'd be possible to use the NPU for prefill and the GPU for decoding and this would be faster than using either alone, but we have yet to see that (even on Windows) for any usefully sized models, just toys like LLaMA-8B.

that's pretty much correct: these apps all do the same thing under the hood. the difference is UX. Mastodon's built for text and the media features are really barebones. no filters, no crop, no EXIF metadata, no albums. so that was Pixelfed's pitch.

yeah he has a long history of saying dumb shit in public and then trying to cover his tracks.

also, having had to figure out some of the Pixelfed code for previous projects, i wonder if he's up to the task of maintaining any of this once the next shiny thing comes along. Fedi software has a lot of quirks in general (comes with the nearly nonexistent budgets) but as a representative issue, the dude managed to build a photo blogging service with no way to export or back up your photos and that hasn't been fixed in seven years.

ultimately, though, if we ignore software quality and developer reputation, Loops is going to live or die based on whether anyone on Fedi actually wants to make short-form video. given existing Fedi culture, plus how expensive it can be to produce and how the RoI is basically zero, i don't think we're going to see much native to Fedi. some might get crossposted by TikTok/Shorts/Reels creators that want a backup location that won't get erased the second someone makes a spurious copyright claim, but i suspect we're just going to see a few months of stolen TikToks and then not much after that.

they're common in databases and performance instrumentation of various kinds (as are other forms of data structure "sketch" like count sketches) but not as common outside those realms.

i've gotten interview questions best solved with them a few times; a Microsoft version involved spell-checking in extremely limited memory, and the interviewer told me that they'd actually been used for that back in the PDP era.

this looks sick as hell. i wonder whether there are viable NE2000 drivers for PowerBooks running classic Mac OS? modern WiFi (even limited by PCMCIA) might be preferable to era-appropriate WiFi. not much you can get an Orinoco card to talk to these days if you can even find one.