daemonologist

For personal use I simply need less than $20 of tokens, even at API rates. If they (Anthropic or OpenAI) offered a $5/month plan that gave you ~$50 of credit I would probably sign up.

Personally, it takes me longer to write the specifications than it takes the model to implement them (and it takes me much longer to review the resulting code, although maybe that makes me old-fashioned). Consequently I do not have enough tasks to run more than one agent at a time.

(2024)

Yeah, I'm thinking the nigh-unreadable AI-speak we get these days makes a lot more sense if this is what they're training on. Or maybe the author has translated the prompts from another language?

My reaction precisely when I first heard the specs for this thing. Espressif works in mysterious ways.

2412 x 2288 per eye (and horizontal FoV is around 70 degrees). Not sure it's really enough for monitor replacement, but certainly an upgrade over Xreal's offerings.

I think they're calling them "VR" because they use camera pass-through for AR, whereas all their other AR products are see-through. (I think this is a reasonable place to draw the line - pass-through AR kind of sucks in my experience. Although maybe a third category would be appropriate.)

They've been using that branding since last June (with the reorg into "Meta Superintelligence Labs"), for better or worse.

Rumor is they are going to do a Quest 4 (late next year).

I agree that these seem like a stopgap short of the thing everyone wants - Vision Pro specs (ahaha) in a pair of regular glasses.

Personally, as a glasses-wearer, I'm not quite convinced it's the right target though - even my 20 gram glasses are uncomfortable to wear all day and I take them off whenever they're not needed. I'd rather have a ski-goggle device (similarly light, but with the weight more evenly distributed, and still with optical see-through) with fewer performance compromises. Of course that wouldn't let the manufacturer dominate "the next form of personal computing" or whatever.

I don't know if there's an overall limit, but their regular tunnels have a limit of 100 MB per request which breaks stuff like Immich.

I get the impression that Nvidia employees don't care too much - I started seeing fully AI-written "documentation" on some of their smaller projects more than a year ago (i.e., before it was even slightly a good idea).

People definitely like teepees and yurts (and thatched roofs, dry stone walls, log cabins, etc.), it's just that at some point the practicality gap becomes too wide to contemplate.

But this is a reasonable point to raise. Although both would serve their purpose a concrete viaduct is much cheaper to build than a masonry one, to the point that it may be difficult to justify the latter no matter how strong the public preference is for it.

Agreed; not writing your own prose is lazy and disrespectful. If you expect/hope that humans will read something, you should demonstrate that it is important to you by spending your own time on it, and you should respect their time by being as clear and eloquent as possible (which Claude et al. are decidedly not, for now anyway).

Yep, across a smaller fov too (100 degree horizontal).

You could do it, but the resolution isn't high enough for it to be a good experience. (2160x2160 per eye sounds like a lot, but with a 110 degree fov it's only ~19 pixels per degree. Which is like sitting very close to a 27" 720p monitor.)

Yeah, the blog post itself loads for me but everything else is really struggling.

You're correct that it needs to run the full model to "verify" a token, but LLM inference benefits from batching - it's much faster to run twice in parallel than sequentially. So the draft model runs ~2-5 tokens ahead, and the full model then runs ~3-6 batches in parallel using those tokens, and can skip ahead by however many results match.

If you have some other source of parallel data (lots of users, many separate tasks) then speculative decoding might not provide any benefit.

Huh, it's true (well, technically the price was 100,000 hugging faces); good catch

Model's not available to the public (or even to Simon, I guess) yet.

and furthermore, this is because the drivetrain is ~always on the right side of the bike - if you want to inspect or admire a bicycle you look at the right side, as you might look under the hood of a car.

(Why the drivetrain is on the right, I don't know. But most bike parts follow open standards so it's quite entrenched.)

I doubt it - AWS hasn't added any non-western models since GLM 5 and MiniMax M2.5 in February, afaik. Might be a deal with OpenAI (GPT 5.4 was the first to be available via Bedrock, in April) or might just be that there isn't a lot of demand due to corporate skittishness around models trained in China.

The no-Google-calendar-sharing-on-mobile is an absolute classic. I have personally seen multiple clumps of people huddled around a phone trying to figure out how to share a calendar with each other - fruitlessly until I butted in, because normal people don't know about desktop mode and even if they did would have no reason to think it necessary.

(If you didn't read the post, it doesn't work in the app either. It only works on "desktop.")

Thanks, but I was interested more in the cross-app communication aspect rather than the data formats ( https://codeberg.org/comaps/comaps/wiki/Content-Provider ).

I once looked into building an app which piggybacked on CoMaps/OSMAnd's data (saving the user from having two copies of their maps, and me from having to duplicate the UI to manage them), but found that in modern Android it was basically impossible for even the user to access data/files that "belong" to an app, without rooting the device.

Neat, I was not aware of this mechanism. Is there any way to get map (tile) data from CoMaps into another app?

You could probably score 100% on ARC 3 if you were motivated enough. I find some of the current problems to be kind of like Chess - mechanically simple, and ~solvable, but it's difficult to force myself to think at length about a monotonous and meaningless problem. The machines do have an advantage on the "energy" front; they've become almost psychotically persistent (and don't get tired after too many prompts).

Anyway yes I think we've had AGI for a while now, even if the GI doesn't quite match up with what we expect from a human.

The Glacial National Park referenced by GP is in Montana; you were perhaps at Glacier Bay National Park.

Regardless, some beautiful photos, thanks for sharing

Bedrock seems to have stopped adding new open-weights models, and mostly only has Anthropic and OpenAI stuff now. You can get Qwen 3.8 directly from Alibaba: https://www.qwencloud.com (proprietary variant) or from DigitalOcean (this variant, probably also from others soon).

On your 5090 you could easily run a smaller model like Qwen 3.6 27B: https://huggingface.co/collections/Qwen/qwen36 or Gemma 4 etc., or as mentioned there's a Qwen 3.8 27B coming out in a few days.

On the first point, I think it's a search of boat names, not boats - they're deduplicated.

Also (mentioned by Tom), this elastic-powered supersonic trebuchet is an excellent video: https://www.youtube.com/watch?v=gdXOS-B0Bus

Most phones won't boot without the battery. Obvious workaround is to connect a power supply (and large cap) directly to the BMS, but in my experience most phones will shut themselves down when they think they've drawn more than the capacity of their battery, even if the voltage is still high. You can sometimes get around that by rebooting the device every couple of hours (and there might be a software/firmware fix in some cases?), but it's a lot less annoying to just accept a small risk of a house fire.