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.
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 can’t tell you why it’s always on the right, but it’s always on the same side because of network effects.
Bicycle frames are not fully symmetric left-right because you need things like a mount point for the derailleur hanger, and optionally affordances to keep the chain off the stays when the wheel is removed.
Those things have to be on the same side as the chain. Bikes designed for disc brakes additionally need a mount point for the brake caliper on the opposite side from the chain.
Additionally, rear wheels are not symmetric: the spokes on the chain side connect to the hub closer to the plane of the rim. That is, they are more perpendicular to the wheel’s rotational axis than spokes on the opposite side (which is why you should always mount a single pannier on the chain side). This asymmetry is to provide space for the gears.
So once the industry decided to put the chain on the ride, you can’t very well make a group set designed for a left chain if you want it to work on the vast majority of frames.
> and furthermore, this is because the drivetrain is ~always on the right side of the bike
While I'm sure this factors into things for advertisements for bike components, there is also just a general preference that westerners have for left-to-right motion. Not just in bike ads, but all ads with (or suggesting) movement. And also not just ads, but movies where directors believe left-to-right motion is associated with progression and right-to-left motion is regressive.
Research has shown that people like to walk counterclockwise (right to left) through supermarkets, which is why they are arranged like this for maximum profit.
Interesting that such a preference exists, and makes sense that supermarkets would therefore be arranged to support this, although as far as I can see this is only to extent of entrance doors typically being "off center" and starting you off to the right. The organization of the store - where the various produce/bakery/deli/frozen-food etc aisles/sections are located seems random from store to store.
It would be interesting to know how people behave if the entrance is to the left vs right. Would they change the direction they walked though the store, or would they just lose customers due to this "awkward" layout?
Since most languages read from left to right, rightward movement tends to read as forward progression. So when showing a bicycle in side profile, having it face right feels more naturally like it’s moving forward.
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.")
The issue is that we change our map file "format" with nearly every release, and handling that without crashing is already a job. For example if we add, say, bomb shelters to the map, an older app using that newer data will have access to it but no idea how to render it, which can cause problems (crashing, errors, blanks, etc.) So we have very specific versioning on a per-map-release / app-release basis, and that's just our project. We've already diverged from Organic Maps and even moreso from the (corporate, closed) Maps.me and living on a totally different data format planet from OsmAnd and Magic Earth. So it's a nice idea but extremely hard in practice.
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.
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.
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.
If you have some other source of parallel data (lots of users, many separate tasks) then speculative decoding might not provide any benefit.
reply