regularfry

5 ms·

61.
by regularfry 2mo ago
For certain specific uses. As a consumer I can still buy Huawei phones. As a business I can still buy Huawei routers, if I feel so inclined.
62.
by regularfry 2mo ago
With their current API approach they're essentially a commodity. They need to start moving parts of the harness behind the API, otherwise they'll remain a commodity. Recursive self-improvement changes the parameters a bit, especi
63.
by regularfry 2mo ago
It's good but (at least on openrouter) it's got an annoyingly tight output token limit. So if it does get stuck in a reasoning pit, it won't work its way out of it in time. It's replaced the Kimi models for me though.
64.
by regularfry 2mo ago
For me it's more an expression of how much mileage you can get out of "glorified", how many tasks devolve down to "if you model language accurately enough, look what drops out" because it turns out that to model lan
65.
by regularfry 2mo ago
The wikipedia definition is: > ...a hypothetical type of artificial intelligence that matches or surpasses human capabilities across virtually all cognitive tasks. You can argue that paperclip maximising is an inevitable consequence of t
66.
by regularfry 2mo ago
I've just grabbed myself a Nape Pro trackball. It's really quite neat. ZMK firmware (which was actually the main thing I wanted) and a configurable positioning angle. I'm still getting used to it but so far it looks like a
67.
by regularfry 2mo ago
Control vectors for the win? It feels like the way to fix this is to pull out a control vector immediately after processing the handbook and use that to steer later inference. That should stop the drift over long distances, but it doesn&#
68.
by regularfry 2mo ago
CSG on meshes isn't really where FreeCAD struggles or focuses. That's more BRep, which is a harder problem than CSG. Blender used to struggle with boolean operations, though. Seems to have got much better some time in the last c
69.
by regularfry 2mo ago
When did you try it? I got a bit confused by your comment because there's no obvious reason why the accelerometers in an android phone would have a problem reporting forward/backward acceleration any less accurately than lateral.
70.
by regularfry 2mo ago
If you're needing to share code for domain objects between microservices something has gone badly wrong.
71.
by regularfry 2mo ago
The very first time I came across the term, it was used to refer to something so small that you'd comfortably rewrite it rather than fix it if it was wrong.
72.
by regularfry 2mo ago
It's not just the modular organisation that you've got to get right though, although that's definitely a big thing. Deployment independence matters too: if I've got a perfectly modular monolith but I have to coordinate
73.
by regularfry 2mo ago
One team owning many services is fine, as long as they can deploy them independently of any other team. Less critical (but still useful) is being able to deploy their own services independently of each other.
74.
by regularfry 2mo ago
This isn't mimicking what a human would do, because it is an exceedingly rare human who would write their own renderer for a task needing a CAD file generating. The human who is capable of doing both is rare, let alone the human who ca
75.
by regularfry 2mo ago
The whole point is that if you don't maintain high quality, any software will degrade to the point where the factory breaks. "Human-quality" here isn't quite the right measure because human software factories also bog
76.
by regularfry 3mo ago
I got usable token rates (10-20tps from memory, so marginal) with the Qwen A10B a while back, well before all the new speculative speedups landed in llama.cpp. There's an unmerged branch which allegedly supports this, but I don't
77.
by regularfry 3mo ago
What's your hardware? I've got 64GB RAM and a 4090 here, wondering if it's worth a play.
78.
by regularfry 3mo ago
Couple of other more businessy reasons: - SynthID hides the watermark in the sampling RNG. No randomness -> no watermark. - If you want to distil on the model outputs, you want temp=0 outputs. No temp=0 -> worse distillation.
79.
by regularfry 3mo ago
The team. The organisation. The reviewer, if they care about team throughput.
80.
by regularfry 3mo ago
That is the context switch I'm referring to. It's worth the reviewer suffering a 15 minute (say) flow state loss to avoid the multiple hours the ticket would otherwise have been waiting to be picked up.
81.
by regularfry 3mo ago
For us, review latency was obviously a problem so we eliminated it (for most tickets, anyway) by making the handover synchronous. Get on a call with the reviewer when it goes into the review column. The win from eliminating review latency
82.
by regularfry 3mo ago
The reasoning trace would be far more interesting, and that's not exposed.
83.
by regularfry 3mo ago
Yeah, it's only a chat template change. So if you don't fancy re-downloading the whole model and can just hack the new chat template into your process, it's a lightweight test to do.
84.
by regularfry 3mo ago
That doesn't make the burden of proof magically mine.
85.
by regularfry 3mo ago
Worth knowing that Unsloth have just put out another Gemma 4 release from Google's upstream updates which should improve reliability. Bugs in the chat template affecting tool calling and other issues, apparently. https://www
86.
by regularfry 3mo ago
I suspect the hard bit is that it sometimes needs to back up and redo, and that's an interface they haven't got figured out. I'm fairly sure I remember Dragon Naturally Speaking doing it in Word years ago though, so the inter
87.
by regularfry 3mo ago
Backwards. The government knows who has kids because they do things like issue birth certificates and operate state schools. In a lot of places it's legally very difficult not to be listed as a parent on government records. So why sh
88.
by regularfry 3mo ago
https://en.wikipedia.org/wiki/Dune:_The_Butlerian_Jihad
89.
by regularfry 3mo ago
I get my team to review my code, AI generated or not. Goes through the same process as anything they produce. Two pairs of eyeballs on everything.
90.
by regularfry 3mo ago
Parsing used to be "AI". If you look at proceedings of old AI conferences you get this impression that anything interesting you might program a computer to do has passed through the field at some point.

More ›