msdz

6 ms·

31.
by msdz 2mo ago
Okay, strange, but that’s a bug, not a web app “feeling barebones” wrt. core functionality for most users, at least IMO. Has that been reported on kagifeedback.org yet?
32.
by msdz 2mo ago
What do you feel is missing for a free news offering? I don’t personally like it all that much, because it feels like it’s missing the depth I like in well-written reporting, but for those who just want a quick overview of topics once a d
33.
by msdz 2mo ago
This one struck me in particular as unconvincing: > The problem is that C have practically no checks, so any safety checks put into the competing language will have a runtime cost, which often is unacceptable. Why would all possible chec
34.
by msdz 2mo ago
> .... Why would you do job interview when they expect you to run some code on your own system, on your own time? Because both the company and you know it’s the most effective job interview “filter” in SWE roles. > on your own time It
35.
by msdz 2mo ago
Generally I agree, with two caveats/thoughts: Providing not just any a baseline, but a correct and useful one, is ever more important the less the model is grounded in world knowledge – misunderstandings probably compound faster if t
36.
by msdz 2mo ago
Great article, even if it will be interesting to see whether things continue to develop in such a direction or not. > There's a version of this future where the model card stops listing a knowledge cutoff at all, because what's
37.
by msdz 2mo ago
Thank you so much for sharing these. I just laughed my ass off a bunch, and I’ve only yet read this one latest post in the series. My favorite bit, I think (along with the running gag of the constant use cases/connections that somehow
38.
by msdz 2mo ago
Good point, but to be honest, I interpreted the original comment, and worded mine with that in mind as well, to be more a critique of the “model downgrade (from frontier) due to a cybersecurity-related request” rather than a general compute
39.
by msdz 2mo ago
> Maybe it was a downgraded session, who knows. Off-topic, but imagine us collectively being okay with (or powerless to do anything about) this sentiment about any other software service provided like two years ago. How the times have
40.
by msdz 2mo ago
No, as the sibling comment mentioned, your understanding was correct there. What’s more, the only technical difference in speeds could be, and likely also is with the HC1 chip, between prefill (prompt processing) and decode (text generation
41.
by msdz 2mo ago
> At such output speed, I wouldn’t expect reasoning. As the sibling comment to yours mentioned, if they had a reasoning model “hardware-ified” onto a custom chip (as is their plan for IIRC this or next year, a new ASIC), it’d output fast
42.
by msdz 2mo ago
About five hours from now, the sections of Europe indicated on the map will experience a total solar eclipse, with other regions of the continent experiencing a partial one. The map shows feeds to watch it.
43.
by msdz 2mo ago
Now I have to ask – which is the worst?
44.
by msdz 2mo ago
I believe those are car license plate numbers.
45.
by msdz 2mo ago
I imagine at such a low parameter count, there would be little to no world knowledge whatsoever, and the entire focus is on getting the structure of tool calling etc. right…? But yeah, in terms of “physical” AI, robotics definitely comes to
46.
by msdz 2mo ago
Exactly. If you can free engineering capacity from tedious busywork (fixing bugs), you have them available for implementing/delegating to implement experimental features into new revenue streams. Why fire anyone in this scenario? Growt
47.
by msdz 2mo ago
It’s really pitiful, because I was ready to leave a comment saying how actual apologies and taking responsibility have gotten rather rare (mostly because they leave the publishing party with lasting feelings of cringe/shame/other
48.
by msdz 2mo ago
Couldn’t you instruct them to not display as much knowledge for a test situation…?
49.
by msdz 2mo ago
Nice article, but I don’t fully understand why the author argues for such an aversion to static functions. > Moreover, it breaks the systematic approach of sending messages to an instance (often presented as one of the key arguments in f
50.
by msdz 2mo ago
Interesting idea, so I had to check! Prompt: “Top language pick for an Android app? Respond with only the language.” Responses (admittedly via the aichat CLI, not an agent harness; and temp=0): Kimi K3: Kotlin GPT 5.6 Sol: Kotlin Claude Fab
51.
by msdz 2mo ago
> countless of LLMish sloppy-pasta code that's now running and going to be run on man-made JVMs. Don’t worry, I think most LLMs, when starting a greenfield codebase, don’t reach for Java, so it may be a smaller amount of code affect
52.
by msdz 2mo ago
Well at ≈15k tps, I think the bottleneck will not be the model (reasoning or tool calling), and attention will be shifted to the harness’ engineering again. Probably, the usual initial suspects for “what makes computation slow” will becom
53.
by msdz 2mo ago
It’s not reasoning, the hardware demo uses a 3.-something generation Llama 8B. But it’s proven they can automate this (they didn’t etch eight billion weights by hand after all, obviously), so now the interesting question is whether they can
54.
by msdz 2mo ago
> IMO there's no excuse for new software sticking dotfiles in my homedir in 2026. Agreed, and also, tinfoil hat time: I believe they opt for this so that state and config files don’t need to be distinguished (it all goes into ~&#x2F
55.
by msdz 2mo ago
Is it also a solved problem for smaller models, as the article mentioned?
56.
by msdz 2mo ago
Soooo… when can we expect an encrypted handoff.md to fully prevent session portability, then? (Only half-/s)
57.
by msdz 2mo ago
Unfortunately I don’t recall anymore, since this took place at my father’s PV installation. I’ll try to post an update when/if I get exact numbers, though!
58.
by msdz 2mo ago
Oh, I didn’t interpret the above question as asking in that direction; but yeah, that’s of course something I didn’t attempt to answer with my comment. Although I’d be intrigued in the answer to that small-narrow vs. large-broad model quest
59.
by msdz 2mo ago
Thanks for the detailed writeup, these all make a lot of sense. My only other tell (hadn’t committed the time to reading the rest of the paper nor looking into the author yet) so far was Zenodo, unfortunately that definitely has become a te
60.
by msdz 2mo ago
Not who you responded to, but I’ll give it a try. Imagine our LLM as this one-dimensional matrix (instead of unfathomably many, many more dimensions): 0.728201017 -1.9293839303 2.93020202929 5.82290101728 -0.2844920917 … And then, all of th

More ›