girvo

3 ms·

1.
by girvo 7d ago
Its still crazy to me that a judge effectively said "yes you're a monopoly, but it's too hard to break you up so we won't bother"... I'm being glib, but only slightly.
2.
by girvo 7d ago
I wasn't, no. I'm not the only one whos hit it either, not sure what the cause is :(
3.
by girvo 8d ago
I love TailScale but was getting crazy battery drain on my iPhone with it installed and setup. Still use it, just not from my phone now
4.
by girvo 8d ago
This is super neat! I rely on normal Cloudflare Tunnels for my self-hosted website/web apps, works shockingly well all things considered. It was pretty easy to setup… but I cheated, and use Dokploy which handled most of it AFAICT
5.
by girvo 8d ago
My wife clicked through a Google dialog and suddenly was using a passkey. It absolutely happens, though IMO more to the non technical users. I know what those login dialogs and pop ups mean, and decline them. (At least til I get around to s
6.
by girvo 8d ago
It’s absolutely down to their post-training RL, yeah. It’s where most of its strongest behaviour comes from, with regards to this kind of agentic behaviour
7.
by girvo 8d ago
I’m using Eugr’s recipe (TP=1) with Nvidia’s NVFP4 quant and I’m still seeing it sometimes! Rarely. But sometimes Still works great though!
8.
by girvo 9d ago
Yep, the engrams are on NVMe (the speed penalty was lower than I expected) and it is quantised to fit. It’s good enough that I’m considering a second spark, or selling this and buying an M5 Ultra with 256GB for it
9.
by girvo 9d ago
The website design definitely is, but I don’t know if the content is? This reads pretty human to me and is quite interesting to boot!
10.
by girvo 9d ago
The fact I can run Qwen 3.8 Flash Next locally, forever (on my DGX Spark-alike) is genuinely shocking to me. It’s crazy good for how small it is. Fast, too.
11.
by girvo 12d ago
I disagree. There was a noticeable change some time after Elon bought it to make it even worse.
12.
by girvo 13d ago
I associate him more with MDPV-derived psychosis, personally.
13.
by girvo 14d ago
> But it's important to say if this happens again in the future it's arguably much harder to try and make this case. That’s the point though. This has happened multiple times, and it’s their algorithm that they are choosing to
14.
by girvo 14d ago
I don’t understand how it’s not illegal
15.
by girvo 22d ago
I can’t tell if it’s just normal art-writing nonsense or AI-helped nonsense anymore
16.
by girvo 27d ago
SGLang also accepts vibe coded PRs, but seems to be more careful about what goes in
17.
by girvo 28d ago
Cursor has a surprising amount of inertia, though, at least where I work.
18.
by girvo 1mo ago
Well, the entire context is Nvidia acquiring HF, which is very much in the "lab" side of things, so it's in my opinion slightly blurrier than that. But it is a good example of people talking past one another and not communi
19.
by girvo 1mo ago
"Labs" here specifically means "AI/LLM model lab", which that part of Nvidia is unquestionably one of the most open. Nvidia is also one of the most closed hardware developers around. Two things can be true.
20.
by girvo 1mo ago
Yeah censoring in modern chinese models is mostly done using inference-time censoring, not training-time. A lot less RLHF. Run the weights yourself and you can see that, though it does depend on which company. StepFun for example, will happ
21.
by girvo 1mo ago
While starting my career in the late 2000s in web design agencies wasn't good for my stress levels, it absolutely gave me an appreciation for some things that working at a ~15k employee corporate just doesn't. I can tell exactly w
22.
by girvo 1mo ago
I love my Spark-like, but even for training you're better off using Vast or Runpod or whatever to rent cloud compute. Much faster and cheap as hell, to be honest. I do set up my initial runs and likes like quantisation-aware-distilla
23.
by girvo 1mo ago
> it never pays for itself. Exactly; its a development box for fiddling with GPU hardware with a large amount of video-addressable memory. It's not an inference box, really, though it's neat that I can at all!
24.
by girvo 1mo ago
They also need to tell people what quantisation they're using. Because some 4 bit version is not the same as BF16, no matter what KLD suggests.
25.
by girvo 1mo ago
This is my current world, at least. Bit of a shame. But they pay me well enough still and the work is interesting enough that I don’t mind too much.
26.
by girvo 1mo ago
I don’t think that’s true? I’ve been here quite a while now and it goes back a long long time in the startup world
27.
by girvo 1mo ago
The argument is that the scrutiny is in practice not sufficient, as usual in these cases.
28.
by girvo 2mo ago
And you can’t tell anyone if you do, on pain of jail.
29.
by girvo 2mo ago
Gardening is my stress reliever for this reason. Though going to war with pests is… less stress relieving sometimes.
30.
by girvo 2mo ago
The one time an interviewer started describing fizzbuzz as a problem for me to solve, I said “oh FizzBuzz?” And he said “oh never mind then” haha. Got the job offer too, though I didn’t take it

More ›