I tried to sign-up with Hetzner instance last night - after all the signup etc, it expects me to enter my passport information for "verification". Fuck that.
I re-tried Firefox couple of weeks ago, in the latest episode of a series of tries to "finally" migrating to it. Same CPU fans blowing like jet engines randomly. None of that with Vivaldi (which is anyway all Chromium/Blink underneath) - so came back to it.
Firefox has so many nice things like containers but basic performance issues are still unresolved.
I think that depends on what sites you are using/things you are doing with it. The only time the fans turn on for me is one specific dashboard in Home Assistant. Nothing else I connect to even moves the needle on temperature/CPU use
I first touched a computer after completing my university degree and I still remember the happiness I felt by simply running a DOS command and seeing the expected output.
It does't matter when the plug finds the socket - it is always electric.
I went through the blog. I started using Claude Code about 2 weeks ago and my approach is practically the same. It just felt logical. I think there are a bunch of us who have landed on this approach and most are just quietly seeing the benefits.
This is what the blog writer wrote in email informing about the vulnerability:
> I am offering a window of 30 days from today the 28th of April 2025 for [the organization] to mitigate or resolve the vulnerability before I consider any public disclosure.
> Please note that I am fully available to assist your IT team with technical details, verification steps and recommendations from a security perspective.
He is offering a window of 30 days and that he will consider public disclosure only after that window. He didn't say that this was the full and final window. He didn't say that he will absolutely and definitely disclose. He is being more than co-operative by willing to offer his time and knowledge in this matter, even if he doesn't need to.
If they are not Google, then instead of push-and-shove legal threats, they could have been forthcoming and said something like, "We are not an IT company with expertise in this matter. We will definitely need more than 30 days to resolve this matter. Please let us know if you are agreeable to a longer time Window of <n days> before you consider disclosure."
To top it all, they ask to keep this matter away from the authorities despite:
> The Maltese National Coordinated Vulnerability Disclosure Policy (NCVDP) explicitly requires that confirmed vulnerabilities be reported to both the responsible organization and CSIRTMalta.
So he followed the law and that is bad, how?
> I don’t think cc’ing the national agency was that necessary given the scale of the problem that necessary given the scale of the problem.
Children's addresses were publicly accessible via the vulnerability - does the urgency solely require the matter to be large scale to be taken seriously?
> Maybe should’ve just given them a call and have had a friendly chat over the phone. You would’ve helped them and stayed friends.
The same could be said about the company. Why are only people expected to be nice and friendly while it is fine for companies to issue legal threats?
Github is showing me unicorn - is there an Linux equivalent? I have a old Thinkpad with a puny Nvidia GPU, can I hope to find anything useful to run on that?
Building Llama.cpp from source with CUDA enabled should get you pretty far. llama-server has a really good web UI, the latest version supports model switching.
As for models, plenty of GGUF quantized (down to 2-bit) available on HF and modelscope.
I'm curious how much of "hardcoding" is in the chip? Can it have parts that don't need changing much and "offload" the rest into some sort of high-speed/bandwidth interconnect?
Will we reach a state where we have chips on which models can be "flashed" like CPU firmware?
Or eventually will we reach a state where none of these tricks will be needed because like run-of-the mill Intel/AMD commodity CPUs, we will have full-power AI chips which will be part of an bigger/integrated mother-chip? Then what will happen to companies that do LLMs-as-a-service? Will they be forced to join and adapt becoming hybrid model+hardware shops?
I'm not knowledgeable enough about hardware but throwing these random ideas out in hopes of thought-provoking responses to learn from.
These chips are large by fab standards and even with state of the art processes we likely won't see any kind of integration on consumer tech any time soon, but I imagine they will absolutely see instant demand if they can deliver on what they laid out in the post.
You are in the "not-enough-AI" stage - keep increasing your usage but try to keep it very gradual to avoid entering the "do-more-AI-within-this-budget" stage too soon. It seems like the firing would be in following stages:
"Since the car wash is only 50 meters away (about half a football field), you should walk.
...
When driving might make sense instead:
You need to move the car into the wash bay.
..."
So close.
Interestingly, Sonnet 4.6 basically gave up after 10 attempts (whatever that means).
This was quite accessible. If I had to pick one point, I wish there was more "handholding" from gradient to gradient-descent i.e. in the style of the math-focused introduction of the function with one parameter, two parameters etc that was done. It felt a bit of sudden jump from the math to the code. I think the gentle introduction to the math is very valuable here.
reply