He makes a point, though: bug bounties exist to incentivize people to find and report bugs to a company. We talk about white, gray, and black hats, roughly based on their level of ethics. For black hats – and some gray hats – money is one of the big reasons they look for vulnerabilities.
They are different drugs in the sense that methamphetamine will cause more serotonin release and will act longer than amphetamine, but they will both cause increase in blood pressure, increase heart rate and increase internal temperature. The rest of your comment is right though.
I read your blog and I can't agree, i've installed arch Linux before and only had to read the arch wiki page on it and it worked fine with none of the issues you described.
Same thing with gentoo, that I am daily driving now.
I sure had hickups but I managed to solve them by sitting and reading the manuals/wikis of the distros directly, or the programs I was dealing with. Maybe the problem is that you read an obscure forum to get your answer? I don't know.
Predicting a word is the final objective, as in the output of the model is a probability distribution of the next token. However, choosing the right token is more complicated than just regurgitating the training data (and you won't encounter an exact example in the training data, so you need to interpolate). This makes the model learn abstract representation of things that it is able to manipulate before outputting this back into token. RL also complicates this because the "fitness" is now some arbitrary metric computed over an entire sequence of tokens.
If anyone else is looking to make graph with typst, this can be done with https://cetz-package.github.io/ -- which is inspired by Tikz from the LaTeX world -- or https://lilaq.org/ which seems more appropriate for this type of data plotting
I took a look at the repo and it's probably the fault of the the SVG of the graphs, not of typist itself. Now, you could have used typst libraries to generate the graphs but back then (2 years ago I think?) it was probably a struggle.
The problem is that humans use this as a coping mechanism for things they don't understand: I don't understand why the printer doesn't work, so I give it a mind of its own.
This is harmless for inconsequential stuff like a chair, but when it's an LLM, people should at least understand it's behavior so they don't get trapped. That means not trusting it with advice meant for the user or on things it has no concept of, like time or self-introspection (people ask the LLM after it acted, "Why did you delete my database?" when it has limited understanding of its own processing, so it falls back to, "You're right, I deleted the database. Here's what I did wrong: ... This is an irrecoverable mistake, blah, blah, blah..."
No, the cache is a few GB large for most usual context sizes. It depends on model architecture, but if you take Gemma 4 31B at 256K context length, it takes 11.6GB of cache
note: I picked the values from a blog and they may be innacurate, but in pretty much all model the KV cache is very large, it's probably even larger in Claude.
On HN, I often see comments like this, complaining about Cloudflare blocking access to pages. It makes me wonder if it’s due to a particular setup that triggers bot detection – like Tor or no-JS – that HN readers often use, or if Cloudflare has too many false positives.
I’m more hopeful that MIDI completion/in-filling models will be easier for musicians to control and use. But right now, the most popular tools are things like Suno, where you barely have any control and it spits out an entire, possibly mediocre song. It’s the same vein as ChatGPT image generation vs. Stable Diffusion, where you can do much more controllable inpaints with the latter.