bityard

9 ms·

121.
by bityard 3mo ago
You're saying it's not bribery if the CEO does it?
122.
by bityard 3mo ago
The McDonald's drive through.
123.
by bityard 3mo ago
Does OpenAI not have mandatory compliance training which forbids them from giving any government a bribe as a condition of doing business?
124.
by bityard 3mo ago
Every company I have worked for in the last 15 years or so has required me to sit through mandatory training in which we are specifically told never to offer any government a bribe in exchange for the ability to conduct normal business ther
125.
by bityard 3mo ago
No, that's about right in my (very limited) recent experience. It's _very_ easy to spend as much at a fast food restaurant these days as you would at a sit-down restaurant, especially if you don't do one of their combo meals,
126.
by bityard 3mo ago
I've never corrupted an SQLite database _file_, but an FTS5 index is remarkably easy to mess up if you don't read and understand the docs to the letter and which parts of them apply to your exact use case. When you do a thing wron
127.
by bityard 3mo ago
There are several variants of Qwen 3.6, the MoE models are performant on Strix Halo, but the 27B dense model (the one spoken about in TFA, and generally regarded as the best of the group in terms of quality) is not so performant: https:&#x
128.
by bityard 3mo ago
Halving the precision of the weights is not a free lunch...
129.
by bityard 3mo ago
No, even MoE models need to fit into (V)RAM. MoE has faster inference because only a subset of layers are used to predict the next token, but the set of layers used changes with every token.
130.
by bityard 3mo ago
Quantization is a trade-off, though. The quality, while still perhaps good enough for many tasks, is not as good as the full 16-bit weights that the model was designed for/released with.
131.
by bityard 4mo ago
> it's very easy to add a TFTP server and configure everything to serve whatever you want. In your own homelab or in a small company, sure. But the nice thing about proxyDHCP is that in a larger company, if the network engineering t
132.
by bityard 4mo ago
Crazy theory: Maybe Codex watched too many YouTube videos. I have noticed on YouTube that many younger narrators say, "I quickly did X," when "I did X" would have been more just as well and usually more truthful. Why did
133.
by bityard 4mo ago
Yes, it definitely is.
134.
by bityard 4mo ago
Congratulations on almost 20 people! And sorry to hear about the 351/500th of a person, but I guess these things happen sometimes.
135.
by bityard 4mo ago
It's a surprisingly common misconception that models contain any metadata at all about themselves in their weights. If you ask them, "What model are you?" they either retrieve the answer from the system prompt, or they halluc
136.
by bityard 4mo ago
The real question is: should the file be edited in emacs or vim?
137.
by bityard 4mo ago
The general consensus is that local models will continue to improve drastically, but hosted models will as well. There will _always_ be a pretty big gulf of capability between what you can do with a desk full of hardware at home vs a few ra
138.
by bityard 4mo ago
I'm not sure to what degree you can influence how a model thinks, but you can definitely hide the thinking tokens and tell the model how you want it to talk to you. For example, the Claude web UI has an Instructions field where I have
139.
by bityard 4mo ago
I don't know what timeframe your friend served, but when I was in the Air Force, leadership was constantly making a big deal about FWA (fraud, waste, and abuse). Now I wonder if it was in response to them finding out about schemes like
140.
by bityard 4mo ago
To put it bluntly and perhaps a bit cynically, on the tree of bad things that people do to other people, this is pretty high-hanging fruit. Right up there next to scam phone calls that prey on the elderly while claiming to be from Microsoft
141.
by bityard 4mo ago
pi.dev is more like an agent developer kit. It's basically a substrate upon which you spend hours/days/weeks building your own agents or coding framework. It's pretty much the neovim to claude's vscode.
142.
by bityard 4mo ago
> Turns out they're as human as software engineers. Lawyers start out as humans but something about going into law school and then private practice, and feeding them after midnight turns them into... something else entirely.
143.
by bityard 4mo ago
That seems like a problem an LLM could solve. ;) (Assuming Forgejo has a reasonable REST/whatever API.)
144.
by bityard 4mo ago
From the aesthetic, this has probably been vibecoded to hell and back. (And it looks like the files were all yoinked from https://github.com/6512345/keygenmusic ) Archive.org also has some bundles of keygen music, but y
145.
by bityard 4mo ago
Somebody at some point is purchasing them, probably not the spammers/scammers themselves. It's very unlikely anybody is placing spam/scam calls with regular cell phones when VoIP numbers are easy and cheap to get, and when Vo
146.
by bityard 4mo ago
Well, I tried to file an FCC comment using the link in the article but reCAPCHA doesn't think I'm a real person. I gave up after about completing about 20 puzzles successfully. Our democracy in action.
147.
by bityard 4mo ago
Wha-huh? This is the decade where every company bought into flat design, monochrome, chonky whitespace, extremely radiused corners everywhere. Pulling up my instance, Nextcloud apes that to a tee. You'd be hard-pressed to find any &quo
148.
by bityard 4mo ago
I'm not sure how you are holding it, but I ran Nextcloud on a VPS with under 1GB RAM for many years, now it's on a 12GB RAM cloud instance and is only using a few GB of RAM last time I checked.
149.
by bityard 4mo ago
That works okay if you are the only user. I use quite a few Nextcloud features where access via tailscale is either inconvenient or impossible. My whole family uses the calendar on their phones and other devices, which means they would have
150.
by bityard 4mo ago
How much power does that cluster consume?

More ›