I have a problem that's deeply rooted in Rust ownership model design, that I try approach from different sides and it's literally taking me months of my afternoon/weekend time. Best part: this problem would not exist in pretty much any other language.
Looks like TDD should be mandatory principle for using Claude Code. Way too often I see AI adding tests that still pass even after I revert the fix they were supposed to verify.
Touchscreens are slick and trendy, but when you need to drive a car or set a bottle warmer for your baby in the middle of the night, you really want to utilise that touch sense. Sometimes good old fashioned physical buttons are the way to go.
In detail: Lattice is a description of a (usually) set that exposes `join` (max) and `meet` (min) operations, satisfying commutativity, associativity and idempotency. Semi-lattice only needs one operation with these properties. Join semi-lattice means that this operation is `join` (`merge` in CRDTs)
I'd say that join semilattice is a summary name for a set of mathematical traits that the state CRDT should satisfy.
Using LLMs for explaining has some value added - as long as they don't lie for their own convenience. But: learning by reading LLM explanation gives you a glimpse of knowledge, not an insight. Similar to reading PhD papers but not actually building stuff. It's more like a procrastination for nerds.
So yeah, we reached the point where I'm starting to introduce changes that don't pattern match with LLM's learning corpus. Now we're starting to run into pit of bizarre decision making.
I have a hard time believing in verification languages. For algorithms, sure. But then the real life comes in, you need I/O, some sys calls and it's like opening the can of worms.
Tbh. I haven't seen examples that verify I/O failure modes beyond simplified simulation.
Tbh. I think that this should be a mandatory in education right now - especially in computer science field. There's no magic, no ghost in the machine.
karpathy.github.io/2026/02/12/m...
I'm pair programming with AI to implement custom LSM tree. It's great feeling to "mentor" it through and discuss technical details.
But man, you really need to be an expert in the field to catch it lying (which it does all the time) and silently skipping the critical issues.
Lol, `genawaiter2` (Rust macro lib to provide yield generators) is not compatible with 2024 edition because `gen` is now a reserved keyword xD
It's sad that this anime season is really one of the best in terms of ongoing series in a long time, yet after work/childcare, all strength that's left in me is to watch a low attention, low commitment ones.
If you want to use LLM for writing a blog post, you could a well just give me your prompt and stop wasting your money and my time.
Some of the best arguments I've seen against #LLM / #AI use:
- undermines experience
- redirects seniors away from teaching
- have limited & biased knowledge
- code has always been incidental for problem-solving
- not conducive to truth seeking
blog.fogus.me/meta/LL...
The part of "everyone becomes a manager" in AI that I didn't really think about until now was the mental fatigue of context switching and keeping many tasks going at once, which of course is one of the hardest parts of being a manager and now you all get to enjoy it too
I used to love coding while in travel (eg. train). Zero distractions, pure focus, being "in the zone".
With AI, I feel more like debugging CI issues. Have to be online, feedback loop counted in minutes, non stop context-switching.
I’m a believer in usefulness of AI but shit like this GitHub issue (and follow up) is just disgusting.
Tech bros trying to not be gross. Challenge impossible.
@pvh.ca speaks truth, but there's one more corner in the design space: you can federate the infrastructure instead of centralizing it.
That's the choice we made for iroh relays: you can run your own (it's OSS!), and endpoints connected to your relay can talk to endpoints connected to other relays.
I used to watch war movies and feel bad for soldiers leaving their families to fight on the frontlines. But after first month of fatherhood (+ work from home), now I think: maybe they were the lucky ones...
Looking at my YouTube subscription feed: 7/9 NDC presentations are all about AI. Something tells me, submitting my talk proposals for this year's edition would be a waste of time.
My recent adventure about plagiarising AI inspired me for a new way to fund open source:
1. Start publishing projects on GH under paid, non-permissive license.
2. Wait for the sources to be scrapped.
3. Make a tool to find scraps of your code in other repositories.
4. Start sending invoices.
You push a commit. You wait. A runner picks it up. You watch logs scroll. Something fails. The error looks like someone fed a stack trace through a paper shredder and then set the shredder on fire. You add a run: env step to see what’s going on. You push again. You wait again. A twenty-minute feedback loop for a one-line change. You do this fourteen times. This is your afternoon now. You had plans. You were going to go outside. The afternoon belongs to the CI now. It has always belonged to the CI. You are only now perceiving this truth.
www.iankduncan.com/engineering/...
It's not about using new Y.Doc, but about swapping one collaborative collection with a new one that uses the old one's output as a (stable) input. I hope I explained the pattern, pros and cons in the last code snippet.
Upon prohibiting it to touch outside implementations and focus on translating local directory... it ignored command and plagiarised my work again. It did the same with C# port and one existing impl.
gl;hf to all the people using AI to write code in a domain where GPL source was available.
I've asked Claude to implement a Rust port of JS library given source code in another directory: there's already one such implementation in OSS, mine. Result?
AI blatantly plagiarised my OSS code, including parts that were not present in source it was pointed to port.
When I'll have a newborn, I won't tell anyone, but there will be signs.
My FOSDEM talk is already online (sound starts at 1:15): fosdem.org/2026/schedul...
Flying back from FOSSDEM (8’C) back to Warsaw (-24’C) is going to be brutal.
