the_duke

7 ms·

1.
by the_duke 6d ago
I disabled it once, it has always stayed disabled. So it may or may not happen regularly, but I would not over-index on a sample size of one.
2.
by the_duke 12d ago
We've had that concept for quite a long time now, in the form of Lora [1] and similar fine-tuning techniques. It first got popular for StableDiffusion to teach the image generation models new concepts. We could easily live in a world w
3.
by the_duke 13d ago
Yeah, I'll do a new release with a whole bunch of local fixes later today.
4.
by the_duke 13d ago
I've had great success with a tool I wrote that can print sparse ASTs for code. https://github.com/theduke/smartedit With the skill installed GPT 5.6 usually automatically uses it, and it reduces code exploration
5.
by the_duke 13d ago
Huge gains on some benchmarks, but for coding it sits barely above Fable It will be interesting to see how it performs in the real world ...
6.
by the_duke 13d ago
500 upvotes with 2 comments would have been a new record. ;)
7.
by the_duke 13d ago
It doesn't reduce the price though.
8.
by the_duke 13d ago
Funnily enough the pricing isn't that much worse than on openrouter, where the best price at the moment is $0.24 in / $2.55 out, vs $1 / $1.5 on Cerebras. Sure, 4x input , but cheaper output. Though Cerebras doesn't hav
9.
by the_duke 14d ago
Google has always done quite a lot of benchmaxing for Gemini.
10.
by the_duke 27d ago
It's already mentioned in the issue...
11.
by the_duke 1mo ago
I guess my main point is that the additional features on top of DeltaDB don't seem terribly important to me anymore, compared to the alternatives.
12.
by the_duke 1mo ago
I'm sure this seemed like a great idea a year ago (they first mentioned it with their Series B). But a lot has changed in those 12 months. Frontier models and coding agents have advanced so much that I don't really see much valu
13.
by the_duke 1mo ago
Yeah, I saw the same thing - quite annoying. It can be mitigated through the prompt.
14.
by the_duke 1mo ago
That's not true, pi auto-compacts when getting close to the context window limit?
15.
by the_duke 2mo ago
Codex CLI was open source from the start. The reason they open sourced this is because grok-build uploaded entire directories.
16.
by the_duke 2mo ago
No clue, I also vouched.
17.
by the_duke 2mo ago
I was investigating the design a little. Two big questions: A) You notably don't write a recovery log (WAL/journal) for things not yet flushed, so data can be lost. Do you have plans to add this? I think it would be pretty crucial
18.
by the_duke 3mo ago
Making something 50-2000x faster is pointless? Besides that, Rust code is actually much easier to maintain , thanks to type system guarantees.
19.
by the_duke 3mo ago
Does this support using S3 as the backing store? That would be very powerful for various use cases.
20.
by the_duke 4mo ago
In my opinion cursor actually has one of the best harnesses again at the moment.
21.
by the_duke 4mo ago
Of course they will. Right now they want to prevent the US labs from gaining any sort of self-reinforcing oligopoly on the space, and to let the ecosystem in China flourish. That will all die sooner or later.
22.
by the_duke 4mo ago
This is often due to (privacy) extensions disabling canvas/webgl.
23.
by the_duke 4mo ago
In codex CLI /status works just fine during a turn. Other things don't though.
24.
by the_duke 5mo ago
I exclusively use thinking mode, which is slower but much more likely to double-check things with web search etc.
25.
by the_duke 5mo ago
Config based hooks seem to miss the mark though? The per-repo config is in `.git/config`, so that can still not be checked into the repo itself, unless I'm missing something? So not very useful at all... I get the security implica
26.
by the_duke 5mo ago
That's in part because many EU countries would like to ship the Syrian refugees back to Syria.
27.
by the_duke 6mo ago
Any recommendations?
28.
by the_duke 6mo ago
You shouldn't be downvoted - LLMs could in theory be deterministic, but they currently are not, due to how models are implemented.
29.
by the_duke 6mo ago
I think many have adopted "spec driven development" in the way you describe. I found it works very well in once-off scenarios, but the specs often drift from the implementation. Even if you let the model update the spec at the end
30.
by the_duke 6mo ago
This doesn't make too much sense to me. * This isn't a language, it's some tooling to map specs to code and re-generate * Models aren't deterministic - every time you would try to re-apply you'd likely get different

More ›