It's very important in my opinion to have Codex review Claude's work. They make different types of errors and both get tunnel-vision. I prefer to drive with Claude, but I set up this hook to automatically invoke a Codex code review every time Claude tries to commit. I find it dramatically improves the performance of either of the models individually.
You can just paste the entire gist into your Claude session to have Claude set it up for you
Same. My app is a PWA. Most users won’t install a PWA and won’t repeatedly navigate to a website so it limits the reach. But the advantage is that I can deploy instantly. I love when someone sends a bug report and I can tell them it’s fixed ten minutes later. Pretty great, compared to “it’ll be fixed in there business days” you get with the iOS app store
Does marginalia_nu not use embedding models as part of search? I guess I assumed it would. If you have embeddings anyway, decision trees on the embedding vector (e.g. catboost) tend to work pretty well. Fine-tuning modernbert works even better but probably won't meet the criteria of "really fast and run well on CPUs". That said, the approach described in the article seems to work well enough and obviously provides extremely cheap inference
It does not use any transformer models right now. I've made experiments with BERT-adjacent methods, but not found them fast enough to be useful. Basically, whatever approach is used, it needs to do inference at ~10us latencies to either make real-time result filtering viable, or <1ms not add unreasonable overhead to processing-time result labeling.
Why does it cause tool fragmentation? You can change the font to a normal sans-serif font or to a monospaced font.
Personally I like the default font. It looks weird to have my crappy doodles next to a normal computer font. The default one is very legible but has a style (and ligatures) that make it feel not too neat
That doesn't really make sense. So it's an ad for raycast? But raycast said they didn't know about it. To me the explanation makes perfect sense. "You can use this tool with raycast" seems like a very reasonable tip.
> That doesn't really make sense. So it's an ad for raycast?
It's an ad for using CoPilot and for Raycast.
> But raycast said they didn't know about it.
If I buy a billboard that tells people to go eat at a nearby restaurant, that's ad regardless of whether or not the restaurant knows that I bought that ad.
> To me the explanation makes perfect sense. "You can use this tool with raycast" seems like a very reasonable tip.
Raycast is a paid product. Even though they have a free tier, they only have that to get people to use and like the tool enough to pay for it. They want you to use Raycast so you use CoPilot and pay for it. It's an ad.
Anyone claiming this is just a tip is being disingenuous or is extremely naive. MS knows exactly what they're doing, this wasn't a charity offering. Now they're claiming it was a tip to save face.
Tips are also not acceptable to add to PR text. It’s like the definition of a “weed”. A “tip” in the GitHub UI would make sense. But “tips” injected into my own PR text become unwelcome ads. In any case, what may be helpful “tips” today are only a gateway to straight up paid ads tomorrow. After all, I get told all the time by adtech folks that actually, the ads and all the tracking behind them are good because aren’t I glad the ads are relevant to my interests and that I’m supporting small businesses online whose shops can only exist because of the ad infrastructure. To which I say, no, they aren’t, and that’s a lie.
The situation you describe has dynamics that don't apply when your windows laptop is trying to get you to install an update. A woman can't have 100% confidence that saying no won't trigger a man into rage, so just the question being asked at all is already a bit unpleasant. WinRAR trying to get me to buy a license is not as offensive because I know it won't beat me up for saying no.
However, do you think people accept Microsoft backup because they want a backup?
Or do you think they click yes because it makes the popup go away for good?
Wearing me down until I say yes isn’t the same as just yes.
It’s the same dark pattern for the 10-11 upgrade. My father in law managed to upgrade by accident because it kept popping up. He didn’t really make an informed choice for himself. One day he just couldn’t figure out why everything was different.
If you buy a google pixel 9 (the last version for which google released device trees), you can do anything you want on your phone. My pixel runs a version of android I built myself
These are some multilingual dictionaries^[1] I made. They're designed to be useful to language learners. They have some cool features I haven't seen elsewhere.
1. you can see the frequency of each meaning of each word. For example, the french word "bois" commonly means "drink" and rarely means "wood": https://yap.town/d/french-to-english/bois/
2. Each meaning also comes with a huge number of example sentences, taken from "real" sources (movies mostly)
3. It also has definitions for common phrases and constructions. For example, "c'est" is technically two words in french, but you should just learn it as one unit, so it gets its own entry: https://yap.town/d/french-to-english/c-est/
You can also see a list of phrases for any language of interest: https://yap.town/d/french-to-english/top-1000-phrases/ . The phrases were discovered via the "unigram" tokenization algorithm (originally invented for training LLMs!)
Something remarkable about the phrases is that you can often understand the meaning of a sentence just from the meaning of each phrase in it. The concatenated phrases sometimes even form grammatical english. And of course, if the same phrase has multiple meanings, those are separated and have their own example sentences, just like the words do.
^[1]: And by "I", I mean I wrote code to ask an LLM to generate each dictionary entry.
You can just paste the entire gist into your Claude session to have Claude set it up for you
reply