Hacker News new | past | comments | ask | show | jobs | submit | 8n4vidtmkvmk's comments login

My work has AI code reviews. They're like 0 for 10 so far. Wasting my time to read them. They point out plausible errors but the code is nuanced in ways an llm can't understand.

Isn't the selling point behind Blue sky is that you can customize your feed your way? I don't know the tech behind that but the feed is "open" isn't it? Can they plug into that?

The selling point of Bluesky is that you don't get bombarded with a mob of blue checks every time you post something political

I thought the phrasing of "western distributor" was shady. I'm like doesn't that just mean the person ordering and distrusting the parts is "in the west" (not even necessarily US) but the parts could be from anywhere? It's not saying much at all.


Maybe he just means they're literally countable. Like, accurate to a single digit?


The half blue / half green image still looks better with LQIP than BlurHash. I was getting ready to use BlurHash in my app, might try this instead!

In fact, LQIP looks better than most of the BlurHash examples in the gallery (https://leanrada.com/notes/css-only-lqip/gallery/); not sure if these were cherry picked or what.


Author here: Definitely cherry picked ;)

I did deliberately pick some "bad" examples like the blue+green image, and other multicolor images.

I wanted to add an upload function so people could test any image, then i realised I'd have to implement the compression/hashing in the client. Maybe i should!


I tried getting that working earlier using Claude to convert your script - you can see the result here: https://claude.site/artifacts/b747d94a-2923-4904-8ed1-7330bf...

Here's the transcript and code: https://claude.ai/share/4a562082-b681-4f0c-909c-3c32c34fd050


I could tell and I really appreciate it. It's really helpful to see both the good and the bad.

Great work!


They've screwed this up several times by shipping too soon and then having to backpedal. The web is not something you want to screw up because they don't own every website in existence.


I figured you don't update the major unless you significantly change the... algorithm, for lack of a better word. At least I assume something major changed between how they trained ChatGPT 3 vs GPT 4, other than amount of data. But maybe I'm wrong.


The number is purely for marketing.

If you could get much better performance without changing the algorithm (eg just by scaling), you'd still bump the number.


I'm not sure how it'll ever make sense unless you need a lot of customizations or care a lot about data leaks.

For small guys and everyone else.. it'll probably be cost neutral to keep paying OpenAi, Google etc directly rather than paying some cloud provider to host an at best on-par model at equivalent prices.


> unless you need a lot of customizations or care a lot about data leaks

And both those needs are very normal. "Customization" in this case can just be "specializing the LLM on local material for specialized responses".


Why can't it be E2E encrypted? If you can encode data on sound waves, surely you can encode encrypted data on sound waves.


Seems to me like the overhead it would add would way outweigh the benefit for the use case. I wouldn't want to wait an extra 10 seconds for the audio to finish transmitting every time just to prevent nonexistent spies in my kitchen from listening in.


The data/time overhead would be negligible. Engineering overhead to implement it on the other hand, perhaps not.


> The data/time overhead would be negligible.

Doubling or tripling the amount of data sent would be negligible over a wire, but an audio protocol won't be as snappy. Then there's the matter of trust/decryption. How are those keys being kept safe? What happens if I lose access?


You can encrypt data and keep the exact same original byte size.


If you're using symmetric encryption, sure. And then every dishwasher has the same symmetric key, guaranteeing it'll leak.


Everything uses symmetric encryption, including asymmetric encryption. Using symmetric encryption doesn't mean you'd use the same key for every dishwasher, you'd obviously just pair the devices and generate a new key, like everything does nowadays. Also, my dishwasher doesn't use any sort of encryption, and it still leaks.


Then don't double or triple the amount of data sent. There's simply no need.


No, we're past pnpm and on to bun now


Bun may be fine for people using it to replace the javascript runtime

As a dependency management toolchain I recommend reading: https://dev.to/thejaredwilcurt/bun-hype-how-we-learned-nothi...

I had some issues with it using it briefly (though it was way faster than other installers I've used)


Sorry, everyone’s on pyarn now and next month we’re deprecating that for pbun and then starting from scratch with taquito when that doesn’t solve all our perceived problems


Still using gnu make.


I'll concede that I've moved to bazelisk over make


Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: