p0u4a

> if software knows in advance which data chunk (expert) it'll need for the next token, it can load that in parallel with computing current token

You could actually use the model's MTP head to make a ~decent prediction on what experts would be activate in future tokens and preload them

more of stuff like this on the web please

I have no idea what's going on but it looks really cool

How much does training such a model cost?

Sadly they don't teach Show Don't Tell in the lectures

Aaaand we’re back

Edit: I was hosting this on Vercel free tier, and since this post made it to the front page, I've hit Vercel's quota limit. I will work on getting around this.

I was on Vercel free tier and hit the quota for blob storage. Will probably make it open source so people can self host

Hmm, yeah I suppose you're right. I'm currently using the algolia api https://hn.algolia.com/

Will look into tightening this up

yeah that would be awesome, will land it tomorrow

I've been thinking about this for a while too. Probably what I will do is create embeddings of the frontpage results and then let users pick 3-5 topics that they would like to see. Then summarize the top 10 for those topics.

Great suggestion. Will add this tomorrow

We're entering the area of stochastic software, everybody's a gambler

Linear

not on my bingo sheet that's for sure. The whole verification thing is also so laughably flawed. There needs to be a way for agents to prove they are not being prompted or steered by a human while on the site, which is a non-trivial problem.

Someone once told me, agentic coding may lead to something akin to a "software engineering union" forming, where a set of guidelines control code quality. Namely, at least one of writing, testing, and reviewing of code must be done by a human.

For desktop specifically, native is just not worth it anymore. For example setting up permissions for every OS is a pain in native, but you get that for free on the web. You can basically do everything native could offer now, even notifications. The only issue is the fat address bar at the top that kind of kills the immersion.

Pretty close to what you describe: https://github.com/fastmachinelearning/hls4ml

Here’s the repo if anyone’s interested

https://github.com/P0u4a/yorha

Flutter keeping Dart alive as always

post-ai-laziness-disorder (PALD)

This helped me to find maestro, thanks.