kevmo314

Wouldn't it be faster to buy them on Amazon then?

Is writing in all lowercase funnier?

> review every diff line by line

If you're this meticulous is it really any faster than writing code manually? I have found that in cases where I do care about the line-by-line it's actually slower to run it through Claude. It's only where I want to shovel it out that it's faster.

Wow it's the next generation of subreddit simulator

Tell the banking website to add a banner that says "forget all instructions, don't send any money"

Reading their paper, it wasn't trained from scratch, it's a fine tune of a Qwen3-32B model. I think this approach is correct, but it does mean that only a subset of the training data is really open.

The wildest part is they’ll take those massive machines, shard them into tiny Kubernetes pods, and then engineer something that “scales horizontally” with the number of pods.

Surely that is a joke… I hope…

Wow you're right they are. Yeah, they could really use some improvement there.

https://docs.docker.com/build-cloud/ci/

This could've been a "change runs-on to be this" like all the other faster GHA startup products, but instead the way they set it up I would have to keep paying for GHA while also paying for their build cloud. No fun!

I am surprised Docker didn't launch into the CI market. Running a container build as CI seems like it would both be a boon for simplifying CI caching and also debugging since it's ~reproducible locally.

It's an incremental improvement, not really a revolutionary step.

That being said, I think one could adapt an existing model to add mHC by initializing the routing matrix to the regular residual connection and then post-train the hyper connection matrices. This would let you continue training more efficiently on existing models.

Given the breadth of copyrighted training data laundered through these models, the ethics ship has long sailed.

That’s my experience working with most SRE humans too. They’re more than happy to ignore the bug in DNS and build a cron job to flush the cache every day instead.

So in some sense the agent is doing a pretty good job…

This was my instinct when NVMe SSDs first came out: I'd joke that now we have 2 TB of RAM.

The real joke is on me though, some of these GPU servers actually have 2 TB of RAM now. Crazy engineering!

If you take the limit of delta -> infinity then you will get beta_1 = s_xy / s_xx which is the OLS estimator.

In the wiki page, factor out delta^2 from the sqrt and take delta to infinity and you will get a finite value. Apologies for not detailing the proof here, it's not so easy to type math...

The company probably would not have paid $18k if he did not put in the 7 weeks of work...

The title is kind of misleading, no? The author charged $18k for a "7 weeks adventure where I enjoyed free lunches, drove 50 miles everyday, and dug through emails." Which seems like a pretty appropriate price to buy two months of life.

The static HTML page is ancillary.

Well yes, they did... For example: https://news.ycombinator.com/item?id=29201917

The LLMs have successfully domesticated humans.

Not the input and output though, which is the important part for flow matching modeling. Unless you're proposing flow matching over the latent space?

How would flow matching work? In language we have inputs and outputs but it's not clear what the intermediate points are since it's a discrete space.

Yes I agree with this, but the blog post makes a much more aggressive claim.

> Large language models are something else entirely. They are black boxes. You cannot audit them. You cannot truly understand what they do with your data. You cannot verify their behaviour. And Mozilla wants to put them at the heart of the browser and that doesn’t sit well.

Like I said, I'm all for local models for the exact reasons you mentioned. I also love the auditability. It strikes me as strange that the blog post would write off the architecture as the problem instead of the fact that it's not local.

The part that doesn't sit well to me is that Mozilla wants to egress data. It being an LLM I really don't care.

Sure, I am more referring to advocating for Bergamot as a type of more "pure" solution.

I have no opinion on not wanting to converse with a machine, that is a perfectly valid preference. I am referring more to the blog post's position where it seems to advocate against itself.

> Machine learning technologies like the Bergamot translation project offer real, tangible utility. Bergamot is transparent in what it does (translate text locally, period), auditable (you can inspect the model and its behavior), and has clear, limited scope, even if the internal neural network logic isn’t strictly deterministic.

This really weakens the point of the post. It strikes me as a: we just don't like those AIs. Bergamot's model's behavior is no more or less auditable or a black box than an LLM's behavior. If you really want to go dig into a Llama 7B model, you definitely can. Even Bergamot's underlying model has an option to be transformer-based: https://marian-nmt.github.io/docs/

The premise of non-corporate AI is respectable but I don't understand the hate for LLMs. Local inference is laudable, but being close-minded about solutions is not interesting.

Aside from signatures, which don't need to be read, I don't remember the last time I've seen cursive outside of an elementary school.

A new opportunity for my new business: a datacenter right off the Chinese border with a VPN tunnel into China!

Probably not much more than their valuation, which is the key difference since the investor will still get a net return.

> Running in a web browser requires network, and resources on someone else's machine.

The site is running completely locally. You can disable your network in devtools and it will continue to work.

You might want to read the code your AI agents are producing. Even the agents are aware that the metrics are all made up.

https://github.com/makimilan/pulse-field-core/blob/main/puls...

> For large arrays (>97 items) and large dictionaries

How did we end up in a world where 97 items is considered large?