Hacker Newsnew | past | comments | ask | show | jobs | submit | AlexCoventry's commentslogin

The fundamental ideas in the paper aren't particularly novel. They will probably work as advertised.

This model has fast weights, which actually are modified during inference.

Marketplace for fast weights inbound

It was definitely a thing.

When I was choosing between learning python and perl in the late 90's, it was the context sensitivity of perl expressions which really squicked me. To me, that was the critically bad language decision in perl. You can make context-sensitive expressions in python (using operator overloading, for instance), but you have to go out of your way to do it in one way or another. In perl you can easily do it by accident, and it can result in serious bugs. It seemed masochistic, to me.

> So it seems to me that we are likely to have worse software over time.

IMO, it's already happening. I had to change some personal information on a bunch of online services recently, and two out of seven of them were down. One of them is still down, a week later. This is the website of a major utilities company. When I call them, they acknowledge that it's down, but say my timing is just bad. That combined with all the recent outages has left me with the impression that software has been getting (even more) unreliable, recently.


Maybe it's not as necessary with a codebase as well-organized as Oxide's, but I found gemini 3 useful for a refactor of some completely test-free ML research code, recently. I got it to generate a test case which would exercise all the code subject to refactoring, got it to do the refactoring and verify that it leads to exactly the same state, then finally got it to randomize the test inputs and keep repeating the comparison.

> I think that the code you actually want to ship is so far from what LLMs write

It depends on the LLM, I think. A lot of people have a bad impression of them as a result of using cheap or outdated LLMs.


I wonder if they would be willing to publish the "LLMs at Oxide" advice, linked in the OP [1], but currently publicly inaccessible.

[1] https://github.com/oxidecomputer/meta/tree/master/engineerin...


Disclaimer: Oxide employee here.

To be honest there's really no secret sauce in there. It's primarily how to get started with agents, when to abandon your context and start anew, and advice on models, monitoring cost, and prompting. This is not to diminish the value of the information as it's good information written by great colleagues. I just wanted to note that most of the information can be obtained from the official AI provider documentation and blog posts from AI boosters like Thorsten Ball.


Thanks.

You're welcome. My colleague published the text for it: https://gist.github.com/david-crespo/5c5eaf36a2d20be8a3013ba...

Cool, thanks again to both of you. :-)

This is from May 2025, according to the arxiv watermark. Maybe that should be mentioned in the title.

> There are three blood thinners on the US market and they all cost a lot.

What about Warfarin? Its $21 for 30 pills, according to https://www.goodrx.com/warfarin


Warfarin needs blood tests at least every 2 or 3 weeks. I bet those tests are not cheap in the USA. Furthermore you can't have any food with significant amounts of vitamine K or its precursors. There are pills that cost more but don't require any of that. And actually they cost very little or zero, at least in my European country. Prescription only, of course.

>I bet those tests are not cheap in the USA.

Within 5 minutes of searching I found a PT time with INR test for $24 [0]. Add in the testing fee and it's probably around $30.

[0] https://www.ultalabtests.com/test/prothrombin-time-with-inr-...


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

Search: