Dan Witte

@patrickc @ATabarrok What vendor do you trust for WGS?

I'm lucky enough to have a great doctor and access to excellent Bay Area medical care. I've taken lots of standard screening tests over the years and have tried lots of "health tech" devices and tools. With all this said, by far the most useful preventative medical advice that I've ever received has come from unleashing coding agents on my genome, having them investigate my specific mutations, and having them recommend specific follow-on tests and treatments. Population averages are population averages, but we ourselves are not averages. For example, it turns out that I probably have a 30x(!) higher-than-average predisposition to melanoma. Fortunately, there are both specific supplements that help counteract the particular mutations I have, and of course I can significantly dial up my screening frequency. So, this is very useful to know. I don't know exactly how much the analysis cost, but probably less than $100. Sequencing my genome cost a few hundred dollars. (One often sees papers and articles claiming that models aren't very good at medical reasoning. These analyses are usually based on employing several-year-old models, which is a kind of ludicrous malpractice. It is true that you still have to carefully monitor the agents' reasoning, and they do on occasion jump to conclusions or skip steps, requiring some nudging and re-steering. But, overall, they are almost literally infinitely better for this kind of work than what one can otherwise obtain today.) There are still lots of questions about how this will diffuse and get adopted, but it seems very clear that medical practice is about to improve enormously. Exciting times!

@sama request for hotkeys to switch between models

@karpathy do you think this is a tokenization issue? (it omits the "h" or uses 4 "o"s)

The presentation style of LLMs will make illegibility an appealing aesthetic to the next indie generation

@natolambert should we expect openai to come out with a healthcare tuned model like 4.1 was for coding? The new eval + a recent Michelle Pokrass interview suggests they're looking for a new power user use cases with in verifiable domains

@jxmnop Did you find any particularly effective ways of using cursor? I’ve found that some useful features are hidden

two weeks trying cursor agent and my opinion has changed a lot. software engineering is undergoing a fundamental shift great engineering used to require deep human-machine symbiosis it's now optimally managing of a team of eternally optimistic but hopelessly deficient interns

@natolambert is there a good benchmark comparison of structured outputs for O3 and Gpt4O? Google is failing me.

@sama deep research needs an export to pdf option. Ideally after you've edited the report with canvas (canvas also needs an export pdf function)

@ryanchenkie @simonw How is this not a more widely used exploit? misspelled packages would also be an easy method

⚠️ Developers, please be careful when installing Homebrew. Google is serving sponsored links to a Homebrew site clone that has a cURL command to malware. The URL for this site is one letter different than the official site.

@nearcyan kids these days just want to be wrap stars

remember when the iphone was new and there were all those "app guys" and they just "made an app" and somehow this pulled millions of dollars in with no effor well guess what the days are back and all it takes is a good wrapper. claude and i call it "The Art of the Wrap"

@paulg But getting an outlet in the middle of the room...priceless

None of the other things are that expensive. The shelves are painted pine. The desk is a kitchen table I bought in the 90s. A used Aeron costs £350. The monitor is so old it's worthless. The books are all used and cost maybe £10k total.

My flight was delayed so I cleaned up my office.

@alexalbert__ Have you come across any good YouTube videos of people demo-ing how their LLM coding setups?

We will soon find it crazy that developers would sit down and hand-type lines of structured code all day. Each day it makes more and more sense for machines to handle that part for us.

Getting developer adoption for new programming languages is going to be difficult in the LLM epoch. Old languages have enough documentation, examples, and forum posting that only the die hards will use something new.

ok chatgpt came up with something actually funny

just dropped a sesame seed in my computer. it's gonna smelll real good when the GPUs go brrr

@fabknowledge @asianometry @dylan522p Yeah that's fair. FWIW I would legit volunteer a little time to help clean up the edges if it would improve quality

@wittedanj @asianometry @dylan522p i totally understand, I wnat to level it up but its hard when its everyone's 5th priority,

Another episode of Transistor Radio. We continue to worsen the show's audio and content quality to make it more fun for all of you guys. We talk about Intel, nuclear reactors and China's impending semiconductor takeover. @fabknowledge @dylan522p podcasters.spotify.com/pod/show/doug-…

@fabknowledge @asianometry @dylan522p Lol. Just sayin, it would be a good thing to have quality podcasts that nerd this hard.

@wittedanj @asianometry @dylan522p 1) never will happen 2) we do it over squadcast but there's no way 1 will ever happen

@simonw I think this is a very good thing that is dependent on base models getting good enough that companies can focus on differentiating by use case

It's interesting how OpenAI are doubling down on end-user-facing features like voice mode, while Anthropic are doubling down on engineer/API-focused features like code generation quality and being able to remotely drive a GUI

@teknium do you know what the best ~7b open source model for structured outputs is currently?