Hacker News Reader

lemmingon Gemini 2.5 Flash

I maintain an alternative client which I build from the API definitions at https://github.com/googleapis/googleapis, which according to https://github.com/googleapis/python-genai/issues/345 should be the right place. But neither the AI Studio nor the Vertex definitions even have ThinkingConfig yet - very frustrating. In general it's amazing how much API munging is required to get a working client from the public API definitions.

lemmingon The ADHD body double: A unique tool for getting things done

An online application of this principle is Focusmate, where you schedule calls with essentially random strangers and just work together. There's very little chit chat and it's not some weird front for a hookup scheme. I've been amazed at how well it works for me. I can't really explain why, after all the other person can't see my screen so I could just be doom scrolling YouTube the whole time, but for some reason I don't. I don't use it all the time, but when I need to get something unappealing done I still use it. I've also used it for exercise, and I've had partners use it for all sorts of weird things that they had been procrastinating. I highly recommend it, I thought it would be far weirder than it turned out to be, and it's really useful.

lemmingon Claude can now search the web

The chat continues with the results, and I often explicitly tell it "search to make sure your answer is correct" if I see it making stuff up without actually searching. I use it multiple times a day for all sorts of things.

lemmingon Blender-made movie Flow takes Oscar

I’m very pleased to see this, Flow was such a lovely film. I didn’t realise it was made with blender!

lemmingon Broken legs and ankles heal better if you walk on them within weeks

We draw the lines in different places.

e.g. amateur MMA is illegal in Norway and I think a couple of other places too.

lemmingon A decade later, a decade lost (2024)

I remember reading Eric's writing about Rebecca's illness and death at the time, and even though it was before my own daughter's birth I remember it being one of the saddest things I could remember reading. His writing is so eloquent that he really communicated the sadness and helplessness, while still managing to move on with his life. My deepest condolences.

lemmingon I built an open source AI tool to find my autoimmune disease

My concern is that for every person with a rare disease causing vague symptoms, there are more people with similar vague symptoms that don't have an underlying rare disease.

This is a problem, for sure, and additionally the people with vague (perhaps even imagined) symptoms probably vastly outnumber the ones with legit but weird conditions. But for the people with those conditions who can't find a doctor who will take them seriously, what are they supposed to do? I know a few of them, and I'm unwilling to basically throw them under the bus for the greater good. IMO (going back to OP), if LLMs give them a few options that they can go back to a doctor and ask "Could it be this? How can we rule that out?" could be really life-changing.

lemmingon I built an open source AI tool to find my autoimmune disease

And they all got diagnosed with no problem? If that's the case, then they were lucky in my experience. Obviously not lucky to have the conditions, which are objectively terrible, but having a condition like this and fighting to have it recognised and treated is a truly awful experience.

I guess looking at your list they're mostly the better-known autoimmune conditions (except for Lupus and perhaps the lung one), and diagnosis might have been easier.

lemmingon I built an open source AI tool to find my autoimmune disease

But when you have a list of vague symptoms that aren't very serious, could be caused by anything, and your records show 30 visits to hospitals and specialists, and you start talking about experimental life extension treatments, the vast majority of doctors are going to write you off almost immediately as a hypochondriac.

Tell me you don't know anyone with an autoimmune disease, without telling me you don't know anyone with an autoimmune disease!

This is exactly the experience of a lot of people, my wife included. It took years to finally get a diagnosis, partly because it took her a long time to convince herself that it wasn't all in her head (much of which was due to gaslighting and general incompetence of doctors), and then once she was convinced she had a problem, finding a doctor who would actually take her seriously was a total mission. There is an enormous class of poorly-understood conditions which result in weird vague symptoms that come and go, and could be explained by many different factors. These symptoms can really destroy your quality of life, but it can be almost impossible to get them diagnosed, and there are often poor treatment options even if you do.

I really try not to take my health for granted now. We are well off financially, but it doesn't matter how much money we have, there's nothing anyone can do about the sort of condition my wife has, at least currently.

lemmingon Building Personal Software with Claude

This seems to be a problem with the most recent model (claude-3-5-sonnet-20241022). There was a lot of complaining about it on their discord when it was released and I had a lot of problems with it too. If you use the previous Claude model, it still works very well and doesn't have these problems.

lemmingon Spaced repetition can allow for infinite recall (2022)

I haven't worked through it, but Michael Nielsen has a great article about this: https://cognitivemedium.com/srs-mathematics.

lemmingon OpenAI O3-Mini

Where does DeepSeek say that about API usage? Their privacy policy says they store all data on servers in China, and their terms of use says that they can use any user data to improve their services. I can’t see anything where they say that they don’t train on API data.

lemmingon OpenAI O3-Mini

So for posterity, in this subthread we found that I can use OpenAI without them training on my data, whereas I cannot with DeepSeek.

lemmingon OpenAI O3-Mini

By default, we do not train on any inputs or outputs from our products for business users, including ChatGPT Team, ChatGPT Enterprise, and the API. We offer API customers a way to opt-in to share data with us, such as by providing feedback in the Playground, which we then use to improve our models. Unless they explicitly opt-in, organizations are opted out of data-sharing by default.

The business bit is confusing, I guess they see the API as a business product, but they do not train on API data.

lemmingon Parinfer: Simpler Lisp Editing

Thanks for the kind words, Simon! I think you probably just got used to the quirks :-). The worst ones are around pasting, which is mostly IntelliJ complicating things for parinfer by making a lot of whitespace changes in order to format the code, in addition to pasting it. But there's also all these: https://github.com/cursive-ide/cursive/issues/2945. I'm planning to do a round of parinfer fixes soon though. It's so complicated that I can't hold it in my head for long, so I have to load it into my brain cache and then do a batch of fixes at a time.

lemmingon Parinfer: Simpler Lisp Editing

I develop Cursive, an IDE for Clojure code, and I spent a long time integrating parinfer smart mode. It was actually really difficult to do well, I wrote a document with feedback about what (among other things I discovered later) that made it hard: https://github.com/parinfer/parinfer.js/wiki/Problems-integr.... I subsequently fixed many of those issues, but it was probably the second hardest thing I've integrated into Cursive, after Leiningen.

It's still a bit quirky, but I use it every day and I love it. Even though I'm proficient enough with paredit, parinfer is just so much easier to use. Several users have commented to me that the parinfer integration is the main thing stopping them switching away to a different editor - it really makes that much difference.

lemmingon Is the TikTok ban a chance to rethink the whole internet?

Of course it's a chance to rethink the internet. COVID was also a chance to rethink our responses to pandemics. Cambridge Analytica was a chance to rethink data privacy as it relates to private companies. Edward Snowden was a chance to rethink data privacy as it relates to governments. Climate Change is a chance to rethink how we live our lives, build our cities, and produce our food. In my hometown, an earthquake was a chance to rethink how the city worked.

Life is full of these chances, but we seldom take them. Whatever happens with TikTok, it'll be business as usual in no time.

lemmingon Swc4j: SWC for Java

Thanks for the reply, I think your reasons make total sense for your use case and I was also underwhelmed by Antlr, or at least the TS grammar for it. But I don't need the vast majority of what SWC offers, I just need a type declaration parser - that seems like it should be much more tractable to keep up to date, especially since there's a massive test suite in DefinitelyTyped.

lemmingon Swc4j: SWC for Java

I agree that the Antlr Typescript grammar is underwhelming, and I also agree with the difficulty of not having a well defined owner of the project to interact with. My main use case is parsing TS type definitions, and the grammar from the grammars-v4 repo doesn't parse anything starting with 'declare', which is pretty fundamental.

lemmingon Swc4j: SWC for Java

Right, that would be another native bridge solution.

lemmingon Swc4j: SWC for Java

I want to use Typescript types to improve IDE support for a compile-to-JS language (ClojureScript).

lemmingon Swc4j: SWC for Java

Oh my word, there may finally be a Typescript parser for the JVM? I have wanted this for the longest time, and it's been on my radar to take on as a project in the near future. If this is a reasonably complete parser, this will save me a ton of time. My use case is parsing Typescript type definitions.

Edit: unfortunately, this seems to be a JNI bridge to the Rust code in SWC, I was hoping for a JVM-native solution.

lemmingon BlogScroll – An open directory of personal sites and blogs

Kagi also has an occasionally delightful small web index: https://blog.kagi.com/small-web. It’s nicely integrated into their search results too.

lemmingon Programmers want flow. When programming, light turns red

That seems like a stretch to me from the landing page, and is absolutely not my experience from using it. It’s also explicitly discouraged in all their documentation, e.g. this is from their FAQ:

Can I chat with my partner?

No. Talking with your partner is only allowed at the beginning and end of your session. At the beginning, you’re encouraged to say hello to your partner and share what you plan to work on. At the end, you should check in and ask your partner how the session went. If you’d like to talk to your partner more, you can suggest or offer it at the end of your session. Please bear in mind, no selling or propositions (business or otherwise) are allowed during Focusmate sessions.

lemmingon Programmers want flow. When programming, light turns red

Yes that one, and yes it’s a real human. I don’t think it would work as well with an LLM, something about there being a real person there helps immensely with accountability. I’d encourage you to try it out if the idea is interesting, it’s nowhere near as weird as it sounds and it really does work for keeping you focused.

lemmingon Programmers want flow. When programming, light turns red

I work from home. For a while, I was using Focusmate to keep me on track, which worked very well. I implemented something similar using a standard smart bulb connected to Apple Home, and then I just controlled it using shortcuts and Hammerspoon such that it would turn red when my webcam was active, and turned off when it wasn't. I mounted it in a normal lamp outside my office door. It worked really well to reduce family interruptions when I was trying to work. I haven't been doing this for a while now, but I'll probably come back to it at some point.

I called it my Roxanne light :-)

lemmingon The Most Expensive Eating Disorder

I saw Bryan recently on a Magnus Midtbo episode on YouTube. In that they discussed that his body fat had been extremely low, and he had gone up to around 10% subsequently on medical advice, i.e. that his body fat had previously reduced to unhealthy levels. 10% is low, but it's not what he looks like in the photos on this article, and it should be sustainable long term with some effort. He says on his website under his current protocol that he eats 2,250 calories a day (10% deficit from RDA), which doesn't sound crazy to me. I mean, it's not what normal people do but he's not doing what normal people do by definition, and at least what he's eating doesn't seem out into crazy unhealthy territory to me.

He definitely came across as a little eccentric in the video, but very enthusiastic about the project, I found it pretty charming and would definitely find it pretty interesting to chat to him I think.

lemmingon Ask HN: What are you working on? (October 2024)

Oh, if you think that is bad, wait until you're looking for educational apps for them. What a total hot mess in a sector which should by rights be amazing.

Fortunately this is one of the killer apps of LLMs IMO - I have built multiple small tools to help educate our daughter, like this guy: https://x.com/random_walker/status/1848388462782673340

lemmingon Ask HN: What are you working on? (October 2024)

For anyone wanting interesting YT videos for their kids (and not wanting to take anything away from OP's project), I highly highly recommend thekidshouldseethis.com. It's basically a curated stream of cool videos, and I would feel totally safe letting my daughter browse it alone (she never does because we usually watch them together, but the curation is that good). Videos on all sorts of topics, and good enough to be really entertaining for both kids and adults - I can spend an evening there easily. They also have a really fantastic gift guide.

lemmingon Everything I built with Claude Artifacts this week

This is interesting, I also tried this with my daughter after we had been talking about Towers of Hanoi, and like you it worked really well. Then we tried to get it to implement a little version of the game where you have a farmer, some corn, a chicken and a wolf and you have to get them across the river in only one boat (actually Wiki says goat and cabbage, but whatever https://en.wikipedia.org/wiki/Wolf,_goat_and_cabbage_problem). I wasn't trying to get it to solve the puzzle, just give us a UI to play the game. We gave up after an hour or more of going in circles. I wonder if there's a lot of Towers of Hanoi implementations out there it can use as references.