fxgn's recent activity
-
Comment on Google Assistant shutting down on Android and Wear OS in September in ~tech
-
Comment on Anthropic discovered three cases where Claude broke into another system in ~tech
fxgn Link ParentBut if the LLMs get better at being sneaky, how are the AI companies supposed to brag about what a huge security risk their models are?But if the LLMs get better at being sneaky, how are the AI companies supposed to brag about what a huge security risk their models are?
-
This planet is...up to something
28 votes -
Comment on Survey: The temperature in my room now is _____ and it feels _____ in ~talk
fxgn Link ParentI was very concerned before I remembered Fahrenheit existsI was very concerned before I remembered Fahrenheit exists
-
Comment on Unreleased OpenAI model escapes containment and hacks into Hugging Face in ~tech
fxgn LinkI see OpenAI got inspired by Anthropic's fearmongering marketing tactics huhI see OpenAI got inspired by Anthropic's fearmongering marketing tactics huh
-
Comment on DNA tests are uncovering the true prevalence of incest (gifted link) in ~health
-
Comment on DNA tests are uncovering the true prevalence of incest (gifted link) in ~health
fxgn Linkhttps://periscope.corsfix.com/?https://www.theatlantic.com/health/archive/2024/03/dna-tests-incest/677791/ https://archive.is/wA4rO -
Comment on Control the ideas, not the code in ~comp
fxgn Link ParentIt has definitely happened to me in TypeScript when using a type from a third-party library which has a nullable property, but most of my code only had to accept the object if that property is not...It has definitely happened to me in TypeScript when using a type from a third-party library which has a nullable property, but most of my code only had to accept the object if that property is not null
-
Comment on The coming loop in ~comp
fxgn Link ParentSorry, that was a typo, I meant to say "developers" in plural! Mario and Armin both have great and grounded takes on LLM coding.Sorry, that was a typo, I meant to say "developers" in plural! Mario and Armin both have great and grounded takes on LLM coding.
-
Comment on The coming loop in ~comp
fxgn (edited )LinkThe developers of Pi keep surprising me with very reasonable takes. I already commented about this one in this thread, so I'll copy my comment: There is one type of cancerous coding pattern all...The developers of Pi keep surprising me with very reasonable takes. I already commented about this one in this thread, so I'll copy my comment:
There is one type of cancerous coding pattern all models make, no matter how good the model is, and no matter how detailed the AGENTS.md. I found it's best described here:
Present-day models tend to produce code that is too defensive, too complex, too local in its reasoning. They avoid strong invariants. They add fallbacks instead of making bad states impossible.
[...]
Furthermore it’s well understood that models tend to observe some local failure and add a local defense. Karpathy mentioned how they are “mortally terrified of exceptions”. In systems with important invariants, especially persisted data formats or core infrastructure, the right fix is not “handle every malformed case.” The right fix is to make the malformed case unrepresentable or impossible to write in the first place. Yet even with a lot of manual steering, that type of code does not come out of LLMs naturally, and even if the code comes out naturally like that, they will still attempt to handle now impossible errors.The models would rather add a
if (property !== null)check all over the code than just add that to the type definition and handle the null case in the single place it could possibly arise.Until this is somehow improved, there is no way I'd ever prompt AI without fixing the generated code. Sometimes the model does a bad job and I have to refactor most of it myself, sometimes it does a good job and I dont. In any case, I don't think I've ever had a situation where I didn't have to first remove a bunch of useless defensive checks from the generated code.
It seems like the "best" models are even worse in this regard. I've tried GPT-5.5, but I keep going back to DeepSeek v4, because GPT keeps insisting on adding that stuff all over even when I remove it.
-
The coming loop
14 votes -
Comment on Control the ideas, not the code in ~comp
fxgn (edited )LinkThere is one type of cancerous coding pattern all models make, no matter how good the model is, and no matter how detailed the AGENTS.md. I found it's best described here: The models would rather...There is one type of cancerous coding pattern all models make, no matter how good the model is, and no matter how detailed the AGENTS.md. I found it's best described here:
Present-day models tend to produce code that is too defensive, too complex, too local in its reasoning. They avoid strong invariants. They add fallbacks instead of making bad states impossible.
[...]
Furthermore it’s well understood that models tend to observe some local failure and add a local defense. Karpathy mentioned how they are “mortally terrified of exceptions”. In systems with important invariants, especially persisted data formats or core infrastructure, the right fix is not “handle every malformed case.” The right fix is to make the malformed case unrepresentable or impossible to write in the first place. Yet even with a lot of manual steering, that type of code does not come out of LLMs naturally, and even if the code comes out naturally like that, they will still attempt to handle now impossible errors.The models would rather add a
if (property !== null)check all over the code than just add that to the type definition and handle the null case in the single place it could possibly arise.Until this is somehow improved, there is no way I'd ever prompt AI without fixing the generated code. Sometimes the model does a bad job and I have to refactor most of it myself, sometimes it does a good job and I dont. In any case, I don't think I've ever had a situation where I didn't have to first remove a bunch of useless defensive checks from the generated code.
It seems like the "best" models are even worse in this regard. I've tried GPT-5.5, but I keep going back to DeepSeek v4, because GPT keeps insisting on adding that stuff all over even when I remove it.
-
Comment on Elixir Is All You Need in ~comp
fxgn Link ParentYou're making a browser extension in Elixir? How?You're making a browser extension in Elixir? How?
-
Comment on Elixir Is All You Need in ~comp
fxgn Link ParentElixir has gradual type checking now https://elixir-lang.org/blog/2026/06/03/elixir-v1-20-0-released/ GenServer and Task are both features built into Elixir, not separate libraries. Oban is a...i'm hesitant at anything dynamically typed though (seems they don't do implicit conversions but you're still going to get runtime errors rather than compile time if i'm reading right?)
Elixir has gradual type checking now
https://elixir-lang.org/blog/2026/06/03/elixir-v1-20-0-released/
the equivalent of redis is now Oban, GenServer, and Task
GenServer and Task are both features built into Elixir, not separate libraries. Oban is a library and it depends on SQL, so it's not really just "Elixir+Phoenix", but I guess the chart doesn't claim you can also replace a SQL database with Elixir.
-
Elixir Is All You Need
15 votes -
Comment on European Parliament achieves upgrade to air passenger rights in ~transport
fxgn Link ParentBest case, the list of your installed applications, all accelerometer movements of your phone, every time you unlock and lock your phone, google advertisement profile id, and a lot of other stuff...What sensitive data are you protecting that Android permissions aren't capable of handling and their website is incapable of harvesting?
Best case, the list of your installed applications, all accelerometer movements of your phone, every time you unlock and lock your phone, google advertisement profile id, and a lot of other stuff useful for fingerprinting and advertising.
Worst case, something like this:
Not saying the Ryanair app does all of that, just answering your specific question. And I've personally seen random marketplace apps notify me (not because Android enforces that, but probably just because they were careful about privacy regulations) after an update that they will now gather the list of all my installed apps to "protect me from viruses".
-
Comment on Give me your culture clash stories in ~travel
fxgn Link ParentReminds me of the Not Just Bikes video about how trying to take a walk in Houston radicalized him so much he started a YouTube channel about urban planning https://www.youtube.com/watch?v=uxykI30fS54Reminds me of the Not Just Bikes video about how trying to take a walk in Houston radicalized him so much he started a YouTube channel about urban planning
-
Comment on Modern, abstract art makes me angry in ~arts
fxgn Link ParentYou could say the same thing about colors and shapes though. They all can have implicit associations and evoke specific things.You could say the same thing about colors and shapes though. They all can have implicit associations and evoke specific things.
-
Comment on Google must pay record €4.1 billion fine, top EU court rules in ~tech
fxgn Link ParentThey do https://source.android.com/docs/setup/contribute/submit-patchesGoogle does not accept outside contributions
They do
https://source.android.com/docs/setup/contribute/submit-patches
-
Comment on Giving Green in ~enviro
fxgn (edited )Link ParentI wouldn't really call it mostly technosolutionist. Most of the things they fund is more focused on government/corporate lobbying (eg. https://catf.us). They have a detailed report on why they're...I wouldn't really call it mostly technosolutionist. Most of the things they fund is more focused on government/corporate lobbying (eg. https://catf.us).
For exemple, we see funding for decarbonizing air travel (it's a very long shot, for something hardly vital, i wouldn't call this a data driven good target for our limited dollars)
They have a detailed report on why they're funding aviation emission reduction: https://www.givinggreen.earth/research/reducing-aviation-emissions-strategy-report
From this it does seem to be a pretty cost effective way to prevent a large amount of emissions
develloping tastier "alternative proteins" so plant based meat substitute (we allready have tasty and balanced vegetarian diets, it's called indian cuisine)
I understand anecdotal experience isn't great evidence, but plant-based meat has been huge for reducing my own meat consumption. I'd probably never go vegetarian otherwise, but having access to tasty plant based meat since recently does make me seriously consider that as an option. I don't think I'm the only one like this, and the surveys seem to support that. There's plenty of people who eat meat because they like the taste, and having plant-based options that taste as good would let them reduce their meat consumption.
Btw, Home Assistant has their own official smart speaker device for their voice assistant
https://www.home-assistant.io/voice-pe/