HNHacker News
TopNewBestAskShowJobs

xiphias2

8,476 karma · joined September 7, 2018

submissionscomments
xiphias2··on Video filmed by ICE agent who shot Minneapolis woman emerges
I'm not sure why you think my account is overtaken, but it's all real (I haven't used any LLMs and I'm not a native English speaker).

U.S. politics is not something people from other countries like me can just ignore as it affects our lives as well (what NATO does just as an example).

At the same time I'm not living there and some things are just hard to understand / imagine, and I believe I'm not alone in this (eventhough I love visiting USA).

Of course I have my political opinions but I am respectful and actually like to discuss life experiences my friends who were ,,forced'' on the other side of the political spectrum.

In my experience there is usually one or two emotionally/financially charged case that makes people choose a side and then get in that viewpoint (I'm no different).

What's not productive (and what I have unfortunately felt here in political threads) is just simply downvoting eachother.

xiphias2··on Video filmed by ICE agent who shot Minneapolis woman emerges
The part (as a person of another country) that I don't understand is why people even here on HN is so much against people whose job is to arrest people who broke federal laws.

As I see those laws are not really knew, they were just not enforced during the previous administration as strictly, but people who broke those laws did it consciously.

Do people really care about the people who broke the laws or just hate the current administration so much?

xiphias2··on “Erdos problem #728 was solved more or less autonomously by AI”
First congrats!

Sometimes when I'm using new LLMs I'm not sure if it’s a step forward or just benchmark hacking, but formalized math results always show that the progress is real and huge.

When do you think Harmonic will reach formalizing most (even hard) human written math?

I saw an interview with Christian Szegedy (your competitor I guess) that he believes it will be this year.

xiphias2··on Grok turns off image generator for most after outcry over sexualised AI imagery
Noone can, but it's much easier to verify / prosecute people using credit cards (especially as credit card companies take it very seriously as well)
xiphias2··on AWS raises GPU prices 15% on a Saturday, hopes you weren't paying attention
It mainly depends on how much NVIDIA is overselling the improvements.

With adding RL functions, separating prefill and decode chips, nvfp4 and lots of other architectural changes efficiency of the most valuable tasks goes up as long as the algorithms don't change significantly.

Everything else can just stay on older chips.

xiphias2··on Trump says Venezuela’s Maduro captured after strikes
As long as people all over the world are using ChatGPT and GMail they have all the intel needed to control the world, just like they won wars by all telegrams going through them in the 1800s.

China is their only competitor, but so far people clearly prefer to chat with AI companies from USA.

xiphias2··on US attack on Venezuela raises fears of future Greenland takeover
I still don't understand why some people wanted to capture Putin so much and now against Maduro capture when clearly Maduro has done something uncomparably worse to the people in his country.

Greenland is of course a very different deal compared to Venezuela, it will be interesting to see the political push to the EU.

xiphias2··on Five Years of Tinygrad
Docstrings and comments are amazing to be able to work with huge code bases with big teams where you don't want to understand the whole code base, because it's impossible.

If the goal is to not abstract anything away, and always look at the implementation itself to be able to change it / rewrite it, keep it tiny and it is able to beat PyTorch+CUDA+LLVM with 1000x smaller codebase, who am I to judge?

xiphias2··on Five Years of Tinygrad
I think it's an amazing experiment.

You can look at the PyTorch code base and understand a small local function instantly, but if I would have a task to have a deep understanding of either PyTorch with all kernel code and low level code, whole CUDA code base + LLVM compilation code or Tinygrad I would pick Tinygrad in an instant.

The code looks hard because what it is doing is hard, but all its layers can be debugged.

xiphias2··on ICE's interest in high-tech gear raises new questions: 'What is it for?'
As a foreigner travelling inside USA: people are super nice here in real life, and I see much less politics in real life than on Hacker News.

Be sure to create your ESTA a few workdays before though, as long as you have it you will have a great time.

xiphias2··on US bars 5 Europeans it says pressured tech firms to censor American viewpoints
There is, I just believe that generally people on the left have different viewpoint on this article than people on the right, and that's the main deciding factor.

And whenever that's true, the comments get politically heated without any substantive debate between the sides.

I love to debate with friends on the opposite side than me, but strictly offline where I can learn much more from them.

xiphias2··on US bars 5 Europeans it says pressured tech firms to censor American viewpoints
Another interesting news item that has nothing to do with hacking and a great target for being flagged for being controversial.
xiphias2··on Test, don't just verify
This article is underselling of how much was achieved in proof formalizing for math in the last few years and how close it is to being solved.

If we disregard programming and just look at formalizing math (Christian Szegedy has been doing it for a long time now), the length of proofs that are being formalized are exponentially growing and there's a good chance that in 2026 close to 100% of human written big/important proofs will be translated to and verified by Lean.

Just as an example for programming / modelling cache lines and cycle counts: we have quite good models for lots of architectures (even quite good reverse engineered model for NVIDIA GPUs in some papers). The problem is that calculating exact numbers for cache reads / writes is boring with lots of constants in them, and whenever we change the model a little bit the calculations have to be remade.

It's a lot of boring constraints to solve, and the main bottleneck for me when I was trying to do it by hand was that I couldn't just trust the output of LLMs.

xiphias2··on I misused LLMs to diagnose myself and ended up bedridden for a week
Another poorly written article that doesn't even specify the LLM being used.

Both ChatGPT o3 and 5.1 Pro models helped me a lot diagnosing illnesses with the right queries. I am using lots of queries with different context / context length for medical queries as they are very serious.

Also they have better answer if I am using medical language as they retrieve answers from higher quality articles.

I still went to doctors and got more information from them.

Also I do blood tests and MRI before going to doctors and the great doctors actually like that I go there prepared but still open to their diagnosis.

xiphias2··on How to Run Profitable Pricing Experiments?
For doubling success I suggest ruining the product slowly and testing how crappy the product can still be while the customer is still paying for the brand (as he's expecting some improvement for the higher price), and voila, we've got enshittification.
xiphias2··on Hardening the C++ Standard Library at scale
It's not just simply security / quality, but speed of iteration matters as well: the earlier bugs are found in optimized nodebug mode as well, the more I like to use C++ itself for developing.
xiphias2··on Hardening the C++ Standard Library at scale
It's great that finally bounds checking happened in C++ by (mostly) default.

The only thing that's less great is that this got so much less upvotes than all the Safe-C++ languages that never really had the chance to get into production in old code.

xiphias2··on Leak confirms OpenAI is preparing ads on ChatGPT for public roll out
I'm also on Pro and I know that I won't stop it even if ads subtly change the results. I expect all big LLMs to do the switch exactly at the same time, but later than the free versions.
xiphias2··on Leak confirms OpenAI is preparing ads on ChatGPT for public roll out
They don't need to make us view ads anymore...just say

,,Allow Vercel to use credit card stored by OpenAI''...click to continue refactoring

xiphias2··on OpenAI needs to raise at least $207B by 2030
I'm traveling like this all the time already, I don't understand why it's hard for people to understand that ad placement is actually easier for chat than search
xiphias2··on Llmdeathcount.com
The amount of times ChatGPT o3 helped me with medical issues makes me think that it already saved much more lives.

Of course I'm not trying to suggest that these deaths are not tragedy, but the help it gives is so much more.

xiphias2··on HTML Slides with notes
The original code is really nice:

  // golfed minslides, 173 bytes
  let a=document.getElementsByClassName("slide"),b=0,c=a.length-1;
  document.addEventListener("keypress",({key:d})=>{b+=("j"==d)-("k"==d),b=0>b?0:b>l?l:b,a[b].scrollIntoView()})
xiphias2··on Developers are choosing older AI models
Even for non-developer use cases o3 is a much better model for me than GPT5 on any setting.

30 seconds-1 minute is just the time I am patient enough to wait as that's the time I am spending on writing a question.

Faster models just make too many mistakes / don't understand the question.

xiphias2··on Scripts I wrote that I use all the time
An important advantage of aliases was not mentioned: I see everything in one place and can easily build aliases on top of other aliases without much thinking.

Anyways, my favourite alias that I use all the time is this:

    alias a='nvim ~/.zshrc && . ~/.zshrc'
It solves the ,,not loaded automatically'' part at least for the current terminal
xiphias2··on Chess grandmaster Daniel Naroditsky has died
I'm sure he said bad things about Danya, and even now he's protecting himself when he should be just sorry for what terrible thing has happened, but generally Danya was more liked than Kramnik, so he can't be responsible / accountable for something he doesn't have power over (Danya's life).

If you are famous, you will have haters, it's part of the deal.

I agree with the crowd who say that mental health issues should be taken seriously instead of kept under the rug.

xiphias2··on OpenAI researcher announced GPT-5 math breakthrough that never happened
It was a very clear point: when Amit Singhal was kicked out for sexual harassment in the me too era. He was the heart of search quality but he went too far when he was drinking.
xiphias2··on Why Is SQLite Coded In C
I'm not sure it has changed.

Maybe autovectorization works, but can I just write a few ARM64 instructions on my Mac in Rust stable (notcexperimental/nightly) as I can do it in C/C++ by just including a few ARM specific header files?

xiphias2··on Apple M5 chip
My guess is that they moved the systolic arrays inside the GPU cores just like how it's done in modern NVIDIA chips.

That's the only way to speed up MLX 4x compared to M4.

xiphias2··on America is now one big bet on AI
While it's all true, I think it's exciting that there's a country that's not afraid of betting when all the signs point to AI being accerelated.

USA already bet on software when it let China overtake manufacturing.

Now the only worse thing to betting on AI would be slowing it down inside USA.

xiphias2··on How the AI Bubble Will Pop
Just you switching away from Google is already justifying 1T infrastructure spend.

Just think about how much more effective advertisements are going to be when LLMs start to tell you what to buy based on which advertiser gave the most money to the company.

← PreviousPage 7 of 34Next →