HNHacker News
TopNewBestAskShowJobs

dust42

745 karma · joined February 7, 2025

submissionscomments
dust42··on The Path to Ubiquitous AI
Interesting hardware but I wonder if it is capable of KV caching. Thus (only) useful for applications that have short context but would benefit from very low latency. Voice-to-voice applications may be a good example.
dust42··on How teaching molecules to think is revealing what a 'mind' is
Actually Pavlov did research about the digestive system for which he got the Nobel prize of medicine a few years earlier.

> Did they interview the dogs and ask them if they actively and consciously decide to produce saliva? > Is "dogs have human-like conscious mind" even a scientific consensus?

That's exactly the point - once you have understood the significance of the experiment you understand that it is not important:

A veteran with PTSD can have a surge in adrenaline, heart rate, and cortisol when hearing a car backfiring but he can not suppress it.

Whether the dog was conscious or not about the salivation is completely and utterly irrelevant. In 1907 this was for the first time evidence of a mind-body connection not being accessible to the consciousness. Seriously, forget about the dog. This is all proven beyond any doubt for conscious humans. Nobody cares about what the dog felt.

Associative learning was already known at that time which in its simple form is just circus tricks. The experiment extended this to physiological responses which are not accessible to consciousness in humans.

dust42··on How teaching molecules to think is revealing what a 'mind' is
"It showed that dogs process information from their environment and use it to make predictions"

Exactly that is not what the experiment is about because we all know that dogs will quickly learn the connection between bell and food as dogs are easy to teach new tricks.

If you replace 'dogs' with 'humans', it becomes an empty phrase: "It showed that humans process information from their environment and use it to make predictions" - we all know that.

The groundbreaking part of the experiment was that it showed there are responses which are not part of the conscious mind and which are not willingly controllable by the conscious mind. The dog did not 'decide' to produce saliva.

The experiment was done with a dog because obviously you wont find humans willing to undergo surgery to have the saliva come out of the cheeks instead of into the mouth.

One has to forget about the dog and mentally replace it with a human: now the observation that the human connects the bell with the food is shallow. But the conditioned saliva reflex remains and can't be suppressed - and that is a remarkable insight. It works both with negative and positive stimuli. The latter one being a recipe for a long-lasting happy relationship ;)

dust42··on How teaching molecules to think is revealing what a 'mind' is
Interesting stuff but it hurts so much that the writer has the common misconception of pavlov's dog doing a circus trick. Sure the dog also consciously understands the connection between bell and food. But the physiological reaction of the saliva flowing is not a conscious decision of the dog. Circus tricks with animals existed long before Pavlov. The key discovery is that there is a physiological reaction which cannot be suppressed anymore consciously. That's why PTSD is such a bitch to be treated: even with the stimulus gone, the physiological reaction remains.

That said, the article is still worth a read.

dust42··on GLM-5: Targeting complex systems engineering and long-horizon agentic tasks
> I can only conclude that you're not using these things in any kind of practical way.

I burn about 100M tokens per month. LLMs are like knives, the outcome of cooking depends on the cook and for 99% of purposes not on the knife. There is not that much difference between a $2000 handmade damascus steel knife and a $20 knife.

You can do agentic cooking (aka factory) and you will get ready made meals without human intervention. But it wont make a Michelin star menu.

Same with LLMs and coding, LLMs are an amazing new tool in the toolbox but not a silver bullet. However, that's what they are hyped as being.

Now OpenAI & Co are in the token selling business, which is all fine and dandy but if they manage to become monopolies, then things are seriously in trouble.

Thus if people are fanboi-ing any closed AI I can only conclude that they have already outsourced their critical thinking to an LLM and are happy to go into slavery - or maybe they are hoping to cash in big time on the hype train.

dust42··on GLM-5: Targeting complex systems engineering and long-horizon agentic tasks
They are all just token generators without any intelligence. There is so little difference nowadays that I think in a blind test nobody will be able to differentiate the models - whether open source or closed source.

Today's meme was this question: "The car wash is only 50 meters from my house. I want to get my car washed, should I drive there or walk?"

Here is Claude's answer just right now:

"Walk! At only 50 meters (about 150 feet), it would take you less than a minute to walk there. Driving such a short distance would actually be less convenient - by the time you get in the car, start it, drive, and park, you could already be there on foot. Plus, you'd save fuel and your car would be right there waiting for you after the wash is done."

Here is ChatGPT, also right now:

"Since it’s only *50 meters* (about 160 feet), walking makes a lot more sense.

Here’s why:

* *It’s extremely close* — you’ll get there in under a minute. * *Driving would actually be less efficient* — by the time you start the car, move it, and park, you’re already there. * *Cold starts aren’t great for your engine* — very short drives don’t let the engine warm up properly. * *It saves fuel and emissions*, even if it’s a tiny amount. * You avoid the hassle of maneuvering such a short distance.

The only time driving might make sense is if:

* The weather is terrible (heavy rain, snow, extreme cold). * You have mobility issues. * The car wash requires you to stay in the vehicle and you don’t want to leave it there.

Otherwise, this is a perfect “walk it over” situation. "

Please save us all that talk about frontier and SOTA and that only the closedAI models are any good and the others are all so bad and benchmaxxed. For most purposes a Toyota is just as good as a BMW or a Merc or whatever luxury brand tickles your fancy. Even worse, the lastest 80B Qwen Next is not far from Opus 4.6 but runs on my laptop.

dust42··on Ex-GitHub CEO launches a new developer platform for AI agents
Not even pocket change compared to the billions of VC money burnt every month to keep the show running.
dust42··on Pure C, CPU-only inference with Mistral Voxtral Realtime 4B speech to text model
Same experience here with Whisper, medium is often not good enough. The large-turbo model however is pretty decent and on Apple silicon fast enough for real time conversations. The addition of the prompt parameter can also help with transcription quality, especially when using domain specific vocabulary. In general Whisper.cpp is better with transcribing full phrases than with streaming.

And not to forget, for many use cases more than just English is needed. Unfortunately right now most STT/ASR and TTS focus on English plus 0-10 other languages. Thus being able to add with reasonable effort more languages or domain specific vocabulary would be a huge plus for any STT and TTS.

dust42··on Claude Code: connect to a local model when your quota runs out
The brand new Qwen3-Coder-Next runs at 300Tok/s PP and 40Tok/s on M1 64GB with 4-bit MLX quant. Together with Qwen Code (fork of Gemini) it is actually pretty capable.

Before that I used Qwen3-30B which is good enough for some quick javascript or Python, like 'add a new endpoint /api/foobar which does foobaz'. Also very decent for a quick summary of code.

It is 530Tok/s PP and 50Tok/s TG. If you have it spit out lots of the code that is just copy of the input, then it does 200Tok/s, i.e. 'add a new endpoint /api/foobar which does foobaz and return the whole file'

dust42··on French streamer unbanked by Qonto after criticizing Palantir and Peter Thiel
Well, he destroyed Gawker. Not that I think they were good people. But it was definitely a personal vendetta.
dust42··on French streamer unbanked by Qonto after criticizing Palantir and Peter Thiel
There are now quite a few cases in Europe where the EU or local govs been de-banking individuals. No court, no judge needed. Much more efficient way to shut down critics. We ain't need no people who delegitimize those in power.
dust42··on AliSQL: Alibaba's open-source MySQL with vector and DuckDB engines
He's Chinese and if you had looked into his comment history you'd know this is not someone who uses LLMs for karma farming and looking at his blog he has a long history of posting about database topics going back before there was GPT.

Should I ever participate in a Chinese speaking forum, I'd certainly use an LLM for translation as well.

dust42··on Qwen3-Coder-Next
It is the buffer implementation. [u1 10kTok]->[a1]->[u2]->[a2]. If you branch between the assistant1 and user2 answers then MLX does reprocess the u1 prompt of let's say 10k tokens while llama.cpp does not.

I just tested with GGUF and MLX of Qwen3-Coder-Next with llama.cpp and now with LMStudio. As I do branching very often, it is highly annoying for me to the point of being unusable. Q3-30B is much more usable then on Mac - but by far not as powerful.

dust42··on Qwen3-Coder-Next
KV caching means that when you have 10k prompt, all follow up questions return immediately - this is standard with all inference engines.

Now if you are not happy with the last answer, you maybe want to simply regenerate it or change your last question - this is branching of the conversation. Llama.cpp is capable of re-using the KV cache up to that point while MLX does not (I am using MLX server from MLX community project). I haven't tried with LMStudio. Maybe worth a try, thanks for the heads-up.

dust42··on Qwen3-Coder-Next
Unfortunately Qwen3-next is not well supported on Apple silicon, it seems the Qwen team doesn't really care about Apple.

On M1 64GB Q4KM on llama.cpp gives only 20Tok/s while on MLX it is more than twice as fast. However, MLX has problems with kv cache consistency and especially with branching. So while in theory it is twice as fast as llama.cpp it often does the PP all over again which completely trashes performance especially with agentic coding.

So the agony is to decide whether to endure half the possible speed but getting much better kv-caching in return. Or to have twice the speed but then often you have again to sit through prompt processing.

But who knows, maybe Qwen gives them a hand? (hint,hint)

dust42··on Two kinds of AI users are emerging
From personal experience I'd like to add the last 5% take 95% of the time - at least if you are working on a make over of an old legacy system.
dust42··on Waymo robotaxi hits a child near an elementary school in Santa Monica
Indeed. Rather than having the company telling me that they did great I'd rather make up my own mind and watch the video.
dust42··on AISLE’s autonomous analyzer found all CVEs in the January OpenSSL release
> People writing and maintaining software need to optimize for simplicity, readibility, maintainability. Whether they use an LLM to achieve that is seconday. The humans in the loop must understand what's going on.

In a perfect world that is.

dust42··on Lennart Poettering, Christian Brauner founded a new company
Linux is nowadays mostly sponsored by big corporations. They have different goals and different ways to do things. Probably the first 10 years Linux was driven by enthusiasts and therefore it was a lean system. Something like systemd is typical corporate output. Due it its complexity it would have died long before finding adoption. But with enterprise money this is possible. Try to develop for the combo Linux Bluetooth/Audio/dbus: the complexity drives you crazy because all this stuff was made for (and financed by) corporate needs of the automotive industry. Simplicity is never a goal in these big companies.

But then Linux wouldn't be where it is without the business side paying for the developers. There is no such thing as a free lunch...

dust42··on A few random notes from Claude coding quite a bit last few weeks
> AIs have endless grit (or at least as endless as your budget).

That is the only thing he doesn't address: the money it costs to run the AI. If you let the agents loose, they easily burn north of 100M tokens per hour. Now at $25/1M tokens that gets quickly expensive. At some point, when we are all drug^W AI dependent, the VCs will start to cash in on their investments.

dust42··on Qwen3-Max-Thinking
Max was always closed.
dust42··on Iran's internet blackout may become permanent, with access for elites only
Prepare to go back to newsgroups with NNTP / UUCP. With today's uSD cards that should create a pretty decent, offline store and forward national discussion platform. No programming needed, it's all there already, just forgotten...
dust42··on Qwen3-TTS family is now open sourced: Voice design, clone, and generation
Scroll down on the Huggingface page, there are code examples and also a link to github: https://huggingface.co/Qwen/Qwen3-TTS-12Hz-0.6B-Base
dust42··on Ask HN: Do you have any evidence that agentic coding works?
> This stuff is relatively new, I don't think anyone has truly figured out how to best approach LLM assisted development yet.

Exactly. But as you say, there are so many people riding the hype wave that it is difficult to come to a sober discussion. LLMs are a new tool that is a quantum leap but they are not a silver bullet for fully autonomous development.

It can be a joy to work with LLMs if you have to write the umpteenth javascript CRUD boilerplate. And it can be deeply frustrating once projects are more complex.

Unfortunately I think benchmaxxing and lm-arena are currently pushing into the wrong direction. But trillions of VC money are at stake and leaning back, digesting, reflecting and discussing things is not an option right now.

dust42··on What it's like to be banned from the US for fighting online hate
> You've then proceeded to make a whole host of other assumptions about me rather than engage with any of the points.

I beg your pardon! You said Europeans are at war currently -please don't deny it, it is up there- and being myself European and not being aware of it, I wondered why you are not participating in it then but rather have other people die for your convictions?

So how did I not engage with your points and how is that an assumption? You just stated it!

If you think I didn't engage, then please ask me questions! I asked 9 questions and you didn't bother to answer one of them. Until then I continue to believe that the most fervent supporters of war are those that wont ever participate in them.

Whenever I hear "Ukraine" I personally hear the "fascist, corrupt leadership with golden toilets to sh*t in" and the suffering population who has to endure it. And I know plenty of Ukrainians and am friends with them.

dust42··on What it's like to be banned from the US for fighting online hate
> for only the truly incorruptible would colloborate with Nazis

As I'd assume from your comments you consider yourself truly incorruptible. So does that mean collaboration with the Nazis is acceptable for you? Which actually would be a Russian demand to eradicate nazism in Ukraine?

Are you European? Then let me ask you why don't you join the war if you think the Ukrainians are an ally? Ally against what? Russia conquering the rest of Europe? Germany being next?

I am a friend of the Ukrainians, I have Ukrainian friends and I don't want them to die. I am not a friend of the Ukrainian government nor of people who think this war should continue nor of people who force others to participate in a war they don't want to participate in. The truth is that most Ukrainians don't want to participate in this war. Why not let them decide freely if they want to join the army? Why so many left Ukraine while it was possible? Why are so many soldiers deserting? Why do people get shot at the borders trying to leave Ukraine?

If you are so afraid of Russia, then go there yourself and fight. Why are you still posting on HN? Give those that don't want to die the freedom to not do so. Go yourself and fight yourself. As long as it is others dying for you, your convictions your words are cheapest of cheap.

Always remember: those who are keen to continue a war are not the ones who have to continue fighting it. I can't eat as much as I want to vomit if I see exile Ukrainians demonstrating to continue the war. Why don't they return but let those die that didn't have the means to escape?

dust42··on What it's like to be banned from the US for fighting online hate
Well, the EU was quick to copy this. The Swiss Jaques Baud was slapped with the same measures. I saw only one youtube video of him and in my memory he was well outspoken and considerate - but definitely not mainstream but also not a conspiracy nut.

I simply think it is not right to basically destroy the life of someone without even a court judgement that he did something illegal. He definitely does not fall in the category of hate speech or trying to stir uproar. I think free speech is important for a living democracy. And that includes people with opposing views.

dust42··on High-speed train collision in Spain kills at least 39
There are reports from passengers that the train rattled before the accident. So my guess is a broken wheel rim and subsequently the train derailed at the track switch then also damaging the opposite track. Accident location: https://maps.app.goo.gl/Cek9DgChguXJxVpd6 The italian is about 400m north at the technical building with the two antennas.
dust42··on High-speed train collision in Spain kills at least 39
From the aerial imagery it looks like the accident sequence started at the track switch [1]. The RENFE is rested south of it and the Iryo is north. Quite similar to the 1998 Eschede ICE accident which started with a broken wheel rim and finally derailed at a track switch.

I wonder how anybody knows that it was the Iryo train that caused the accident.

[1] https://maps.app.goo.gl/Cek9DgChguXJxVpd6

dust42··on Erdos 281 solved with ChatGPT 5.2 Pro
Personally, I'd prefer if the AI models would start with a proof of their own statements. Time and again, SOTA frontier models told me: "Now you have 100% correct code ready for production in enterprise quality." Then I run it and it crashes. Or maybe the AI is just being tongue-in-cheek?

Point in case: I just wanted to give z.ai a try and buy some credits. I used Firefox with uBlock and the payment didn't go through. I tried again with Chrome and no adblock, but now there is an error: "Payment Failed: p.confirmCardPayment is not a function." The irony is, that this is certainly vibe-coded with z.ai which tries to sell me how good they are but then not being able to conclude the sale.

And we will get lots more of this in the future. LLMs are a fantastic new technology, but even more fantastically over-hyped.

← PreviousPage 3 of 6Next →