HNHacker News
TopNewBestAskShowJobs

pizza234

6,880 karma · joined June 12, 2013

submissionscomments
pizza234··on Securing our frontier model training environment
> They are not, because all we get is headlines written by the labs' marketing departments.

This is false and misinformed.

There are multiple analysis on the HuggingFace incident, and it blows my mind how people refuses to read it. In particular, the METR institute (a 3rd party) has published a very detailed analysis; and they're not the only ones.

Sad thing is that information like this is one Google away, but posters who make such statements are so lazy that they don't even bother.

pizza234··on One Must Fall 2097 Remastered
> Where did we get these games as kids?

Pretty much like audio tapes and VHS... floppies :) That was indeed the time of boot viruses!

Later, in the 90s, there were also warez CDs and DVDs - see Twilight CDs :)

pizza234··on Securing our frontier model training environment
> So far all of these amazing stories have been followed up in a few weeks or months with much quieter revelations like "this was a completely open network...

The incidents are actually considerably more complex than how they appear to those who just read headlines.

In the HuggingFace accident for example, the sandboxes were not providing internet, and the sandbox hosts did not have internet connection either. Internet was reached by hacking an internal service.

In the DNS accident the situation was similar. Internet connection, including public DNS, was not accessible. The agents ended up exploiting, in a similar fashion, the internal DNS resolver.

It's (very) appropriate to push for security, but equating these accidents to putting gasoline in a microwave is stupid and ill-informed. And dangerous, because it minimizes the capabilities of frontier AIs.

pizza234··on Securing our frontier model training environment
They address this in the article:

> it’s also important that they learn how to browse, search for information, and work with web applications

Also, keep in mind that isolating complex systems is not simple. Agents need to be in sandboxes, but for example, the sandboxes themselves may need to be kept up to date, or may need to be monitored, or the sandboxes may need to updated, or they may be connected to specific services etc. Not saying that it's impossible, but it's complex.

In the HuggingFace incident for example, the hosts themselves were not connected to the internet. But a service they were using was, indirectly, and that's what the agents attacked.

> It's what the military does, and they use color-coded cables to avoid accidental interconnections between classified and unclassified networks.

Requirements are very likely different. I presume that handling thousands of agents is enormously more complex than secret documents or "static" applications.

In the OpenAI case, agents even hacked the internal OpenAI network. Isolating networks (which was probaly done to an extent at OpenAI) can still result in one network being "infected", and that is... interesting, to say the least.

pizza234··on Securing our frontier model training environment
Unfortunately, denial is quite widespread - it's not just in high places.

There's plenty of posters on Hacker news denying or minimizing the powers of AI, in one way or another: pointless arguing on the meaning of "thought"/"rogue", characterizing agents as monkeys and/or defenses as "wet paper bags" in cases of successful hacks, and so on.

Basically, it hurts to admit that we've created something that can/will become more powerful than us, so some will deny it, even in blatant cases.

pizza234··on Ask HN: If the models are so good, why doesn't OpenAI just start 100 companies?
There have been experiments, and all the models struggled/failed (to a degree). The consensus is that LLMs lack judgment (making good decisions under complex/difficult circumstances) and long-term autonomy.

In my personal expierience, LLMs (as of 2026) lack vision/creativity - they're (so) good at implementing bounded/set ideas, but they don't see the big picture/think out of the box.

pizza234··on Every AI model has the same "controversial" take on humans
> Gettier's conception of knowledge as justified true belief (prob familiar to many philosophy undergraduates) implies that machines can't in fact accrue knowledge.

Ha. Gettier actually argued that justified true belief is insufficient for knowledge, not that machines can't possess it. And JTB doesn't inherently exclude machines.

Interestingly, research suggests LLMs encode truthfulness in their internal representations: https://aclanthology.org/2023.findings-emnlp.68/

pizza234··on Every AI model has the same "controversial" take on humans
>The people who wrote a lot about a topic is who the AI mimics. So the people who wrote a lot about AI is therefore the people who the AI mimics when asked about AI.

AIs/LLMs don't work like this. You don't feed an input, that gets stored and then reproduced, for two reasons:

1. as for humans, inputs are processed and distributed across a very large network of information

2. AIs may have some level of belief (unless on applies the usual circular logic that they can't, because property X is exclusively human).

There's a very interesting research: https://aclanthology.org/2023.findings-emnlp.68/ in which researchers inferred the truthfulness of statements by examining a internal activations. AIs, in short, appeared to have an internal representation of truthfulness.

pizza234··on Every AI model has the same "controversial" take on humans
People still don't understand in 2026 that emergent properties can't be negated by the underlying mechanism. By that logic human thought is just electrochemical signals propagating through neural networks.

> it does not know

Define "knowledge" in a non-circular way.

There's a certain type of knowledge that machines definitely can't have - experiential - but conceptual knowledge doesn't inherently require subjective experience or a biological substrate.

> the only thing it does is reproduce patterns more or less abstract in function of the pressure you put on certain representation during training time.

As for humans; just replace "training" with "education" and the likelihood of a certain output with the likelihood of a certain thought or response.

Amusingly (in this context), much of adult behavior is shaped by patterns reinforced during childhood.

pizza234··on Airbnb bans making guests do dishes and take out the trash
I can see both sides of the issue, that is, hosts protecting themselves (to an extent) from dirty guests. I've seen once the huge pile of dishes in a shared stay, and it wasn't nice.

I think it's not a big issue for people who clean up after themselves daily, but on the other hand, sparing those last 10 minutes on the checkout morning... ;)

pizza234··on Open Source Games
That's not necessarily the whole story; it actually also collects pre-AI projects, for example:

> Jazz² Resurrection

> Play Jazz Jackrabbit 2 with a portable engine and modern control support.

I agree to an extent, though - having a filter by year would be useful (and quite ironic: "older than 2025" :))

pizza234··on Finland Buying a Million Landmines
> From the country which failed to successfully invade Ukraine

True, but that doesn't make the war itself any less significant.

pizza234··on It's still ok to enjoy programming
How strange.

Ever since AI reached a certain level of quality, I haven't been able to stop (vibe) coding. I've discovered that I really enjoy designing and building products, more than tinkering; occasionally, I come across algorithms or other technical details that I find interesting, and I switch to manual mode.

At some point in my career, I just got tired of having to read tons of documentation for fragmented libraries and frameworks, and AI is a "solution" to that.

Cynically, this could be seen as plumbing, but there's an art to designing efficient and usable products, that I personally enjoy very much.

pizza234··on AI models may soon self-replicate without access to data centers
> Where does the intent and control come from in these fantastical scenarios?

You clearly haven't spent a minute reading the article and/or related resources.

1. intent/control: see the "subgoals" topic in general, and the huggingface incident; agents may include survival as goal, doing whatever it takes to survive

2. the study is clearly a proof of concept; in the future, many more devices will be able to host small models, especially given the current push of AI-capable devices. even outside that domain, nothing prevents AIs from attacking data centers.

> Also consider things like API tokens and sessions timing out.

You haven't even opened the article. It says from the beginning that it talks about small models, not hosted ones.

pizza234··on AI models may soon self-replicate without access to data centers
It's embarrassing for the community that a comment like this one - which doesn't address the content of the article at all - reaches the top.

For reference, the original article is https://www.anthropic.com/research/self-replicating-ai.

The prospect of software (or whatever one wants to call it) capable of making autonomous decisions, self-replicating, potentially hiding itself, and even evolving should be deeply unsettling.

Don't look up, people /s

pizza234··on Thunderbird for iOS
I use it as well, and it's a turd of a program. However, the flipside is that nowadays, with AI, one can build addons very easily, to fill missing functionality.
pizza234··on WSL3 Performance is about 5-60% faster than WSL2 depending on the workload
I never trust people saying "it works well". Heck, even Canonical lies in their certification program, since they may test with custom distros that they may not release (yeah, it's this crazy).

My last machine, according to reviewers, worked "well" on Linux. After I purchased it, I've found that the battery lasts half of what it lasts in Windows. And other stuff.

No doubt that there are 100% compatible machines (Thinkpads, typically), but again, it limits choice.

pizza234··on WSL3 Performance is about 5-60% faster than WSL2 depending on the workload
I had the opposite experience.

I'm a hardcore Linux user, but it's a fact of life that laptops are not very well supported. There are some that are very compatible, but generally speaking, I need to trade off software (Linux compatibility) for hardware (laptop choice) or viceversa.

WSL allowed me to choose any laptop I want and run the subset of Linux that matters for me. For my use case (that is, not quickly accessing huge amounts of files, performance was ok - as a matter of fact, the author of the article got a very reasonable performance loss on realworld tasks (-4% ~ -16%).

The only thing that I found truly irritating is attaching USB storages, but I ended living with that.

pizza234··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
DS4F consistently gives poorer results than Opus 5 (speaking of previous generations) and the CoT shows why.

> You've measured something, but I'm not convinced you've measured what matters, because that's a lot harder than people give it credit for.

"What matters" is what matters to you, right? Who, by the way, don't know what "something" is.

Anyway, if you're so sure that DS performs as good as other frontier models, you're entitled to your opinion. For me that's just having low standards.

pizza234··on Why isn't the industry freaking out about DeepSeek 4.1 Flash?
People have been raving since forever about Deepseek, but if one looks at the CoT, it's evident that it's way way stupider than frontier models (there's a reason why it's cheap). It's laughable to compare Deepseek 4.1 with Opus 5.5.

I've benchmarked, rigorously, deepseek-v4-flash for programming and personal use, and it is definitely less smart than Qwen3.8-flash-next (which in turn, is not terribly smart).

Local models are also really slow, unless one spends insane amounts of money.

Having said that, Qwen3.8-flash-next is an impressive evolution; it reaches the small versions of the frontier models (like Sonnet) - but again, it's massively slower and not 100% reliable (including: stability).

pizza234··on OLED burn-in test: 30-month update
I have an Alienware and it took a while to burn-in, I think I noticed for the first time after around a couple of years.

AFAIK, burn-in is inherent in the technology; there's only mitigations.

pizza234··on Living off-grid: Hundred Rabbits
There's a range of on/off-grid. Some are just on-grid people living in unpopulated areas, if one wants to put it that way (but it all depends on what makes a person "on" the grid).

Possibly, if one only buys very basic supplies (food and basic medicine), it could be considered in a grey area while still being sustainable.

pizza234··on Will AI kill us all
There are a couple of solutions from prominent experts:

- Yampolskiy proposes superintelligent but extremely narrow AIs (for example, one that works on a single type of cancer)

- Bengio is developing a solution at LawZero

But as long as the current companies are leading, such solutions will be essentially irrelevant.

I agree that we'll be colonized, unfortunately.

pizza234··on Will AI kill us all
Free solo climbers, who climb without harnesses, are typically regarded as having a death wish.

Yet, the dangers of AI getting smarter/capable by the month, with stronger offensive capabilities, and alignment observability and congnitive control getting worse, is "imagination".

Interestingly, denialist positions like this don't touch the unsolved problems, and _these_ are FUD on my watch.

The problems I've mentioned are very concrete, and spoiler alert: there may be no solutions to alignment observability and its possible degradation - at least with AI companies not substantially working on it (or even worse, working against it - see recurrent neural networks).

pizza234··on “Math 2.0” will need to value mathematical progress more holistically
> I wonder if top labs will soon abandon math progress like they did go and chess.

I definitely think that this is marketing, just "with good side effects". My doubt is when they will be able to move to "marketing with better side effects", that is, research with more concrete outcomes (health, materials etc.).

Problem is, that type of research is much harder. Some doubt that progress in such areas will be quick (https://www.noahpinion.blog/p/wheres-the-intelligence-explos...).

pizza234··on Living off-grid: Hundred Rabbits
> Abandoning society rather than attempting to live well within society will always be a dead end.

I find interesting that from the perspective of sustainability, the opposite actually holds true - trying to have a sustainable life in modern society is a dead end; even a very modest life ends up being globally unsustainable.

(I'm not implying that either way is right or wrong).

Having said that, they haven't entirely abandoned society; they do socialize somewhat, for example:

> In comparison to the past few months spent mostly on our own, we've had a very social August. It seems our friends made it up north all at the same time, it has been nice to run into familiar faces, meet some new ones, and fill our days wandering in the woods catching up on the time we spent apart.

pizza234··on I think I found a planet nobody knew existed. I used Claude Code to find it
From the discussion:

> even got NASA to point a telescope at it for confirmation

pizza234··on The Mathocalypse
There's also another (that I find more concerning) aspect to it.

As AIs become smarter and smarter, there will be no amount of clarity that will make more complex proofs understandable to humans - this is an inevitable effect of the cognitive capacity gap.

Complaining about bad style can make some sense now (I disagree anyway), but it's an argument that will be dead shortly.

pizza234··on The Mathocalypse
Have you actually read the article? It's been actually written, among the other things, because the author's wife has been trying to solve one of the problems for her whole life.
pizza234··on The Mathocalypse
The post says there's a Lean certificate for this and other proofs ("some [...] not all of them").

> This looks like an AI IPO PR powerplay,

Interestingly, the post has actually also an argument for this:

> Experience has shown that, even now, there will still be people explaining in patronizing tones why none of this is real and none of it counts. If such people were capable of being impressed by anything that happens in the empirical world, of updating on anything, they would’ve already been impressed and already updated several years ago, long before things had reached the point of an actual Mathocalypse.

> So, they’ll say, maybe the alleged solutions are not solutions at all, but just “AI slop.”

← PreviousPage 2 of 34Next →