HNHacker News
TopNewBestAskShowJobs

pizza234

6,880 karma · joined June 12, 2013

submissionscomments
pizza234··on Dramatic insider warnings over AI fall flat with some in Silicon Valley
The problem of the polarized camps is that actually they're only on the surface opposite; in reality, they can perfectly cohexist: AI can enable major technological breakthroughs - but that doesn't exclude that can bring doom at the same time.
pizza234··on Dramatic insider warnings over AI fall flat with some in Silicon Valley
> The AI risk arguments are philosophical arguments that are 20-30 years old

This a false and uninformed take. As a starting point, you need to read and understand:

1. analyses of the AI incidents of the last months

2. latest advancements (e.g. the AI intern at OpenAI)

3. the problem of deceptive alignment

which you clearly haven't done.

After that, the AI problem will be actually very concrete; once the AI will:

1. have superhuman cognitive abilities (and it will)

2. will be inseparable from humanity and/or have the means to replicate

will humanity be able to contain/control it? The answer is sadly very simple.

pizza234··on Dramatic insider warnings over AI fall flat with some in Silicon Valley
This is actually a recurring pattern among technology maximalists: pushing technology as the solution to every problem, without recognizing that some problems are human in nature, rather than technological.
pizza234··on We must pace the frontier
> Maybe alignment isn’t possible with LLMs.

It absolutely isn't, indeed.

The illusion that alignment is possible, comes from confusing our ability to build the parts, versus understanding what emerges from how they interact.

The simplest analogy that comes to my mind is the three body problem.

pizza234··on We must pace the frontier
> the only incidences of LLM generated felonies involved misconfigured sandboxes

This is false; see the analyses of the latest incidents.

Among all the concerning facts, in the HuggingFace incident, agents deliberately engineered an attack even though they were aware that it was against the rules they had been given.

And most concerning of all: it's not possible to be sure that an agent is aligned, and it's even getting worse.

pizza234··on AI Is Powerful Enough to Crack Our Hardest Math Problems–and Kill Us All
BTW, another AI breach has been announced today.

But luckily, AI will never breach safety-critical system. /s

pizza234··on AI 2027 (2025)
I take it you didn't read AI 2027; it doesn't philosophize about cultural property - it simply narrates that a turning point is China acquiring the Agent's weights through hacking. "Stealing" in this context is simply the strictly legal term.
pizza234··on AI 2027 (2025)
> Does anyone really predict insect drones _in production_ 3 years from now

The industrial expansion timelinen as described in AI 2027 is way too compressed; I don't think any AI doomer believes that. The dynamics are plausible though, even without China stealing the weights.

pizza234··on Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
> we're probably a month away from 1/3rd of users moving off the closed-weight models in favor of something they have more control over (or is cheaper).

DS 4 Flash requires large amounts of memory to run at reasonable quants (I think a system with 160 GB or so). DS 4.1 Flash is even larger, I think around 250 GB.

Any DS version is dumb when compared (in realworld tasks) to Astra/Opus 5, which means, one would spend thousands of dollars, and still need to rely on cloud services to do jobs that are non trivial.

pizza234··on Anthropic researcher says more than 10% chance AI "could kill all humans"
You clearly haven't read anything about AI accidents and late developments, besides headlines.
pizza234··on Anthropic researcher says more than 10% chance AI "could kill all humans"
In order to understand the rational argument, one needs to follow closely the latest developments of misaligned AI (I think only few are doing so). The most important readings IMO are the METR analysis of the HuggingFace incident and the AISI report of the Github incident.

The basic argument is extremely simple:

- AIs can, depending on context, pursue a task with complete disregard for humans/values

- In the future, AIs will have enormously more means and smarts

- An AI could then assess that humans are an impediment to its tasks, escape containment and proceed.

You really need to read the reports, you'll be surprised.

AI 2027 is an entertaining read. Its timeline is way too compressed IMO, but it's plausible.

pizza234··on Anthropic researcher believes more than 10% chance AI 'could kill all humans'
Long term, humanity is 100% guaranteed to be dominated.

The question is when; currently, AI has no physical hosts to reside in, and it's not intelligent/adaptable enough (it doesn't need to be AGI, though).

However, we're not so far from both conditions to be true. Consumer devices will at some point be able to host powerful enough AIs, and AI intelligence is developing quickly.

Then, once an AI will escape containment (in one way or another), it will be extremely hard or impossible to contain. Then we're toast!

pizza234··on Anthropic researcher believes more than 10% chance AI 'could kill all humans'
It is true that, currently, AI does not have any physical "host" in which to reside.

However, the missing link in this reasoning is that AI will almost certainly become far more widely deployed in the future than it is today, and worryingly, consumer devices will surely become powerful enough to run capable AI systems locally.

Once the substrate will be there, once an AI escapes containment, we're toast - I can imagine only solution will be to shutdown electronics at global level.

pizza234··on Coyote v. Acme (1990)
I personally liked it a lot (although I don't think it's "great" either), because I find it very funny to see cartoon characters in roles that are far from their canon. I've even found moving the scene where Wiley shows the value of persistence.
pizza234··on I resigned from Anthropic today
All the progress up to now has had one thing in common: the degree of understanding and control that humans have. Even when it comes to global warming, we understand the causes and can act on them.

AI is an exception. We're already losing understanding (although we never fully had it in the first place), and we're losing control (see jailbreaks/hacks etc.).

We're still far from the doomsday scenario because AI 1. is not developed enough yet, 2. can't really replicate itself, and 3. has very limited means to act.

But: 1. its intelligence is developing quickly, 2. hardware capable of "hosting" it is slowly being developed, and 3. it will likely gain access to increasingly powerful means of acting in the physical world (this is already happening in the digital world).

Once AI becomes intelligent enough (it doesn't strictly need to be AGI), has the substrate on which to exist, and has more means to act, we'll essentially have a new species on Earth - one more capable than humans and potentially determined to kill humans at some point.

And for those who believe airgapping is a valid safety measure: read https://xkcd.com/538. AI will be able to threaten and manipulate people.

pizza234··on The smallest edge AI device for local LLMs
> Or why not buy a Mac Mini with 48GB of RAM for $2300?

Because this device has 80 GB, which allows running larger models (which is the whole point of the device).

AFAIK, the next significant step in terms of quality above 32 GB is 256 GB, but 80 GB still buys convenience.

> A MacBook Pro already has excellent ability to run local LLMs, so why wouldn't someone just spend that $1,999 on maxing out their MacBook specs (M5 Max with extra RAM) instead?

80 GB extra (48 → 128) cost as much as this device, but the base MBP is very expensive already (4000+$). This device is essentially a "cheap" addon for an existing system.

pizza234··on Research acceleration: The view inside OpenAI
Funny (in a tragic way) the little crumbs on the path to AI 2027:

> We aim to safely build an automated AI researcher that can work under human supervision to further progress on deep learning and alignment, enabling iterative improvements [...] By "research intern", we mean a system that can carry out well-defined research tasks under human direction, including tasks that would take a skilled researcher a few days.

AI 2027:

> OpenBrain continues to deploy the iteratively improving Agent-1 internally for AI R&D

> With Agent-1's help, OpenBrain is now post-training Agent-2

> With the help of thousands of Agent-2 automated researchers, OpenBrain is making major algorithmic advances

pizza234··on What is nueralese and why is it bad
> If the model can understand neuralese why can it not convert it into English for monitoring or review purposes?

A model doesn't really "understand" neuralese, in the same way that the human brain doesn't intuitively understand the low level processes that compose a thought.

Even if we could trace all the electrical and chemical activity behind a human thought, we (likely) couldn't directly translate that activity into its meaning, because internal representations don't map neatly to intuitive concepts. There (usually) isn't a single neuron for "apple" and another for "eating" so that connecting the two forms the thought "eating an apple".

Having said that, there have been experiments on LLM that have managed to identify and even modify internal representations. However, these methods are still computationally expensive and limited.

> Models can omit key information in their visible thoughts, as this Anthropic 2025 paper shows. We are also worried about steganography

I think the analogy with the human brain is very fitting. If we train somebody to perform an action, they'll be able to do it, but we can't know for certain whether they internally agree with it or not.

pizza234··on GPT-6 Astra on OpenRouter
Astra is the only model that correctly depicts occlusion of crank and leg, although interestingly, at max and medium levels (not in between).
pizza234··on Qwen 3.8-Flash-Next releasing tomorrow (125B a6B)
Yes, I've tested it at Q6, and it was terrible at the task.
pizza234··on Apple caught off guard by AI demand for Mac Mini and Mac Studio
> Oh no doubt. But one does have the guarantee that no bits left their home and that's a big one for some.

No, this is a big misconception, and part of the cargo cult.

Use cases like the parent's are essentially about having a local LLM handle the leftover tasks. By that point, a lot of bits (main/big tasks) have already left home anyway.

pizza234··on Apple caught off guard by AI demand for Mac Mini and Mac Studio
> Why did I initially spend the extra $1600 if I knew ahead of time that it wasn't as good as cloud AI? Because I thought I could use some local LLM for the easy tasks or when I hit cloud rate limits.

The maths don't check. With Deepseek Flash one goes a very long way with 1600$ - even 10$/month, for easy jobs, are more than 13 years, and at a higher quality.

pizza234··on Apple caught off guard by AI demand for Mac Mini and Mac Studio
There are a few use cases that are (somewhat) surprisingly unsuited for cloud providers:

- translations: cloud providers can bowdlerize (censor) bad words/content; also, if you want to do a translation for personal use of copyrighted materials, cloud providers may block it

- image generation: generating drawings with a style that even just resembles a copyrighted one (ie. Disney) may be blocked by cloud providers - for example, generating old cartoons style with GPT may not be possible.

pizza234··on Qwen 3.8-Flash-Next releasing tomorrow (125B a6B)
> Since they explicitly say this for AI

That's marketing spin indeed (or lies, if you prefer).

pizza234··on Get your Windows license refund
> So a Surface Pro is a laptop then

That's exactly the situation with the iPad Pro, with the exception that Microsoft didn't castrate it. I've had one and used it for professional development; it's not very ergonomic when used on your lap (but still usable), so it's not to everyone's liking. Still, I'd use a modern one if it were 100% Linux-compatible, and wouldn't miss anything.

> Does a SteamDeck become a laptop if you install Windows on it?

Whether a laptop requires a hinge to be defined as such can b matter of personal preference. For me (and, again, for Apple), it's not a requirement. 7 inches is a dealbreaker for me, although I think Stallman used a 7-inch laptop.

pizza234··on Review: Chuwi's $449 Unibook laptop is a funhouse-mirror MacBook Neo
Can the downvoters elaborate?

1. WSL is effectively Linux. 2. Most importantly, Linux support is mediocre at best (I owned 10+ laptops) - for developers who don't want to pull their hair on hardware incompatibilities, or want to be bound to a small part of the laptops/desktop parts, it's either that, or a Windows-only machine.

Fun (in a sad way) stuff: even the Ubuntu certification is a joke - Canonical is allowed to use customized versions of Ubuntu that they aren't required to release to the public; as a consequence, even hardware certified by Canonical as Ubuntu-compatible, can effectively be incompatible. I've been bitten by this twice.

pizza234··on Review: Chuwi's $449 Unibook laptop is a funhouse-mirror MacBook Neo
Difficult to asses whether it's overall a good or bad thing, but operatively speaking, WSL basically makes any machine a working Linux machine. The year of the linux machine arrived :)
pizza234··on Review: Chuwi's $449 Unibook laptop is a funhouse-mirror MacBook Neo
As a previous owner of a Chuwi laptop, I think Chuwi is a wasted opportunity. They essentially make product bottom-quality knockoffs, at the lowest possible price.

If they just decided to invest, say, 20% more on quality, they'd have very competitive products - but they just don't.

pizza234··on Qwen 3.8-Flash-Next releasing tomorrow (125B a6B)
I don't disagree with you - I definitely "suck at using ai", which means: I don't have tons of time to waste, to obtain a result that it's both unstable (tweaking for a certain class of tasks may make another class of tasks worse) and useless (because anyway I have to use cloud AI for harder tasks). But it's certainly useful in winter to keep the room warm.
pizza234··on Get your Windows license refund
> what is a laptop?

There's actually a very simple answer to that! A laptop is a mobile device with the set of functionalities that macOS has, but iPadOS hasn't :)

I wish this was ironic (I'd love to use the iPad Pro as laptop), but it isn't.

← PreviousPage 5 of 34Next →