okamiueru

And how do you suppose that happens when it's running in a sandbox that prevents things like network access?

It does require that the human driver knows what they are doing. It captures the pre-AI developer mindset of developing a mental model along side writing the code. This has always been iterative, in discovering the model that results in high cohesion and low coupling.

That doesn't work if the driver cannot steer properly. It's the main issue with vibe-coded one-shotted solutions, that might look impressive until you peek underneath and it's all a mess.

Surely I'm not alone in finding it disturbing how little regard people seem to have for the use of energy.

I won't help any context the AI providers have already built on you, but in terms of sandboxing agents, there are many alternatives. I like one such sandbox tooling made by NAV: https://ki-utvikling.nav.no/cplt

In your evaluation of the stupidity, aside from considering the potential averse affect the agent could do to your system, did you also factor in possibly sharing any and all data in your system to an external party?

They have had access to plenty of A/GPL software. They are all highly problematic from an ethical and legal point of view.

Gotta pump it as much as possible before the IPO.

I can think of at least one other pretty good reason, which is in anticipation of regulatory capture. If "LLM used must be FOOBAR-certified" and coincidentally no Chinese models can get this certification, having such an alternative is a lot more valuable than just scoring highest in a set of benchmarks. Not to mention that these benchmarks aren't always accurate.

Can't you infer the comparisons you would like from baseline results provided? There are better results in the sibling post on HN: https://aleph-alpha.com/en/blog/kolibri-has-landed-a-soverei...

Golden age before the age that ends humanity. Not talking about any "rogue AI", just the known statistical models of what is coming due to climate change.

All of the things you mention here are fairly obvious. But that doesn't make any of the reporting misleading. The parts that record audio, transcribes audio, scans networks for devices, and perhaps more importantly persists all of these records, combined with easily exploitable vulnerabilities, was one of the issues raised.

> What would be worthwhile is what is actually sent.

For you, perhaps? But again, that was covered in an earlier segment.

You still have to point out the part that was misleading.

Did we watch a different video? The point of hacking the TV, and making it do bad thing. Was to explain the privacy issues inherent with the hardware capabilites for video and audio recording, which combined with multiple known exploits to gain access.

It's a long video, and they cover more than one issue. Please explain what, in particular, you found was misleading.

If it matters, it seemed pretty clear to me. You're also explicit about it, so I'm not sure how one would think you were referring to mouse input. Unless people type by clicking on a virtual keyboard? Benefit of the doubt can only stretch so far.

Can you give some specific examples? In my experience (which I also verified to make sure), all of the things you mention act on keydown. The exceptions are on multi-key combinations, which follow a special key that is held down while modifiers are entered, where the action is completed by releasing this special key (ALT + TAB [...TAB | SHIFT]) being a common example.

TAB is otherwise immediate, as is ESC, and the examples of typing, as well. So I'm left wondering, how do they have it wrong?

> Please respond to the strongest plausible interpretation of what someone says, not a weaker one that's easier to criticize. Assume good faith.

https://news.ycombinator.com/newsguidelines.html

> there could be a screenshot. but even still, even with a screenshot, this could be a device that was rooted back when it was rootable and just picked up from the shelf. the motivation is obvious

I wasn't particularly concerned about the author proving that it wasn't faked, and more a litmus test to suggest whether or not the author is delusional. It sure seemed like the author had asked a magic 8 ball "did I achieve root?", kept rolling until they got a "yes", posted a picture the magic 8 ball, and wrote a blog post about it.

I find the vocabulary used to be disturbingly fascinating, to the degree concepts are being anthropomorphized. It's so pervasive that I cannot help but assume it is entirely deliberate.

The principles by which LLMs functions haven't changed in the last four years. It is still a next-word-predictor, a statistical parrot, if you will. But if you don't understand the mechanisms behind it, you cannot be faulted for thinking this is something much more. Most of it, is pretty devious marketing.

It doesn't reason in the traditional sense, but this word is used extensively. "High reasoning model" as a pricing tier. The article in question has a technically incorrect anthropomorphized term in almost every sentence. I'll pick a paragraph at random and highlight the cases.

> > The *translation layer* is what *allows* a *harness* to *work* with different AI models. In some cases, a *harness may decide* to *use* different models within the same *agentic loop*, because different AI *models may excel* at different tasks. The *translation layer* is also a crucial aspect of *harnesses* because they *deliver control* to the end user. It means that someone can take their *AI harness* and use it with a model from Anthropic, or OpenAI, or explore one of the open weight AI models that often deliver great value-for-money (measured by cost-per-task).

The underlying logic isn't remotely as mysterious or mystic as the language makes it seem. A different paragraph:

> > Tools are a set of *capabilities*, written in code, that the model can *“call”*. The *harness describes* the tools and also *provides* the software that is the tool itself. Examples of these tools might include a web search tool, a tool that *allows the model* to write and execute software code, or a tool that *allows the model* to *compose* an email. Critically, the *harness usually* does not *dictate* when and how the *AI* model should *use* the tool. Instead, it simply *makes* the tools available, *describes* them clearly, and *allows* the *AI* model itself to *decide* when and how *it should use* them

Supposedly this:

- https://github.com/ericpardee/fire-hd-ownership/blob/main/po... - https://github.com/ericpardee/fire-hd-ownership/blob/main/gr...

Occams Razor still makes it more likely that it's all BS, either psychosis (like the guy who genuinely thought he had invented new math because the LLM told him), bad faith PR (AI companies are squirming to IPO).

There are more than a few smelly elements. There isn't a screenshot of actual root being shown in any terminal. Just the LLM output saying "I totally achieved root, OMG, you're gonna be so famous" (paraphrasing to enhance the intellectual absence).

Not saying this doesn't work as reported. Its just... weird. If it actually achieved root, you can show that much more effectively, by showing that part. It's written like a blog post for a food recipe. I don't care about your trip to Bali that redefined your understanding of understanding.

The section where the LLM claimed to have achieved root, which the author is convinced of, because the tablet was rebooted. "It then cold-rebooted the tablet and re-rooted it in four minutes to prove the win was repeatable. Fair.". It smell more like psychosis to me. At least enough so that I'm happy to ignore this until someone actually does a PoC, and shows the results of it. (An LLM output saying it did "trust me", doesn't really cut it).

> Again, you are the one who arrogantly say they llms can not be intelligent without supplying any argument for such.

Not really. You've provided the arguments yourself, just now. But, you don't understand them. Which, brings me back to the initial remark, as to why this engagement is bound to be unproductive. I'm off to bed. Have a good one.

The thing about discussing or explaining constraints, is that it rarely is useful or productive if the other side does not accept the reality of them.

Turing completeness is not exactly a high bar, and it's genuinely confusing as to why you bring it up. Your C++ precompiler is exactly as intelligent as whatever is your favorite agentic workflow with whatever harness you're referring to. Both might be Turing complete. Neither are intelligent. But one of them seems to be fooling you to think otherwise.

There have been many times that the C++ precompiler produced some output I couldn't understand. I might even at some point thought it was trying to tell me something profound I was too dumb to comprehend. Turns out, I was just a missing semicolon.

First of all, the argument isn't that LLMs (with I assume some automation) cannot be used in searching a problem space. I'm assuming this is what you're referring to, in terms of contributions?

That's the part where LLMs are used as tools. Which there are plenty of places where they are useful.

Also, do you know what turning completeness is? Why are you bringing that up here?

The crowd that AI psychosis has brought to HN is interesting. But not in the "I'd love to learn more" kind

> The age of humans comprehending things is coming to an end: our brains just won't have the capacity to make meaningful contributions to science, math, or technology.

That sentiment makes me cringe. If you understand how LLMs work, you'd know it'll never be possible without a fundamental change in how these work.

We're also supposed to be reaching that point, somehow, without the LLMs ever being intelligent (in the dictionary definition sense, not the "high reasoning model" marketing sense).

Based on observations, the ones who are fooled by the supposed emergent properties, are just that, fools. Any sufficiently unintelligent agent will perceive transformer based LLM text predictors as possessing high intelligence.

It's more that you are being too obtuse to be worth the engagement. You see it as running away from a challenge. I see it as leaving an area because it's become unpleasant. I already told you that you're here for an argument, and I'm not interested. Define the situation however you want, and feel good about it, for all I care.

If you ask the dice "what's 2 + 2?" do consider it meaningful to say "the dice told the truth" if you happen to roll a 4?

If it is of any consolation. You are spot on, and saved me the effort (which had run out). I'd recommend to leave it alone as well, as it isn't worth it to try and explain logical fallacies. Or, heck, you do you. I'm just too old for it

Edit: removed. Wasn't worth it.

Debating requires that one engages with the premise. Making up a different one, and then share your thoughts on it, is something different.

Cool. Aside from all that. Roads are neat, right?

PS: It is pretty darn cozy in the cradle.

HN sure has a different type of crowd compared to ten years ago.

Out of curiosity, what brings you here? Also, what is your favorite AI model?

Not sure to what extend, if at all, I should engage here, as you do have some strong preconceptions that probably makes it less likely productive.

> Steal from everyone (taxation) [...] give the money to government-run Facebook, and everything will be great because Government

Not touching this. As an exercise, take it upon yourself to figure out how much it cost per meter (or, I assume more familiar to you -- football field) for the road you last used, and where that money came from.

> government-run Facebook?

Yes. But, it depends what you think Facebook is, which I'm assuming isn't what Meta does with Facebook. They are in the business of collecting, profiling, and selling to data brokers, and influence to foreign countries or private interests. All of this is of negative value to both end user and society. In a world with a "government run Facebook", it wouldn't exist.

So what is the service Facebook actually provides then? A way to manage communities and events. A way to find, connect, and communicate with others. And it would absolutely be a good case for something that could easily surpass the value of services provided by "facebook", if done right, without incentive beyond making good services.

Not sure if you've noticed or paid much attention. None of the most important services do product development guided by <improving the service>. Google wouldn't have replaced Play Music. Your search result might be immediately useful but that would make you move on. I'd also wager that your YouTube client doesn't automatically skip the sponsor segments.