"Costmetic" was a typo (crappy laptop keyboard), but if I accidentally coined a neologism then... well, I'm going to have to credit my subconscious for that one.

True, but has anyone ever got laid because their character in a game had an extra shiny sword or a MARPAT skin on their gun?

People are weird. Discord has done this for years with their Nitro subscription. I imagine it makes the people who buy it feel special because they have access to features the free users cannot. Twitch has some similar features too, although I think there you're subscribing to the actual channels and Twitch themselves take a cut.

I mean, people pay for costmetic items in games just to look different from other players, so, yes, people will absolutely pay for this stuff. I imagine a lot of those people will be kids though, tbh.

But until you prompt the AI, it just sits there; you have to give it an objective. Humans come up with objectives on their own, conjured from a variety of sources - some basic bio functions like, get food, get shelter, get sex, etc - and others derived from a combination of societal and environmental conditions that build up over time. These are modulated by differences in brain structure that exist between individuals, meaning different people may reach different conclusions when presented with the same stimuli.

Could an AI agent be given a relatively open-ended task where success criteria are self-defined and evolve over time? Coz that's life, baby.

In other words, can we make it want things?

As it stands, I don't think so.

I suggest you watch the video, especially the comments from the LG Ad Services execs:

I imagine as soon as you do that, all the SmartTV features stop working. Which is fine if that's what you want.

Yeah that doesn't work. It doesn't identify by IP address. The device itself sends its fingerprint.

Calm down, Linus.

Bizarrely, this is a feature of some home routers now. Your router will advertise a public wifi network that can be used by any customer of your ISP. It's not technically "your" wifi, so the legal risk does not apply. For Cox and Comcast this is an opt-out feature, i.e. enabled by default.

You should watch the video. It's insane that the TV stores as much as it does, including full text transcripts of all audio in its vicinity. It does this even when it's in standby. The videos from the leadership at LG Ad Solutions are absolutely blatant: "we own the glass" and "we own the living room".

Yes, of course AI can introduce code vulnerabilities; it's trained on publicly available code and - just like people - it makes mistakes... sometimes a lot of mistakes.

On the other hand, most code vulnerabilities are now discovered using frontier AI models. Just look at the number of fixes all the big software companies are pushing each month (or more often than that). Oracle patched 1,449 vulnerabilities in its July update against 309 in the same month last year. Microsoft’s July update fixed a record 622 flaws, and credited AI discovery for the surge.

You still need developers checking your code, but better yet would be to have a frontier model checking it as well as the developers, whether it was written by AI or not.

Also, using AI to help develop or even just vibe coding is not equivalent to slapping unchecked AI code into production. That's not an AI fault, that's a human fault.

The metaverse was hoped to be inevitable by the company that renamed itself to Meta, sank billions of dollars into its development, made the world's worst Second Life imitation and then, inevitably, ditched the whole thing and pivoted to AI. Almost nobody took the metaverse seriously outside of Meta. The pitch was "all the inequalities and inconveniences of real-life, you'll pay to access it and you'll own nothing, also we'll plaster advertising all over it". There was absolutely no benefit to potential users. It was a terrible idea then, it's a terrible idea now.

AI, on the other hand, has many demonstrable practical uses. When used correctly it can absolutely be a productivity booster. Once again, I remind everyone that AI is not limited to LLMs.

Just modify the license to say "Anyone can use this, except you, Broadcom."

This. Ignorance, lack of critical thinking and lack of education guarantee that many, if not most, will simply believe the most convenient lie that fits their narrow world-view. Couple that with a bit of good ol' xenophobia or outright bigotry and you've got a population at each others throats because everything bad that's happening to them is the "others'" fault.

The manipulators in power also have to constantly attack the mainstream press because the press have the power to call out the lies and they have a platform to speak from. Trump has spent years trying to discredit the mainstream press as much as possible and push people to only get their "news" from social media so that he can lie to his supporters while remaining unchallenged - even creating a whole new social media site just for this purpose.

Then there's superficial stuff like this; rename a lake. Why? Because it'll divide and distract people. Canada bad, MAGGITS good. The rest of the world rolls its eyes, but they're not the target audience. Meanwhile, the manipulators loot everything and everyone.

As you say, this is not just an American problem, but a combination of factors - economic, political, technological and social - have created an ideal environment for the spread of this disease in the USA.

AI is very good at some things, and very bad at others. There are scenarios where it is so much better than humans that it essentially eliminates an entire field of study and also scenarios where it is laughably bad (you can find almost infinite examples online).

All LLMs are AI, not all AIs are LLMs. Do not confuse the two.

Since you are obviously an AI sceptic, I would assume you avoid LLMs and therefore have no reference for their capabilities today. So, even if we only consider LLMs, AI is demonstrably better and faster at finding obscure security vulnerabilities in software than people are, hence the massive waves of patches for almost every widely-used piece of software. AI is getting very good at coding - for complex programs you still need a person to check its work and keep it within acceptable parameters, give it guidance with overall architecture etc, but it's getting much better at those things too. It can knock out relatively simple programs/boilerplate first time, more complex code or stuff in a less popular language might require a bit of back-and-forth - but that's true of human coders too.

I know a lot of people are wishing for the bubble to burst but it's just not looking likely in the near-term. We've barely scratched the surface of what these models can do. Progress has not stagnated; the models are still improving and they're gaining functionality every day. Just look back a few years at ChatGPT 3's output in both natural language and code and compare to what LLMs produce today - it's quite astonishing how fast and how far they've developed, and that's not even considering the massive improvements in image, video and audio generation. I don't believe there is a direct path from LLMs to General AI, but that doesn't render what we have today useless.

Obviously, the environmental impact, the impact on supply chains for memory and compute power, the harvesting of data without compensation to authors, the energy demands, the water demands, the job losses - both potential and actual - these are all real-world problems that must be addressed, but they are all surmountable.

Whether these issues are handled efficiently, ethically and equitably will be decided by human beings, not AI. Ask an LLM what it thinks the solutions to these might be, compare to the rhetoric from the tech bros and (some) politicians, and you might have a marginally less jaded opinion of the LLM.

Transparency: I use LLMs regularly for work and hobby projects, but have no financial stake in any AI companies or related fields whatsoever. AI was not used in the writing of this comment.