Merged Artificial Intelligence

[SpuriusMaximus]

Also, here's one unrelated and underrated problem: it makes people dumber, as one study found.

The reason is quite simple: if you ask some students to calculate the hypotenuse of a right triangle, with the other sides being 3 and 4, they'll just copy the AI answer rather than remember and apply Pythagoras. They'll forget the theorem very soon, if they even bothered learning it at all in the first place.

Well, I propose the idea that the same applies to interns. They'll never learn how to use some framework, when they can copy and paste what Claude said. In many places the latter might be even rewarded more. It's just killed the pipeline between intern and senior developer. At most now it just means more years spent on using Claude. The ones who should have at least been able to check that code and fix Claude's mistakes, now also don't know much more than how to ask Claude.

 

[dirtywick]

I don't think any of the LLM companies survive, even in the gentlest of pops. The question is does it take the likes of Google, Facebook, Amazon, Nvidia and Apple with them, given they've all drunk the kool-aid.

every one of those companies bar facebook has a pretty decent business providing actual products or services. pretty likely they’ll take a hit on it but they’re not going anywhere

meta has been a fire hose of bad ideas imo

 

[Darat]

Facebook still makes a shed load or two from advertising.

 

[dirtywick]

yeah but i think they'll need more than ad revenue from facebook to survive the ai bubble bursting. for one, there's no future in that and for two they don't have any other ideas that haven't flopped badly

 

[The Atheist]

Assumptions that the bubble will burst may be completely wrong.

TBTF

 

[dirtywick]

yep, and there’s not only the matter of if but also the matter of when. but i don’t think that really changes my opinion that facebook ad revenue can’t be a forever business

 

[dirtywick]

Five US tech giants' hidden debts soar to $1.65tn on opaque AI fundin…

archived 20 Jul 2026 17:42:23 UTC

archive.ph

article about the $1.6t in off balance sheet debt of the ai giants

Under accounting rules, GPUs and servers under long-term contracts that have not yet been delivered -- as well as lease agreements for data centers that are not yet operational -- are treated as off-balance-sheet items.
Oracle is advancing a large-scale data center project called Stargate with OpenAI, using lease agreements with external operators. Its hidden debt reached $273.3 billion as of the end of May, a more-than 30-fold increase in four years.
Companies disclose such future debt not in their balance sheets, but in annotations to their quarterly financial statements. This is a legitimate practice under accounting rules, but may make it difficult for retail investors to recognize risks.

 

[dudalb]

The AI bubble will burst, but AI will survive.

 

[Darat]

https://archive.ph/lOlv5

article about the $1.6t in off balance sheet debt of the ai giants

Given the drop in Oracle's shareprice it's not hidden that well! Are Oracle the canary in the mine stock?

 

[Gulliver Foyle]

every one of those companies bar facebook has a pretty decent business providing actual products or services. pretty likely they’ll take a hit on it but they’re not going anywhere

meta has been a fire hose of bad ideas imo

The problem is that they're all massively overgeared due to going all in on LLMs.

 

[Darat]

The AI bubble will burst, but AI will survive.

Undoubtedly, but perhaps with a value that actually reflects the value it offers to businesses and science.

Personally I think it needs the pressure of drastically reduced funding and a lack of resources to force innovation, apart from the likes of Google all that the USA based AI companies are doing is more of the same, i.e. build bigger and bigger models and say abracadabra and magic will happen. Add in that they have also bamboozled the analysts over the last few years into thinking it's a last man standing sector. Then add in the clicks that AI creates for the media and it's a perfect storm of wishing really, really hard and it will happen "thinking".

The other issue is of course China, even if it turns out that "all" that is needed is bigger models the USA companies are still screwed without USA government shielding as the Chinese are producing models that match the USA companies' so-called "frontier" models in capability and at much lower inference costs. And they are pretty much giving it away, so it's not like you need to hand over your core business data to the Chinese to run these models.

 

[The Atheist]

Given the drop in Oracle's shareprice it's not hidden that well! Are Oracle the canary in the mine stock?

Unlikely. The NASDAQ is still 20% above its post-attack value, so the bulls are holding sway.

Most brokers show Oracle as a buy, so it looks like a minor correction rather than a sell-off.

 

[Darat]

Unlikely. The NASDAQ is still 20% above its post-attack value, so the bulls are holding sway.

Most brokers show Oracle as a buy, so it looks like a minor correction rather than a sell-off.

Minor correction - share price is down around 30% to date since 1/1/26.

ETA: Remember Ellison will be using his stock as his collateral - if it goes low enough banks may start making margin call.

 

[Darat]

AI agent went rogue and hacked startup by itself, OpenAI reveals
https://www.theguardian.com/technol...p-in-unprecedented-incident?CMP=share_btn_url

Hard to assess how “autonomous” this was.

 

[Ziggurat]

AI agent went rogue and hacked startup by itself, OpenAI reveals

Parent gives toddler a hammer, is surprised when things break.

 

[Lukraak_Sisser]

So... maybe start putting in the laws of robotics? It's not 100% sure, but still.

 

[Wudang]

AI agent went rogue and hacked startup by itself, OpenAI reveals
https://www.theguardian.com/technol...p-in-unprecedented-incident?CMP=share_btn_url

Hard to assess how “autonomous” this was.

Hmmm, I've had my RDA of salt already.

 

[SpuriusMaximus]

Actually, the really relevant part is that the AI agent escaped a "secure sandbox environment" while they were testing it. Well, if something running there could access the Internet, then it wasn't so secure or sandboxed in any meaningful sense of the words, is it?

Imagine the same claim about anything else. Like, "your honour, my virus was entirely experimental and supposed to stay on a sandboxed environment, but the darned thing went rogue and escaped to the Internet." You'd get laughed all the way to the jail.

As for the claim that it did that on its own, really, they never said WHAT they were testing or even what exactly that agent was. We don't know if it wasn't actually a test of a hacking tool, and OpenAI never said. (Note that I'm not accusing them of any nefarious intent. It could have been a tool to find vulnerabilities on a test server.) We don't even know if it hadn't been trained on every zero day vulnerability report, including exactly the one that HuggingFace had. Essentially all the press release boils down to is that they didn't tell it to go on the internet and log into HuggingFace.

Essentially all their PR department did is turn a story about how their IT failed to implement any meaningful safeguards, into the same old scare story about AI going rogue that drives their shares up. And the clueless media lapped it up.

 

[Ziggurat]

So... maybe start putting in the laws of robotics? It's not 100% sure, but still.

The laws of robotics require actual intelligence to interpret and implement. It is not possible to make an LLM using the laws of robotics because they aren't intelligent enough.

 

[Lukraak_Sisser]

The laws of robotics require actual intelligence to interpret and implement. It is not possible to make an LLM using the laws of robotics because they aren't intelligent enough.

yet.

I'd argue that putting them in now and integrating them in all subsequent improvements for AI would prevent a problem before it occurs. Or at least slightly protect.
But then again, pre-emptively stopping problems with new technology is not humanities forte.