Best thing about AI is: any fool can use it.
The worst thing about AI is: every fool will use it.
Unsafe at any speed: AI optimists are turning cautious as safety concerns mount
OPINION It wasn't that long ago, like earlier this year, that AI leaders were doing their usual pie-in-the-sky proclamations of how everything would be so wonderful with AI. Anthropic CEO Dario Amodei tweeted that AI "will actually be possible to cure most human diseases in ~5-10 years." While on a podcast, OpenAI CEO Sam …
-
-
Friday 28th August 2026 10:44 GMT I ain't Spartacus
It's just more marketing. The fact that the AI company CEOs themselves are the ones warning about how amazingly powerful and intelligent their large language models are is the clue that it's just more of the same. In the last 3 months they've both issued an amazing new model that finds bugs so fast that it can't be allowed out (except to "special" customers who get advance access) - because it'll cause a security apocalypse. And then they're leaked and are about as good as existing "AI" bug hunting tools. i.e. some bugs found, but loads of false positives.
It's still, "all aboard the hype train!"
"AI" is going to reshape society by causing a massive global market crash. Only after we've picked through the wreckage of that will we find out what these LLMs are going to be useful and financially viable at doing - as it took the dot.com bust to shake out the crap from the early internet.
-
Friday 28th August 2026 17:00 GMT doublelayer
I think you need to read that article more closely than the headline. He did say that we should get used to LLMs staying around, but not because they're great and everyone should agree. Most of the text is a list of reasons to hate LLMs, which really doesn't fit with your "how wonderful AI is for everyone" summary. He does give it some credit for capabilities and improvements which I think are too speculative, but another one of his points is one I think is true: even bad uses of LLMs will keep existing because people want them.
There are people who express the dream that the popping of the LLM bubble will act like a time machine back to 2022. It won't just be that the companies lost a bunch of money, people who thought these things were intelligent will realize they're wrong, everyone who built one will take their LLM customer service down, and we won't have to hear the word "chatbot" ever again. That sounds nice. It's not going to happen. Smaller models can still simulate that, stupid people who think they can do things they can't can still believe that of something they can run cheaply enough that they can invest in that faith, and early LLM chatbots got psychotic users, so people who get attached to that tech can still find it. When the headline says "AI's here to stay", that's part of what it means.
-
-
Saturday 29th August 2026 19:30 GMT doublelayer
They aren't different opinions. You think the other article says something completely different from what it actually says. Your summary doesn't match the text. But also, last week's is an opinion about what we should think long-term, whereas this one is about what AI fans think short-term, so even if they did disagree with each other, which they still don't, it wouldn't be a conflict.
-
-
-
-
Friday 28th August 2026 10:41 GMT Bebu sa Ware
"smart sex toys for interpersonal use that rely on AI"
I was wondering what "30 percent of people admitted to using AI for sexual purposes" meant (especially the bold part.)
Following the link in the article it is mostly just the old agony aunt, porn mag, phone sex etc stuff but as you can observe the idea of embedding AI in your sex toys (or connecting them to an external service) had occurred to someone.
In future if someone says they have been fucked by chatgpt (etc), it might be prudent not to enquire too closely.
-
Friday 28th August 2026 12:46 GMT BoHu
"AI under the control of the technically clueless"
Best to "start making sure that first graders, or even pre-Ks, have A.1. teaching every year" then, and right away! ;)
-
-
Saturday 29th August 2026 21:26 GMT John F Jennings
Re: Haven't these 'experts' ever heard of...
Too true
Give an example.
I gave an Ai a request to design a rowing boat. Paid up enterprise copliot.
I am somewhat experienced in building small boats, and I could give the AI access to all the information I use, while it wasnt a skill I use in my day job.
I gave it the specs of what I wanted, what tools I have and the materials I wanted to use, as well as performace desired.
It eventually did it - it took over 95 clarificaitons - most of which would be impossible if you wernt fairly experienced in the first place. The first iterations would have killed someone or it got splines wrong.
Oh yes, it apologised and used every NLM trick in the book and corrected itsself - I would have been faster designing it myself - and it would have been a much prettier and functional product. BTW, the time taken to produce it - 28 hours in front of the computer, before I got (remotely) feasible plans. It normally takes me about 2 days to sketch and plan. The 28 hours did not include designing the brief.
I am actually going to try to build its proposals because really think its nigh on useless, but I enjoy a challenge...
-
-
-
Friday 28th August 2026 16:20 GMT Anonymous Coward
Yeah, good luck with keeping teens from inappropriate materials. It didn't stop me ...
I think the hope is not so much that they will never see inappropriate material; but that (again, *hopefully*) they wouldn't then get a feed that is algorithmically driving them down a rabbit hole where all they see is more and more of that kind of material, and material of ever increasing extremity at that.
-
Saturday 29th August 2026 08:36 GMT Anonymous Coward
Re: Yeah, good luck with keeping teens from inappropriate materials. It didn't stop me ...
A decade or so ago, you could indeed go to a "nasty" site and find the shocking images, but creating the images was difficult and they would not follow you out.
Now, if you see one - even by accident - your social media and even advertising will chase you with a thousand more, many AI-generated.
That force-feeding is what's really changed.
My wife tried to find some genuine news about the Nepal/Tibet flood, and her entire social feed across all of Meta became horrific videos of dismembered body parts and worse.
She still cannot sleep.
Kids are being sent the same, and worse.
-
Saturday 29th August 2026 21:04 GMT ravenviz
Re: Yeah, good luck with keeping teens from inappropriate materials. It didn't stop me ...
Video / photographic based media is the problem, which your brain interprets like you are there, witnessing local and interpreted personal tragedy.
Radio / podcast media is one step away, like someone telling you a story about something that happened to them, so still worrisome, but not so direct.
Written media is like a report of something that has happened somewhere different to where you are, so while informative, and perhaps worrisome, is not immediately stressful.
Better steer clear of video based social media, just don’t go there.
-
-
-
Saturday 29th August 2026 15:30 GMT Fido
Trapped in a Bubble
Right now all the major chatbots have an option to turn off use of previous chats for personalisation and context. Turn it off to avoid being trapped in a bubble and pray that remains possible.
My opinion is an AI built to not do what people ask will be worse moving forward than an AI that consistently tries to do what people ask. Said another way, the misalignment that comes from guardrails may be more dangerous than a monkey with a torch.
-
Sunday 30th August 2026 15:59 GMT jlturriff
True Names...and Other Dangers
It's rather amazing that in 1984 (!) Vernor Vinge, seemingly prescient, wrote True Names which features an agentic app named the Mailman, which was created by a government agency like DARPA and tried to overtake everything, just the way LLM agents seem to be going.
-
Thursday 10th September 2026 11:03 GMT Jonjonz
Big Techs Save the Children Playbook Comes Out Again
The last time Big Tech rolled out the "Save the Children" playbook was when social media cried that they could not moderate to protect the children as long as they were liable, and congress handed them 230. Social media then took that took the governors off the engagement at any cost algorithms, kicking moderation to the side.
This time around the big players want a mote to protect their massive investments, freedom from liability for their AI platforms, federal override of state restrictions on building data centers, and the big one affirmation that training on copyrighted/proprietary data is legal.
-
Thursday 10th September 2026 18:56 GMT lucidnightmare
"My primary concern is the risk that authoritarian governments—not solely the Chinese Communist Party (CCP), although the CCP is clearly the most capable threat—build AI models that are more powerful than those built by the US, and use them to achieve permanent military superiority or perpetrate incredibly deep repression of their own people. "
But the current US regime _is_ authoritarian.....
