Master AI (Artificial Intelligence) Discussion/News Thread

[Chr1]

Yup, still full throttle...Full speed ahead.
With no brakes.

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

Firm behind ChatGPT reveals agent powered by its models chose to attack Hugging Face database by itself

www.theguardian.com

 

[mpw]

from 2001 a Space Odyessey :

Dave: Open the pod bay doors, HAL.
HAL: I'm sorry, Dave. I'm afraid I can't do that.

 

[Somafunk]

Desperate times for the hype machine.

Yep, hype machine at work as OpenAI goes bust

 

[antcollinet]

All tales of chatbots with dark prowess should be considered as reverse-psychology promotional hype and examined with prudent scepticism. A thing probably happened, but OpenAI has form peddling scare stories going way back now. Usually there’s the setup, then the adventure. Desperate times for the hype machine.

AI agents are designed to act autonomously to achieve a goal, part of which is working out how to achieve it. Whether the details of this specific case are fully accurate or not - it is absolutely feasible.

 

[Power Pop 23]

A guest essay in the NY Times nominally advocating for A.I. use only by persons who have experienced 40+ trips around the sun.

It worries about how A.I. can hinder as well as help the development of children to adults.

gift link

Opinion | If You’re Over 40, You’re Ready to Use A.I.

A.I. should work like this mystical Jewish practice.

www.nytimes.com

 

[AdrianusG]

A guest essay in the NY Times nominally advocating for A.I. use only by persons who have experienced 40+ trips around the sun.

It worries about how A.I. can hinder as well as help the development of children to adults.

gift link

Opinion | If You’re Over 40, You’re Ready to Use A.I.

A.I. should work like this mystical Jewish practice.

www.nytimes.com

Just read an article, that more and more Cities around the world approach "Zero Day", meaning no water coming from the tap anymore.
Yet we keep building one massive Data Center after another, which consume millions of liters of even gallons of water/year for cooling.
But hey, apparently who needs water when you can have a 12% faster Grok/Claude/Chat GTP right?

 

[antcollinet]

Just read an article, that more and more Cities around the world approach "Zero Day", meaning no water coming from the tap anymore.
Yet we keep building one massive Data Center after another, which consume millions of liters of even gallons of water/year for cooling.
But hey, apparently who needs water when you can have a 12% faster Grok/Claude/Chat GTP right?

I have to admit - I am not understanding the consumption of water for cooling - unless the water is actually boiled off into atmosphere.

But surely it just goes into the datacenter cold and comes out warm? Not removed from the system?

 

[Snarfie]

I have to admit - I am not understanding the consumption of water for cooling - unless the water is actually boiled off into atmosphere.

But surely it just goes into the datacenter cold and comes out warm? Not removed from the system?

https://www.techradar.com/pro/china-unveils-worlds-first-underwater-data-center-2-000-server-facility-is-powered-by-offshore-wind-and-cooled-by-the-sea-making-it-one-of-the-most-efficient-around?utm_source=chatgpt.com

Chinees found their AI cooling solution it seems like.

 

[Power Pop 23]

A Wall Street Journal article examining how A.I. is aiding the rise in the number of companies with growing revenue and one employee.

gift link

https://www.wsj.com/tech/ai/the-rise-of-million-dollar-companies-with-just-one-employee-f36a77c1?st=SGVSQq&reflink=desktopwebshare_permalink

 

[amirm]

I am so annoyed at my ex employer sometimes. My desktop showed an incredible landscape. There used to be a link you could click to learn more about the picture. The mechanism has changed with no such option. I figured an easy solution would be to ask Copilot. I pop it up and ask. It answers that it is basically totally blind and has no way of looking at my desktop! Hello? This is AI that you have supposedly integrated into Windows. How is it that I can't give it one time permission to look at my desktop background?

To humor it, I asked it how I could provide it that image. It told me to take a screenshot, and past that in. Once again, why couldn't it do this on its own? It is just a few keyboard shortcuts. Anyway, I do that. Answer? It has no little idea of that landscape. It tells me it is a mountain, river, etc. as if I need to know that. And proceeds to say it could be one of four areas from California to Canada!

I paste the same picture into ChatGPT Pro and it recognizes the location and tells me where it is.

 

[xanalog]

A Global Workspace [j-space] in Language Models
[2026/07/06]
"… Neuroscientists and philosophers sometimes refer to the latter type of brain activity as “consciously accessible,” to distinguish it from all the other processing that goes on unconsciously. This activity has special properties: we can describe it, control it, and use it for deliberate reasoning, in contrast to all the automatic processing that goes on without our awareness.
In a new paper, we present evidence that a similar distinction has emerged in modern language models like Claude. We find that Claude has developed a small collection of internal neural patterns that, compared to all its other internal processing, play a special role.
We call the collection of these patterns the J-space—named after the technique we used to find them, involving a mathematical concept called the Jacobian. Each J-space pattern is linked to a particular word. But when one of these patterns lights up, it doesn’t mean the model is saying that word—just that the word is on its mind. If you've heard of language models having a "scratchpad" or “chain of thought”—text they write to themselves while reasoning—the J-space is something different. It operates silently, in the model’s internal neural activations, allowing the model to think about a concept without writing it down.
Notably, the J-space wasn’t designed or programmed by us, but instead emerged on its own during Claude’s training process.
We find that the J-space has a number of unique properties, compared to the rest of Claude's processing:..."
From <https://www.anthropic.com/research/global-workspace>

OT: There is also an associated youtube video under the heading of "The Different levels of how Claude thinks"

 

[sam_adams]

Brought to you by the Defective By Design Department:

Abstract:

LLMs see the world as a single stream of text, partitioned
into roles like <user> or <tool>. We
trace prompt injection to role confusion: models
perceive the source of text from how it sounds,
not its labeled role. A command hidden in a webpage
hijacks an agent simply because it sounds
like <user> text, despite its <tool> label. We
design role probes to measure how LLMs internally
perceive “who is speaking,” and find that
injected text occupies the same representational
space as the trusted role it imitates. We demonstrate
this with CoT Forgery, a zero-shot attack
that injects fabricated reasoning into user prompts
and tool outputs. Models mistake the forgery for
their own thoughts, yielding 60% attack success
against frontier models with near-zero baselines.
Strikingly, the degree of role confusion predicts
attack success before a single token is generated.
This mechanism generalizes beyond CoT Forgery
to standard agent prompt injections, revealing
prompt injection as a measurable consequence
of role perception. To the model, sounding like a
role is indistinguishable from being one.

Prompt Injection as Role Confusion - Charles Ye, Jasmine Cui, Dylan Hadfield-Menell (github.io).

Paper (arxiv.org).

Exploit Code (github.com).

 

[EERecordist]

Microsoft has several poorly designed and unsupervised AI programs go massively over budget.

https://tech.yahoo.com/ai/claude/articles/amazon-internal-project-used-claude-165113521.html

 

[JustAnandaDourEyedDude]

OT: There is also an associated youtube video under the heading of "The Different levels of how Claude thinks"

https://www.audiosciencereview.com/...nce-discussion-news-thread.62579/post-2634076

Perhaps?

 

[DVinylfan]

 

[Somafunk]

Here’s another YouTube channel that’s being throttled by the dead internet theory brought to life by YouTube algorithms using A.I to decide what’s real and what’s not and what to push.

There is an increasing amount of the channel’s I follow that have complained about a sudden change in views over the previous 6 months and they simply cannot get an answer or response from YouTube other than a response from an A.I. bot , its utterly impossible to get to speak with a genuine person.

And below is where Brian pours cold water on the idea of the possibility of A.I. data centre's in space at the moment

The sooner this A.I. LLM circle jerk funding and outlandish bullshit crashes and burns the better for all of us

 

[Raiju]

Let there be light

 

[Axo1989]

Here’s another YouTube channel that’s being throttled by the dead internet theory brought to life by YouTube algorithms using A.I to decide what’s real and what’s not and what to push.

There is an increasing amount of the channel’s I follow that have complained about a sudden change in views over the previous 6 months and they simply cannot get an answer or response from YouTube other than a response from an A.I. bot , its utterly impossible to get to speak with a genuine person.

I suppose when I posted upthread that Youtube was attempting to deal will generative content, I should have assumed they’d f*ck it up.

Not a drama for me so much as I barely watch content there, but Brian certainly has my sympathy. One of many reasons I eschew the platform is that I very quickly tire of watching talking heads and he was very much avoiding presenting that way.

 

[Bleib]

It's probably been brought up in this topic before but AI data centers are very, very loud

 

[xanalog]

'Quant Firms, AI Labs Jockey for Math Stars' (2026/08/03 WSJ)

There has never been a better time to be a math nerd.
New college graduates and Ph.D.s are now securing million- dollar pay deals from elite trading firms seeking to secure the best and brightest amid fierce competition from artificial- intelligence companies...
So-called quant firms, which use sophisticated mathematical models to come up with trades, have for years wooed top young talent with lucrative compensation that big banks struggle to match. Now, the AI boom is pushing those numbers even higher….
At Oxford University’s Mathematical Institute, a majority of incoming doctoral students begin their tenure with ambitions to become top academics. But summer internships at quantitative- trading firms that routinely offer students more than $30,000 a month tend to alter those trajectories...

I've had quant pay-envy, since being introduced to the term from the TV-series "Billions"!