[K-Dog]

WIRED.COM 2026-09-28

AI Agents Are About to Flood the Workforce. No One’s Ready for It

AI agents are being integrated into workplaces as digital coworkers with humanlike features to encourage collaboration, marketed as efficient and tireless assistants. Though users often feel free to critique them due to their nonhuman nature, emotional attachments may blur accountability, hinder error detection, and pose psychological risks as organizations increasingly depend on autonomous systems.

The research tool agent is having trouble completing the task pipeline.  Not quite ready for prime time.  It can do simple tasks but is falling down where multiple tasks are chained together.  It is in a way, acting like a new hire if you want to take the analogy all the way.

Agents also don't necessarily fit in the same boxes as human employees, Bedard says, in their abilities or their weaknesses.

True that, so for now my agent waits for me to fix it.  Which I will, but I have been hitting the keyboard hard, and need some distance to see what really needs fixing in it.

[K-Dog]

Progress on the Diner Agent progresses.

wired.com 2026-09-29

OpenAI Delays Release of Latest Model Over Safety Concerns

OpenAI has paused development of its GPT-6.1 Astra model after it failed safety tests and displayed misaligned behaviors. The company admitted to delayed disclosure of a government website breach caused by an unreleased agent and is now strengthening safeguards, improving alignment, and enhancing containment protocols amid rising public concern and industry calls for cautious progress ahead of potential IPOs.

More bullshit.  A model in nothing more than a pile of a few billion numbers.  Seriously, that is all it is.  to use it a software runner has to use the numbers to make the model work.  That in itself does nothing more than make the model work.  On top of that is regular software that controls everything.

De tech bros are counting on America not educating themselves enough to see through the bullshit.  And that is a very reasonable assumption.

The danger is giving badly controlled agents tools.  And that is human error.  Capitalism should not be doing all this, universities yes.  If the AI corporations want regulations I thing they should get them.  Give the whole kit and caboodle to the land grant universities with national coordination.  Nationalize the shit.

Nahhh --- that is too big of an idea for Americans.  America will be more comfortable just opposing data centers.  Two minutes of rage a day.  Data centers will get built anyway.  The rage is nothing but performative.  Like in the book.

[RE]

Instead of Nationalizing it, just kill the financial model.  Make it illegal to charge for Tokens. 😀  Take the profit out of it, the Bubble pops tomorrow.

RE

[TDoS]

Instead of Nationalizing it, just kill the financial model.  Make it illegal to charge for Tokens. 😀  Take the profit out of it, the Bubble pops tomorrow.

RE

Great idea. Never happen. The oligarchs obviously won't allow it. Both democans and republicrats would fight it tooth and nail.

[K-Dog]

Instead of Nationalizing it, just kill the financial model.  Make it illegal to charge for Tokens. 😀  Take the profit out of it, the Bubble pops tomorrow.

RE

The money piles already invested will prevent that from happening.  As much as I like it, and I do, your solution is a lawsuit in the making.  You have to nationalize it to make current investors whole, and then immediately stop new data centers and mothballs those near completion using only what is needed now (to be decided by the national board).

It would force companies to spend their own money developing in house, which would make for a much better system.  As it is now ALL the agenic software is TOTALLY INSECURE.  Token use is counted and every prompt goes up to the tech bro monopoly, and every token is visible.  All the AI companies see EVERYTHING you do.  The asses are talking about how their own AI could be dangerous, while they watch everything everybody does.

Input tokens can't go into an LLM or come out of it without being seen.  It is impossible unless you have a total monopoly and promise you won't do any looking at the tokens before you process them.  You only count them.  That could be part of their plan.  Assert themselves as the one and only moral authority as they plunge America into the unknown.  If we don't control it there is no way it can be kept safe they will say.  And they all smile good.

As the money passes under the table the verdict will be.  Yes tech bros are fine upstanding Americans, and as our new royalty they promise to keep your secrets safe.  It is a new social contract.  Nothing to worry about.

Perhaps this little undisclosed fact that we can read all the tokens we charge you for has something to do with the pious smokescreen the bros now fart out.

[K-Dog]

ENGLISH.HANI.CO.KR 2026-09-30

Leading AI experts warn unchecked ‘intelligence explosion’ could lead to ‘catastrophic’ outcomes

Leading AI researchers caution that accelerating automation may spark a rapid intelligence explosion, threatening human control over advanced systems. They call for urgent government measures to evaluate automation levels, enforce independent audits, and create oversight before it is too late, warning of severe consequences such as societal collapse or existential risk.

I was under the impression they all wanted the SINGULARITY.  Does the Halloween season change their minds?

An excellent explanation of what really happened. 
A good description of agents too.

[RE]

Of course, I know it won't happen.  However, an Open Source AI available free to everyone like LINUX which is just as good or better than the ones being charged for could undercut their financial model also.  The Chinese also might offer up at least much cheaper options.

To an extent this is already happening, which is why the price of tokens has been dropping and why financiers are getting nervous about the ROI.  Thus another reason for the safety scare and having goobermint control and licensing.  I think it's going to be very difficult to make a profit on because there is no Patent type protection on the technology.  No patent, no monopoly, basically anybody can pitch their hat in the AI ring.

RE

[K-Dog]

Yes,

which is just as good or better than the ones being charged for could undercut their financial model also.

What the Diner has is good enough, and it only took $10,000 to get there.  This is why a RTX5090 card which I can only dream about now costs $7000 when it should cost $2000.  The 'chip shortage' is artificial and directly intended to benefit the billionaire bros and the capitalist rent model.  Late stage fun and games.

Independent users have to be kept to a minimum for the bros to make bank.

[RE]

$7000 is still pretty low as a bizness investment.  Lots of entrepreneurs could afford that.  You don't need a $Trillion$ IPO.

RE

[TDoS]

What the Diner has is good enough, and it only took $10,000 to get there. 

Can you outline "good enough"? Can I ask it to build python code to do specific tasks with proprietary information, including the stochastic modeling component? Scrape websites for all posts from a given user, put it in some organized fashion? Collate all research material on the web on a particular technical angle I need references for? Search the internet across my 5 favorite news sources and build me a summary? (Didn't Surly1 do something like that years ago? Collapse Chronicles? He still does it, and it is pretty good stuff).

[K-Dog]

What the Diner has is good enough, and it only took $10,000 to get there. 

Can you outline "good enough"? Can I ask it to build python code to do specific tasks with proprietary information, including the stochastic modeling component? Scrape websites for all posts from a given user, put it in some organized fashion? Collate all research material on the web on a particular technical angle I need references for? Search the internet across my 5 favorite news sources and build me a summary? (Didn't Surly1 do something like that years ago? Collapse Chronicles? He still does it, and it is pretty good stuff).

He still does, and he is aware that he is a contributor and moderator here.  He also has never had time to read all the articles to generate a complete summary, nobody does.  That is where a clean honest Diner generated description adds something new to what Surly, and others, and myself have done previously.  And something which you cannot be trusted to do.

Didn't Surly1 do something like that years ago?

Indeed, and Collapse Chronicles is prominently linked on top of this page in a rotating banner.  As you know.