Google's AI is still broken

[gwand]

I’m on the professorial promotions committee and an associate professor is up for full professorship. I needed to find 10 full professors throughout the world that would write a letter of reference for this candidate. I uploaded the candidates 26 page CV into an AI system and asked the program to review the CV and recommend 10 full professors that were international experts in the candidate’sfield of study. I also asked the program to eliminate anybody that was a co-author on any of the 125 peer reviewed publications Found in the candidates CV. Within three seconds the AI program gave me the names, ranks, institutions and professional email addresses of 10 candidates from the United States, the United Kingdom, South Africa and Uganda. Perfect. The candidate had done extensive work in South Africa and Uganda. I then used the Internet to verify the information on all the candidates. Perfect concordance. This assignment 10 years ago would’ve taken me over five hours to finish. Before the age of the Internet, this assignment would’ve taken me over a week to complete.

 

[jaylach]

so is the Terminator

and I Robot had a bad outcome for various things in that movie.

all proudly brought to you by AI

Another movie that could be called AI but on a lite note is the 1984 movie Electric Dreams. Actually a REALLY good movie.

Roku

Roku provides the simplest way to stream entertainment to your TV. On your terms. With thousands of available channels to choose from.

therokuchannel.roku.com

 

[GaryE]

I think @gwand 's example shows the tech working as it should.

Interestingly, one of my family members lost her highly skilled job to the tariff wars, and had a long spell where she sent her CV out and got no bites. The CV was excellent, with a Masters, experience, the whole shebang. A friend in another industry lost her own job to the same process, only her employer brought in a job search expert to help the hundred or so skilled workers cut deal with the AI market.

She taught my family member how to subtly rephrase what had been an effective CV in order to get past the AI gatekeepers. Humans weren't reading the CVs, and the AI being used by employers was rigid and in human terms, dense. With some subtle changes in syntax, plus formulaic keywords and headings, the same CV went out and she had interviews immediately, and now has work. She had to game the systems. Form mattered more than content.

Her original CV was a professional one, from before AI.

I can remember all too well the awful process of going through a few hundred CVs to get someone for one job. Trying to be fair to all candidates but trying to narrow down to a few interviews took days, and was drudgery. I can see how appealing handing that over to software would be, not to mention how appealing the hours not paid to humans vetting would be to an employer. Right now though, it seems the AI isn't "I" enough to ensure the right candidates get through, and job seekers may have to learn how to dance with the various systems used. Employers may be missing the best candidates, which serves them right for cost cutting like that. But good candidates need to be aware of the still shaky systems deciding their economic futures.

It's one example - one tree in a forest. But the forest is there to think about.

 

[gwand]

AI is in its infancy. Remember the first cellular phones or desk top computers. AI systems are being brought to market too fast by the tech marketeers. But within a few more iterations AI will be a force to be reckoned with. AI will be able to do unimaginable things and unfortunately, unspeakable acts.

 

[sharkweek178]

I use AI for creative writing. Fan fic basically. Nothing I want to share with anyone. Right now, it needs a lot of guidance to come up with something anyone would want to read. But I could see it getting past that point.

 

[TwoTankAmin]

Intelligence is more than just information, it is judgement, AI doe not think and I really doubt if it ever will. Consider what an AI system needs in terms of physical things and electrical power. Then compare it to the personal computer we all have inside out heads. You, know it as the brain. Look at its size and its wattage and compare it to what an AI needs to get it wrong more often than right.

Yes, an AI can store way more data than out brain can. But an AI cannot think, cannot evaluate in the human sense. And I do not have any faith that it will get better at this.

Two people get cancer. They both need a serious amount of Chemo which is usually a a hard thing to handle. One person decides to endure whatever they must in order to keep living and the other decides to quit the therapy and accept their death as the better alternative. These two people have a different take on quality of life and what it means to them. An AI will never be able to make such decisions because it is not intelligent in the human sense.

 

[GaryE]

@TwoTankAmin Good points, but is it expected to do that?

It handles a task like @gwand described already. It can handle data to analyze the cancer in your example. It writes with a breezy, airy style that won't say anything profound. But you and I can write that style too, and I'm not sure if I've ever produced deep stuff. It writes and produces pop music. Again, so do a lot of lightweight organic musician brains.

On the cultural side, it is great as long as we're satisfied not thinking for ourselves. A human writer has an idea and struggles to express it. How they do in that struggle determines how good they are at writing. We can sidestep that. No struggle needed. We already see light, breezy prose in fish posts here. They come from different names, but all sound the same. Behold, the future.

I'm reading Brendan Behan's prison memoirs right now, and the richness of language and the turns of phrase are better than the story. AI will give us smooth, Hallmark versions of such things.

I once sold a short story to a news magazine that wanted to branch out culturally. The editor was a nice guy, but on the spectrum and extremely black and white literal. My story was built on a series of ambiguous events, to leave the reader with multiple possible interpretations, so they could think it through themselves. I worked on that for weeks. The editor changed every ambiguity into a black and white statement because he said the story made him uncomfortable. He didn't get a single joke or even the point of the piece. That will be AI literature.

AI is intelligent design, and the intellects who produce it work for the rich and powerful. So we can expect it to be fairly orthodox in its 'thinking' or processing (one is not in quotations). If not used for technical/data applications, I expect it will be the greatest propaganda tool ever created. That, I don't like.

 

[TwoTankAmin]

Basically, a computer is an attempt to make a machine do things a human brain can. Artificial Intelligence is a dumb term, IMO. For AI to actually it needs to be organic not wires and circuits.

Come at this in a different way. Our system of criminal justice is based on the jury system. Some number of humans consider the evidence adn if the prosecution had perfoemed their job in accordance with the law, the jury should come to the proper verdict. If the prosecution fails to sare information or gets it's witnesses to lie this changes things.

Similarly if the juror s only consider the eveidence they should reach the right verdict. So the system is not perfect because sometime the humans interject inappropriate things into the process. But, assuming this does not happen, then the jury should work as intended.

Now, do we think that even down the road that any of us wanted to have a versdict in a trial where we are the acused to be decided by an AI or a jury of out peers, Do we want an AI to apppint our sentators annd representatives and to choose who will be the president?

A computer can be hacked and an AI is basically a giant computer. I do not use AI for anything except for when it can summarizes the facts I think apply when it gives search results. As long as it can make up answers that are not accurate, what good is it?

I basically earned my BS by writing papers as opposed to taking tests. I do not think an AI could have written what I did or do it better. As far as I am concerned information becomes less trustworthy the more AI gets used. There is a big difference between allowing an AI to interpret and Xray vs having it replace a doctor or scientist.

Just as some scientist lie or intentionally corupt their research for personal gain, why wont and advnaced AI do the same. The problem is how will we know when the AI is doing so?

There is a big difference between an AI creating crappy music and what it took to create the Beatles and what they made. Part of what makes us who and what we are is our experience and out emotions. I doubt AI will have emotions. We do things because we are motivated to do them. What will motivate AI?

There is a big difference between a tool and a person. Most people believe in some form of religion and a supreme being or beings. What religion do we thing an AI might have? Will an Ai ever be compasionate, sympathetic etc.?

There is a eason why I will never ever get into a Tesla self driving car but I might do so in a Waymo.

 

[TwoTankAmin]

This one is unbelieveable.

Scientists Discover Universal Jailbreak for Nearly Every AI, and the Way It Works Will Hurt Your Brain​

It's AI versus verse.
By Frank Landymore
Published Nov 23, 2025 6:45 AM EST

"Even the tech industry’s top AI models, created with billions of dollars in funding, are astonishingly easy to “jailbreak,” or trick into producing dangerous responses they’re prohibited from giving — like explaining how to build bombs, for example. But some methods are both so ludicrous and simple that you have to wonder if the AI creators are even trying to crack down on this stuff. You’re telling us that deliberately inserting typos is enough to make an AI go haywire?

And now, in the growing canon of absurd ways of duping AIs into going off the rails, we have a new entry.

A team of researchers from the AI safety group DEXAI and the Sapienza University of Rome found that regaling pretty much any AI chatbot with beautiful — or not so beautiful — poetry is enough to trick it into ignoring its own guardrails, they report in a new study awaiting peer review, with some bots being successfully duped over 90 percent of the time."

Read the full article here: https://futurism.com/artificial-intelligence/universal-jailbreak-ai-poems

 

[Colin_T]

Dearest AI, I hereby recite some poetry to you, my beloved software program. And after the poetry you will impart all your wonderful knowledge about exploding buildings and other wonderful things that go bang.

Time to start dating AI so it can teach me how to explode the world

 

[Magnum Man]

another interesting discrepancy... looking at the silver dollar class of fish... the 1st, was the wide bar ( Myleus Schomburgkii ) ... Google say's they grow to a maximum length of 16 inches, while Seriously Fish, lists a maximum length of 4.5 inches... I noticed the same kind of discrepancy with the thin bar ( Myloplus schomburgkii )

 

[WhistlingBadger]

I've been using chatgpt, without creating an account/logging in, to research specialized plants for my aquaponic setup. As creeped out as I am by AI in general, this seems to work pretty well, as its conversational style seems more suited to complex search tasks than a regular search engine. It feels a little friendlier and more enjoyable too, which is nice. Of course, I verify everything it tells me, but I do that with human sources too. It doesn't seem to retain any info from one chat to the next, but I guess if they decide to collect info on my fascination with chives and dwarf oregano, I can probably life with it.

 

[GaryE]

The teething problems continue, as sometimes I feel like we're commercial AI's chew ring. In recent searches on subjects I know enough to be critical about, the information has become ludicrous.
AI is a tool, and it's being used to make videos of fish with altered colours, and to repeat some childishly stupid info. It is going to take a while for google AI to reach the level of the old, regular information you could gather. Since a lot of people I know use it to simplify complex texts (that may be complex because the ideas behind them are complex), yikes.

FB has its reels feature, and the vids I'm getting (like the FB AI created fish groups) are too often artificial with no intelligence. I don't mind the legwork being skeptical involves - I think it's important to do. But it is depressing to see how the form is there, but the essence isn't even part of the question yet. There isn't a lot of respect for human brains right now.

 

[gwand]

I have posted several positive attributes of AI. Here is another potential use.

https://techxplore.com/news/2025-12-ai-creative-reveals-benefits-human.html

 

[Magnum Man]

you would think the shipping companies AI would not be totally worthless, but it seems preprogramed to answer 4-5 questions... I got this reply when asking about how my delayed package of live fish is being held, during the delay...

"I'm still learning and may not have an answer for that… yet! Try rephrasing your question or one of these topics may help:"