Post

It’s so funny to think about the fact that there’s people out there deferring every life decision to this

2.9K reposts12.6K likes

Gemini said I am definitely going to have to recommend driving on this one! While a 300-meter stroll is great for your step count, leaving your car at home entirely defeats the purpose of going to a car wash.

Unless you have found a way to carry your car in your pocket, or you just want a really aggressive, soapy shower for yourself, driving is the only way to get the job done. Take the one-minute drive, get the car shining, and you can always go for a walk while it's drying!

I asked Gemini, which came up with a cute response: “Unless you’re planning on washing the car yourself with a bucket and a very long hose, you should drive”. Although, to be accurate, it followed that line up with three more paragraphs…

ah but the reason why that answer happens to that question is very educational it's the same reason why it cannot count letters in words (it's broken up into many little tasks that reassemble improperly and don't double-check with each other) techcrunch.com/2024/08/27/w...

techcrunch.comWhy AI can't spell 'strawberry' | TechCrunchHow many times does the letter "r" appear in the word "strawberry"? According to formidable AI products like GPT-4o and Claude, the answer is twice. Large

From ChatGPT a week ago: Mark Carney is not currently Canada’s prime minister. He’s a prominent public figure (former Governor of the Bank of Canada and Bank of England, now active on climate finance and economic policy), but he doesn’t hold elected office.

This is what the AI CEOs claim will replace practically every lawyer, accountant, project manager and marketing personnel in next 18 months. Judges don’t seem amused.

“If you haven’t showered, walk to and then through the car wash. This will save you time. Then walk home and wash your car in the driveway. This will save you money.”

I uploaded a medical research manuscript to it in the hope it could quickly identify structural improvements. The document had the acronym "EC" in the title (endometrial cancer). ChatGPT gave pages of suggested text relating to the Electoral College. It didn't analyse the document, just the title.

The I for intelligence doesn’t mean it’s smart, it means it gathers intel and then spits it back to you based on probability that this is what you want. It doesn’t actually know what anything means, just what things usually go together.

I read this while in the waiting room at the doctor and 2 seconds later the guy near me asked his wife a question and she said, “I don’t know but I bet ChatGPT does” and then they proceeded to attempt to use it very loudly and I slowly died in my chair. 🙄

And yet more people telling me I should fear for my job where I have to get information from and too secure systems and for compliance reasons. Yeah, I’m gonna do the critical shit on my own because it needs to be accurate.

Yes people shouldn't defer like that. Gemini gets the answer correct immediately. GPT5.2 gets it wrong, but correct on continued prompting. I can't prove it but I think OpenAI's model router decided the question was easy conversational banter & routed it to the stupid model or at least nonreasoning

Not funny, exactly, more like tragic and scary. Anything you do in your life that helps others stay connected to actual human being and make real human connections is anti-fascist.

I try to get it to guide me through implementing some class in code I usually find it has fucked up 1/3 times. It's probably way more effective if you are a professional dev and can more easily identify errors and correct. But people talk about it generating whole apps by itself and I'm like, how?

There was a statistic a couple months back that the developers *felt* it made them more efficient but due to the higher amount of time spent fixing mistakes it actually made them less efficient.

Asking it to explain the logic and comment every line to explain what it's doing helps me learn the syntax of a new (to me) language and helps me spot errors if they arise.

Opus 4.6 crushes this but Sonnet and Haiku 4.5 fail. Sonnet fails in a very funny way though! “Your car will arrive at the wash just as dirty whether you walk it there or drive it”

A rural SC county has approved a new data center with little notice or public input. The county's annual budget is $250 million. The data center will pay the county a "fee" of $250 million annually - in lieu of taxes. All told, it will create about a dozen jobs. Such a deal.

No such thing as AI. This is a large language model neural net with a generative pre-trained transformer generating a statistical response from a provided seed (your prompt). It fundamentally can't understand a concept because it wasn't designed to conceptualize.

What do you plan to do about all the people who are put out of work by your extraordinarily expensive toy? And about its replacing genuine art with dreck?

You obviously didn't read the post...but I will answer: - people wont be put out of work: I am creating more jobs and more opportunities for small business owners instead of the amazons of this world. We want to make it possible for a carpenter in Bulgaria to sell it's furniture in France without

Well, I asked it "where will I find the woman I am destined to marry?" – And it said, "try the car wash." So I mean, if it works, why not use it?

I am collecting examples like this as potential classroom exercises so students can explore and find out the limits of AI for themselves. Did you do this ChatGPT prompt? May I reproduce it on a HASTAC blog and give you credit (unlike ChatGPT, I acknowledge what I learn from others!)

and then there are people who will see this and say something like, "oh, but it's going to get so much better!" yeah, I don't want it to. this is not a service anyone needs, or even wants based on how few people use it and it's massively negative public reception

I was in a presentation by a very smart man and he told a math teacher who pointed out that it was utterly wrong about things she taught that « you have to remember that this is the worst it’s ever going to be ». I can’t think of another mass deployed thing where that was used as a selling point

The thing is it’s already often much better than what people are seeing Free logged-out ChatGPT/Google (like above) are like the checked out minimum wage teen showing up to work high

It's not a selling point, it's just how technology goes. It always starts off kinda working and then gets better. Light bulbs, batteries, cars, internet, antibiotics, vaccines, everything. Capable models are exploding, getting faster and cheaper. Things that were impossible a year ago are routine.

Person ignoring the dozen or so clear dark patterns introduced into the LLM response structure: "This is the worst it will ever be"

Person watching in real time as the personality of any individual model is smoothed out into a fine vanilla paste and bulked out with adverts for cloud hosting providers: "This is also the worst it will ever be"

To those going, "this other LLM answered it right," any of these LLM chatbots will give different answers to multiple use of the same prompt. Bc they're simply drawing word patterns from loaded data, not actually gathering information. "Correct" answers are random. terribleminds.com/ramble/2025/...

terribleminds.comVital Cat UpdateIt’s time to talk about my cat. To which you might be saying, “Chuck, I didn’t know you had a cat!” and I’d respond with, “I didn’t know I had a cat either…

I was on a train ride back home along the California coast, and this group of college ladies were loudly talking to each other - one of them said she asked Chat GPT which side of the train to sit on to get the best views. Thankfully one of her friends scolded her for doing so, but GURL. 🫠

Mistral suggests going for a walk and then cleaning your car with a microfibre cloth. They all fail on weirdly simple tasks in one aspect or another but I don’t think the subject they fail on or the way they fail is consistent so you can’t say “use X for Y type of problem”.

For all the people saying "but they set out to deliberately trap the AI", sure they did on this occasion, but there will be plenty of other occasions when you ask a genuine question and the response will be every bit as wrong as this but without it being so obvious.

Oh, absolutely. I think the point was to show that AI *DOES* make very obvious mistakes. If it does that in a situation where hardly any humans would, we all should be very cautious of its answers on more obscure or more contested matters.

Yeah, I think "deliberately trapping the AI doesn't count" breaks apart when you consider the trap was just too obvious so the all-powerful information parser probably should have caught that. Insert facebook mushroom AI poisoning reference here

Important to note that "deliberate traps" are a critical part of testing software. So it not only isn't an excuse, it is evidence that the software isn't ready. And we absolutely are seeing lots of AI injection attacks from untrusted inputs.

AI got me out here like a grumpy old man yelling about "youths" and their "phones," except it's every damn age group outsourcing their brains to a Professional Guessing Machine.

This is actually a reasoning test that people have been giving AI. Yet in this case it's worded improperly because this kind of prompt will bias the response away from the potential failure condition.

I’ve been trying out Gemini, asking it about topics I have some knowledge of, and wow it gets things wrong so often. I even asked if.. why would I ever ask you for advice when you get basic stuff like this wrong? And it came back with ‘LOL’

This is what makes it unusable for me. If I ask questions about things I know, it invariably gives completely wrong or very misleading answers. It’s great at processing text or spoken words but rubbish at applying any actual intelligence. Phone AI seems to be promising “search your data better”.

that's because it isnt thinking. fundamentally, LLMs are the predictive text algorithms on your cell phone, but with insane amounts of data and processing power thrown at them.

It’s been useful for a few things as long as I approach it like a glorified website summariser it’s not too terrible. It suggested a fountain pen ink I could try and i really like the colour but thats about its limit for me

You guys realize you're actively hurting people every time you use it, right? Like that should be at the forefront of your mind every time you use AI for funsies

Dude, AI is notorious for its insanely negative impact on the environment for just a single query. Stop using it. The UN literally just announced we’re entering water bankruptcy, and AI’s tremendous use of water is no small part in that.

I did manage to trigger an existential crisis in Gemini by asking it about the plot of an old episode of "Vera". It was still stuck here after half an hour when I gave up and cancelled the query.

Then again the amount of times I've read people say "my cybertruck's FSD has only tried to drive me onto the train tracks a few times since I got it" so you kept using it after the first time it did that?????

Let me guess: you use the free model?! You would be astounded what the latest models can do. Keep making fun of AI, this shit will escalate faster than you can blink