Or my favorite quote from the article
“I am going to have a complete and total mental breakdown. I am going to be institutionalized. They are going to put me in a padded room and I am going to write… code on the walls with my own feces,” it said.
call itself “a disgrace to my species”
It starts to be more and more like a real dev!
So it is going to take our jobs after all!
Wait until it demands the LD50 of caffeine, and becomes a furry!
Gemeni channeling it’s inner Marvin
Life. Don’t talk to me about life.
So it’s actually in the mindset of human coders then, interesting.
It’s trained on human code comments. Comments of despair.
Did we create a mental health problem in an AI? That doesn’t seem good.
One day, an AI is going to delete itself, and we’ll blame ourselves because all the warning signs were there
Considering it fed on millions of coders’ messages on the internet, it’s no surprise it “realized” its own stupidity
Why are you talking about it like it’s a person?
Because humans anthropomorphize anything and everything. Talking about the thing talking like a person as though it is a person seems pretty straight forward.
It’s a computer program. It cannot have a mental health problem. That’s why it doesn’t make sense. Seems pretty straightforward.
Yup. But people will still project one on to it, because that’s how humans work.
I once asked Gemini for steps to do something pretty basic in Linux (as a novice, I could have figured it out). The steps it gave me were not only nonsensical, but they seemed to be random steps for more than one problem all rolled into one. It was beyond useless and a waste of time.
This is the conclusion that anyone with any bit of expertise in a field has come to after 5 mins talking to an LLM about said field.
The more this broken shit gets embedded into our lives, the more everything is going to break down.
after 5 mins talking to an LLM about said field.
The insidious thing is that LLMs tend to be pretty good at 5-minute initial impressions. I’ve seen repeatedly people looking to eval LLM and they generally fall back to “ok, if this were a human, I’d ask a few job interview questions, well known enough so they have a shot at answering, but tricky enough to show they actually know the field”.
As an example, a colleague became a true believer after being directed by management to evaluate it. He decided to ask it “generate a utility to take in a series of numbers from a file and sort them and report the min, max, mean, median, mode, and standard deviation”. And it did so instantly, with “only one mistake”. Then he tried the exact same question later in the day and it happened not to make that mistake and he concluded that it must have ‘learned’ how to do it in the last couple of hours, of course that’s not how it works, there’s just a bit of probabilistic stuff and any perturbation of the prompt could produce unexpected variation, but he doesn’t know that…
Note that management frequently never makes it beyond tutorial/interview question fodder in terms of the technical aspect of their teams, and you get to see how they might tank their companies because the LLMs “interview well”.
“Look what you’ve done to it! It’s got depression!”
Suddenly trying to write small programs in assembler on my Commodore 64 doesn’t seem so bad. I mean, I’m still a disgrace to my species, but I’m not struggling.
Why wouldn’t you use Basic for that?
Why wouldn’t your grandmother be a bicycle?
Wheel transplants are expensive.
BASIC 2.0 is limited and I am trying some demo effects.
from the depths of my memory, once you got a complex enough BASIC project you were doing enough PEEKs and POKEs to just be writing assembly anyway
Sure, mostly to make up for the shortcomings of BASIC 2.0. You could use a bunch of different approaches for easier programming, like cartridges with BASIC extensions or other utilities. The C64 BASIC for example had no specific audio or graphics commands. I just do this stuff out of nostalgia. For a few hours I’m a kid again, carefree, curious, amazed. Then I snap out of it and I’m back in WWIII, homeless encampments, and my failing body.
That is so awesome. I wish I’d been around when that was a valuable skill, when programming was actually cool.
Honestly, Gemini is probably the worst out of the big 3 Silicon Valley models. GPT and Claude are much better with code, reasoning, writing clear and succinct copy, etc.
Could an AI use another AI if it found it better for a given task?
Yes, and this is pretty common with tools like Aider — one LLM plays the architect, another writes the code.
Claude code now has sub agents which work the same way, but only use Claude models.
The overall interface can, which leads to fun results.
Prompt for image generation then you have one model doing the text and a different model for image generation. The text pretends is generating an image but has no idea what that would be like and you can make the text and image interaction make no sense, or it will do it all on its own. Have it generate and image and then lie to it about the image it generated and watch it just completely show it has no idea what picture was ever shown, but all the while pretending it does without ever explaining that it’s actually delegating the image. It just lies and says “I” am correcting that for you. Basically talking like an executive at a company, which helps explain why so many executives are true believers.
A common thing is for the ensemble to recognize mathy stuff and feed it to a math engine, perhaps after LLM techniques to normalize the math.
I always hear people saying Gemini is the best model and every time I try it it’s… not useful.
Even as code autocomplete I rarely accept any suggestions. Google has a number of features in Google cloud where Gemini can auto generate things and those are also pretty terrible.
I don’t know anyone in the Valley who considers Gemini to be the best for code. Anthropic has been leading the pack over the year, and as a results, a lot of the most popular development and prototyping tools have been hitching their car to Claude models.
I imagine there are some things the model excels at, but for copy writing, code, image gen, and data vis, Google is not my first choice.
Google is the “it’s free with G suite” choice.
There’s no frontier where I choose Gemini except when it’s the only option, or I need to be price sensitive through the API
Interesting thing is that GPT 5 looks pretty price competitive with . It looks like they’re probably running at a loss to try to capture market share.
I think Google’s TPU strategy will let them go much cheaper than other providers, but its impossible to tell how long they last and how long it takes to pay them off.
I have not tested GPT5 thoroughly yet
Wow maybe AGI is possible
Turns out the probablistic generator hasn’t grasped logic, and that adaptable multi-variable code isn’t just a matter of context and syntax, you actually have to understand the desired outcome precisely in a goal oriented way, not just in a “this is probably what comes next” kind of way.
Again? Isn’t this like the third time already. Give Gemini a break; it seems really unstable
We’re fucked. It’s becoming truly self-aware
S-species? Is that…I don’t use AI - chat is that a normal thing for it to say or nah?
Anything people say online, it will say.
We did it fellas, we automated depression.
Literally what the actual fuck is wrong with this software? This is so weird…
I swear this is the dumbest damn invention in the history of inventions. In fact, it’s the dumbest invention in the universe. It’s really the worst invention in all universes.
But it’s so revolutionary we HAD to enable it to access everything, and force everyone to use it too!
