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.
Google replicated the mental state if not necessarily the productivity of a software developer
Gemini has imposter syndrome real bad
As it should.
Is it imposter syndrome, or simply an imposter?
This is the way
Imposter Syndrome is an emergent property
Wait, you know productive devs?
I was an early tester of Google’s AI, since well before Bard. I told the person that gave me access that it was not a releasable product. Then they released Bard as a closed product (invite only), to which I was again testing and giving feedback since day one. I once again gave public feedback and private (to my Google friends) that Bard was absolute dog shit. Then they released it to the wild. It was dog shit. Then they renamed it. Still dog shit. Not a single of the issues I brought up years ago was ever addressed except one. I told them that a basic Google search provided better results than asking the bot (again, pre-Bard). They fixed that issue by breaking Google’s search. Now I use Kagi.
I remember there was an article years ago, before the ai hype train, that google had made an ai chatbot but had to shut it down due to racism.
Are you thinking of when Microsoft’s AI turned into a Nazi within 24hrs upon contact with the internet? Or did Google have their own version of that too?
And now Grok, though that didn’t even need Internet trolling, Nazi included in the box…
Yeah, it’s a full-on design feature.
Yeah maybe it was Microsoft It’s been quite a few years since it happened.
That was Microsoft’s Tay - the twitter crowd had their fun with it: https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist
Gemrni is dogshit, but it’s objectively better than chatgpt right now.
They’re ALL just fuckig awful. Every AI.
Not a single of the issues I brought up years ago was ever addressed except one.
That’s the thing about AI in general, it’s really hard to “fix” issues, you maybe can try to train it out and hope for the best, but then you might play whack a mole as the attempt to fine tune to fix one issue might make others crop up. So you pretty much have to decide which problems are the most tolerable and largely accept them. You can apply alternative techniques to maybe catch egregious issues with strategies like a non-AI technique being applied to help stuff the prompt and influence the model to go a certain general direction (if it’s LLM, other AI technologies don’t have this option, but they aren’t the ones getting crazy money right now anyway).
A traditional QA approach is frustratingly less applicable because you have to more often shrug and say “the attempt to fix it would be very expensive, not guaranteed to actually fix the precise issue, and risks creating even worse issues”.
Weird because I’ve used it many times fr things not related to coding and it has been great.
I told it the specific model of my UPS and it let me know in no uncertain terms that no, a plug adapter wasn’t good enough, that I needed an electrician to put in a special circuit or else it would be a fire hazard.
I asked it about some medical stuff, and it gave thoughtful answers along with disclaimers and a firm directive to speak with a qualified medical professional, which was always my intention. But I appreciated those thoughtful answers.
I use co-pilot for coding. It’s pretty good. Not perfect though. It can’t even generate a valid zip file (unless they’ve fixed it in the last two weeks) but it sure does try.
Beware of the confidently incorrect answers. Triple check your results with core sources (which defeats the purpose of the chatbot).
Is it doing this because they trained it on Reddit data?
That explains it, you can’t code with both your arms broken.
You could however ask your mom to help out…
Im at fraud
call itself “a disgrace to my species”
It starts to be more and more like a real dev!
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.
“Look what you’ve done to it! It’s got depression!”
Google: I don’t understand, we just paid for the rights to Reddit’s data, why is Gemini now a depressed incel who’s wrong about everything?
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.
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
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.
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
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.
Wow maybe AGI is possible
Skynet but it’s depressed and the terminator just makes tik tok videos about work-life balance.
There’s personal time for sleep in the grave.
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.
Wonder what did they put in the system prompt.
Like there is a technique where instead of saying “You are professional software dev” you say “You are shitty at code but you try your best” or something.
Wonder what did they put in the system prompt.
Pretend to be having a mental breakdown so I can write my fluff news article.
