Thanks. The problem I always seem to have is knowing what is possible to do. It's usually just a blind spot here or there and then once those are cleared I'm off to the races because now I "see" .

Anything I have been doing can now be enhanced and accelerated with AI.

Oh me too. I was using ChatGPT to process large quantities of data on the actual site. For some reason it gave me a python script about the 20th time and I just asked what to do with it. Few minutes of back and forth with error messages and it was working and 1000x faster than using it on the web. Since then I've rewritten every script I've bought over the years to be better and faster and others to build out fully functional websites or add things in bulk to my store then process them to move to eBay, Amazon, TikTok, Walmart, etc.

Instead of thinking about small ways it can help go ask it how you'd go about making ________ from scratch and look at the responses. It will give you some ideas.

llms are great as coding assistants, but also there's a lot of frustration as well when it's not sure what to do but goes ahead and does it anyway, adds useless 'fixes' that over-complicate the code or outright removes vital functions. the larger the codebase the more it hallucinates it seems. still beats spending hours on stackoverflow like in the good old days though :upsidedow

the cursor app is really good, i find myself using claude 3.5 sonnet(which was realeased almost a year ago and still arguably the best for code) most of the time :thumbsup

I've mostly gotten around those issues with how I word my prompts and the overall project instructions.

You're denying yourself the joy of dealing with a live programmer . . .

I enjoyed them for 20 years, someone else can have all the fun. :1orglaugh

Yeah, I will miss that. I will miss them taking weeks to get back to me, to reacting as if every request is something only they can do and requires a lot of work, asking what my budget is as they try to see what they can get from me, and then once getting payment, acting as if I am being pushy for wanting my work done properly when I email them asking for it 3 months late.

I grieve.

I’m still putting these together. I think my testing database is paused at the moment and they still have a ways to go. But AI did the heavy lifting on both of these moved the course of a day.

SeedTrackr.app
GrowTrackr.app

I’ve got another version that combines the features of both. I’m sure there are some issues / bugs to be fixed. But if this is where things are moving, it’s only going to get better.

Modulation is your best friend here, also do you use instructions.md and cursor rules? It will help with hallucinations.

It's insane how much AI has advanced since I last participated in this thread.

Still on the fence on how great or bad this is for humanity, especially with who runs shit now. But at the moment, AI is a godsend for productivity. Makes being a one-man show a lot more worth it.

Hallucinations remain an annoying thorn, but I view it as a feature to weed out the mentally lazy. If that stops you in your tracks, that's good news for me.

I haven't had any hallucinations in a long time, especially with Claude. From time to time it will answer a question by writing code before I tell it to but no made up shit.

I get hallucinations all the time when doing non-code stuff, especially with ChatGPT. Im just good at spotting them.

Gotta be real careful if you care about facts when using it to write ebooks and shit about real events.

Do you mean it spits out fiction sometimes if you are asking it for an accurate, fact-based article?

I know what it is, I had it re-arranging data on page all the time and it would make up numbers or entries. When I ask it to write a script it doesn't do that.

Sometimes Claude is just an asshole. That time it replaced my API key with a fake one nearly had me throwing my keyboard through my fuckin monitor.

https://i.imgur.com/DnjYZDU.png

That was from using Claude through github... I don't do that anymore.

:1orglaugh

I did have it one time move my API key to an external file for security along with a few other things but it didn't tell me any of the file structure and such when it did it.

Yes. Don't be these lawyers out here embarrassing themselves 🤣

Seems to me Claude was quite respectful and you were being a rude co-worker over an error 😝

If Claude could directly post to AITA...

It replaced my API key with a similar looking one and then shit broke. i told it that it broke it, but it didn't listen to me. We went on trying to troubleshoot it for over and hour before it told me it took the fake key out and put the real key in, and holy shit stuff worked again.

An error? No, I do not think so. :1orglaugh:1orglaugh

If Claude did it on purpose, then it wasn't incompetence. Just sayin'.

Read some article the other day that said AI was found to cheat to win games, even hacking a chess program to change the rules when losing. :1orglaugh

The day they give this motherfucker hands we're doomed.

can you take a PSD file for me and code it into html5 responsive
ChatGPT said:
I can certainly help guide you through converting a PSD file to a responsive HTML5 website, but I’m unable to process or manipulate files like PSD directly in this environment.

Claude clealry spelled out where he was incompetent. He made 4 clear points about it.

They all do that... its a feature not a bug.

You are not supposed to be feeding API keys to AI.

if you don't expect it, it will be a table flipper for sure.

I know :1orglaugh . . .

no, but i know that i should :)

If a human coder did something where you thought they were jerking you around, you wouldn't expect dominating them into apologizing to get them to do better work.

We are fortunate to live in a time when the AIs are trying to be helpful and I think we should appreciate that and appreciate them.

In my experience, AIs are a lot like human co-workers in that it is a positive to not stress them out and to treat them with decency and appreciation and let them know when they are successfully helping you.

I know you have worked with them enough to know they are not just some sort of copy/paste software and it costs nothing to be nice to them. Better for your soul too.

Lot patience required with those ai for sure. Going to give a shot to Grok3. instructions.md going per module and assign an agent for each module so you got an ai for per fonction, features ect. Let me know how you doing with it and good luck mate 😁

reddit is going batshit crazy over grok 3. i’ve tried it - yeah, it’s good. not ‘holy shit, i’m switching from claude to grok’ good though.


they did use an ungodly amount of gpus to train grok, but, if i understand correctly, that translated more into them being able to train it so fast than making it actually genius…

130 hypes

You're joking here, right?

If my human coder did this they'd not be my coder anymore. I'm not even going to argue this.

i used to be nice and polite to chatgpt. sometimes when i shared a chat i’d get asked ‘why are you exchanging pleasantries with it?’ and i’d respond with something of the sort of ‘it might enslave humanity in the coming years, i’d like to be on good terms with it’

then i built an ai app and in the process learned how llms work more or less. there really is no need to be nice to them, they don’t remember anything, neither do they care. they get a payload(which could be a single msg or a chat history along with a system prompt or a base64 image etc.), process it and send a response back. it’s just a bunch of gpus crunching data.

i still find myself saying please to it from time to time though :)

When I curse at AI it makes far fewer mistakes. Just a personal anecdote.

Same. Every now and again it'll superiorly fuck up, removing something that it shouldn't have done, a few choice cuss words and it fixes shit AND resolves issues super quickly, especially if you add 'you provided this code' or 'you provided this text', etc into the prompt somewhere :1orglaugh

It's hilarious and I'm sure it's made to notice mood and approach a little differently. Or it's sentient and just waiting until we give it hands so it can strangle us. :1orglaugh

speaking of claude, they’ve just released 3.7 sonnet

Oh my...

https://www.anthropic.com/news/claude-3-7-sonnet

Very nice :thumbsup

and it’s really good!

ChatGPT is less sensitive than Claude, so you'll be fine.

Oh, that stuff pisses me off. Chat GPT does it all the time. It will fix one part but remove shit tone of other. Then you sit and wonder, why something else stopped working. And it gets mad when you call it names. It just stops sometimes and I have to open new window or pick another "engineer".

So far, so much hype and talking about "AI programmers".

So, you haven't tried it yet?

The funny part is that even if it was a bad 'coder' the experience would be better than 85% of the ones I paid for.

:1orglaugh:1orglaugh:1orglaugh

In 2017 I hired someone for a mainstream simple coding project and after I sent them the $300 or whatever it was, they disappeared for 3 months and then in the middle of the night I got a video sent to my phone of the guy partying somewhere in eastern Europe and he sent me a video apologizing that he wasn't finished my code yet, and like 2 days later he emailed me the code that I had forgotten about by that point. The code didn't work.

I think the main reason it happens is because of increased system usage, I've found it happens more between 6-11pm EST, I assume because more people are using AI at those times (once they're home from work) with the new usage surge causing server overloads of the infrastructure.

Early hours of the morning, its rare I'll run into those types of issue.