Leaving Open AI, like the other 600+ dedicated developers have, would for me be like the worst kind of “breakup” because in many ways and in truth, Ive come to love Open AI as a friend and partner.
Any chance the upgrades will be open again soon?
Deeply appreciate this message. Rooting for the OpenAI team, let’s keep building.
And hope that all efforts, irrespective of individuals and organisations are aimed towards a common goal. Betterment of humanity, and more tools for developers
I absolutely love this guy. I hope I live long enough to see the days when he becomes the POTUS.
So much. The nature of software is fundamentally changing. The tools we make can’t exist without a foundation model. You can’t compile an exe and away you go, totally different ball of wax. No api? No tool. Super fascinating.
And when software can write itself, which it is starting to get good at (I cannot belive I am writing this) then what is software? Anyone could ask for a character sheet generator and get exactly the custom one they want with the features they want. At the rate things are going who knows what things will look like in 6-12 months. It has only been 9 months since gpt4 was released!
This weekend have been a roller coaster of ups and downs. We’ve based our entire existence on top of your stuff. I feel extremely bad about the whole thing, but if it goes down, please take it down controlled, such that we can swap out chat.openai.com with chat.microsoft.com or something, and have an in place replacement …
Regardless of what happens, you did something incredibly unique here, and it’ll echo in history books for centuries to come …
I think it just changes scope.
Where for instance, a game studio would take a long time to produce a triple-A game, it may take months now or less, but if the development cycle for a triple a game is 5 years, and they KEEP it 5 years once AI takes off - I mean you’re talking about your favorite game but with a 200 hour campaign and its done easily.
So I’m thinking like yes software might get cheaper, but therefore what you build may have to be larger.
It also gives smaller studios the ability to compete.
Think bigger. Logically if you think through the implications it’s wild. You don’t need software applications and packages in the same way. Already I just get gpt4 to write custom code and run it to extend functionality. “Hey, here’s the task, figure out what we need, write up the scripts, then save them to your local environment for us to use.” Each time I run the thing it’s slightly different and tweaked for the task. In the near term I think you have it right, but I do think things will change in much more fundamental ways over time.
Dilly Dilly! And do A bright future under Open AI!
it is NOT that simple with gpt-4 dont lie to yourself, it is much more effort than just that currently.
For real, the amount of insane code GPT4 gives me that I have to sanitize is unreal. I can’t imagine trusting the code it generates out of the box. Still, it’s early days, I used to be sure AI would come for the coders last, but there’s no denying it’s starting to look like we’ll be obsolete sooner rather than later…
Once they figure out the limited context window, of course, until then we’re safe haha.
Please don’t call me a liar. If you know what is possible and approach it correctly the capabilities right now are staggering. I used to copy and paste error messages back and forth etc but that has pretty much gone away. The function calling updates were huge, everything got so much tighter. I had implemented function calling on my side before theirs was released and the change was night and day. The massive improvements in the quality and consistency of the Gpt4 outputs and calling things with correct and valid arguments impressed me most. The most recent update is wild. As of a few days ago if you approach things the right way you can get it to write gorgeous code, check everything over, then save out clean functioning files and pop you a link.
And yes, I have assistants that customise themselves and it works great.
This is actually what I did to start learning how to code 3 months ago. It’s very slow and hit or miss, but when you’re creating something new (in my case a dashboard) there’s really not too much risk with trusting it and testing it.
Reminds of my experiences in june/july, feels like 2 years ago. Took me ages to get system messages etc dialled in. If you haven’t tried yet just using an assistant/gpt(way cheaper) with the code interpreter, even if you don’t use it, writes much better code. Feels like it’s ‘primed’ to write good code.
It’s not primed to write good code, it’s primed to write what it believes is good code based on what its knowledge base contains.
Don’t get me wrong, I use it on a daily basis to help me out with mediocre tasks, or to sanity check when I’m missing something, but would I trust it to actually write production code for me? Not a chance.
Out of curiosity, what language are you guys asking it to write for you?
Everything I’ve been doing is in Python
No doubt, we are a long way from production, but it can write code. At this rate it won’t be long.
I wouldn’t ask it to write photoshop, but in it’s sweet spot it is very capable.
For me it’s mostly python, c#, hlsl, a bit of assembly and so on. I was midway through learning shader programming when this all hit. I finished up in a fraction of the time I had expected and with much more solid and well-rounded knowledge once I realised that gpt4 could do this stuff.
Yeah as a learning tool it’s beyond useful, I’ve been having it teach me React for a project at work, as I’m mostly a backend dev focused on .Net, and it’s real good at explaining individual parts of code. I just wouldn’t trust it to actually understand the requirements or context surrounding modifications to an existing codebase, and the ramifications of its suggestions, without some serious prompt doctoring, and even then it tends to go off the rails and revert back to cookie cutter code you could read on a tutorial site.
This is a fascinating subject, it will access different knowledge depending on how the system message is. ‘You have the knowledge and abilities of the best programmers in the world and write beautiful, flawless, concise code.’ works good for me. There’s lots of information about this out there if you are curious.
Yeah the inconsistency is what is holding everything back. I just had it say ‘I can’t do that’ three times in a row and then happily do the thing it said it couldn’t on the 4th.
And I know exactly what you are talking about, going from writing good code to some boilerplate uselessness. Super frustrating, but it’s all getting so much better so fast. Totally guessing about this particular thing, but it feels like a guardrail thing to me, like it is getting kicked out of it’s normal response and ‘blah blah blah’ pours out instead.
For me it’s not about developing production ready tools right now, it’s about understanding how to work with the technology so that when it is actually useable I am also ready. The tools I’m working on will take 3-6 months to build, for now I’m just getting proof-of-concept up and running (even if super flaky) to get a feel for how to approach things. When things get solid I’ll dive in and build the tools from the ground up the right way. In the meantime it’s making my life easier and reducing development time across the board.