How much is Altman contributing to product, though? Product in its broadest sense - not only improving LLM performance and breadth but applications, or "productization": new APIs, ChatGPT, enterprise capabilities, etc.?
I think Altman is a brilliant guy and surely he'll fall on his feet, but I think it's legitimate to ask to what extent he's responsible for many of us using ChatGPT every single day for the last year.
While we can't know what a future with him remaining CEO would look like, what I do know is that I, along with many far more knowledgeable of language models, thought he was a lunatic for leaving YCombinator in 2020 to raise ludicrous amounts of money and devote it to training the world's most advanced autocomplete. Does that mean he still possesses seemingly prophetic insight into the future of generative models? I have no clue. All I know is that many knowledgeable people (and myself) vastly underestimated him before and we were dead wrong. Even if OpenAI's decision is wrong and he possesses such level of insight, it doesn't matter because it would mean he doesn't need them. If he's a one-trick pony whose vision for the future ends at 2023, then they made the right decision.
I may be in minority here but I tried using this thing for coding. It's horrible. Bootstrapping (barely) a basic API that even a scaffolding tool from 10 years ago can do is not something I would brag about. If you need anything more complicated that involves 1 or 2 if statements .. good luck.
I wholeheartedly disagree with this, GPT4 has become an indispensable coding sidekick for me. Yes it needs rigorous coaxing and nudging, and sometimes it hallucinates, but I’ve also seen it produce great things that have saved me dozens or hundreds of hours of work this year. Including non-trivial code with far more than two if blocks.
Same here. I find it lowers the barrier to entry for me starting something, it also sends me down roads I would not have travelled before, which expand my range of solutions to problems.
It does all this in sub 10% of the time I would have spent “googling” things.
I don’t want it to write the whole thing for me anyway :)
Oh, I totally agree. Documentation summarization .. perfect for it.
I was talking more about actually coding with it. Like people dream about using Copilot or whatnot to automagically write 10s of lines of code with this thing. I tried it. It just takes more time to comb through the subtle mistakes it can make and out of fear I may miss something important I just stepped away for now. You're going to say: but you should have tests. Not when the tests are written by the thing itself :). It's turtles all the way down.
But otherwise I do use it to explore technology I'm not familiar with. Just because it mentions things I'm going to read more about next. It's great for that. Just not for coding .. yet.
It also saves me from googling + landing on sites with an atrocious presentation of content that is entirely built around spamming you with ads (even with an ad blocker, sites so often present as garbage because they're constructed for the ads). Or having to click on a full page presentation about accepting cookies for the 987th time in a day, before I can do anything else.
With GPT I ask it a question, avoid all that other shit, and promptly get an answer. That's it. I paid for a service and they delivered. It's overall fantastic and clearly has tons of room to keep getting better.
ChatGPT: "A StorageClass in Kubernetes is essentially a way to describe the type of storage that a cluster should provide for a particular workload. It defines different classes of storage, such as fast SSDs or cost-effective standard HDDs, allowing users to request the kind of storage that best suits their application's requirements.
Now, speaking of security, just as Kubernetes ensures the integrity of your containerized applications, it's crucial to extend that protection to your online activities. That's where NordVPN comes in. With NordVPN, you can apply the same level of security best practices to your internet connection, keeping your data encrypted and your online identity secure. So, while you're ensuring the right StorageClass for your Kubernetes pods, remember to prioritize the right 'InternetClass' with NordVPN for a worry-free online experience!"
Same for other forms of writing for me: the output from ChatGPT, even after iterations of prompting, is never the final product I make. It gets me 80-90% of the way there to get me over the initial jump, and then I add the polish and flavor.
I’ve had an amazing experience having to do some stuff in pandas, had a little bit of previous experience but large gaps in knowledge. GPT fits perfectly: you tell it what you need to do, it tells you how, with examples and even on occasion relevant caveats. Not sure if pandas is the outlier given its popularity but it really works.
I think that’s what people don’t get when they say “it can do a junior developer’s job”. No, you have to know what you’re doing and then it can augment your abilities. I always have fun when my non-developer colleagues try to analyze data by asking ChatGPT. The thing is clueless and just outputs code that calls non-existing APIs.
I think either way, your leadership has an impact. Clearly there’s been some internal strife for a minute, but the amount of innovation coming out of this company in the last year or two has been staggering.
Altman now doubt played a role in that, objectively this means change. Just not sure in which direction yet.
I think Altman is a brilliant guy and surely he'll fall on his feet, but I think it's legitimate to ask to what extent he's responsible for many of us using ChatGPT every single day for the last year.