OpenAI released their ChatGPT. Damn, it is good. This might be GPT4. Starting a 🧵 with observations...
First, it has a memory, something a lot of folks have been working on.
OpenAI has created a great UX for collecting a lot of great human feedback, a necessary ingredient for continual improvement. They are winning the data flywheel game.
It's responses are fast and high quality. The conversation feels fluid.
More great poetry.
It's smarter than text-davinci-003 on some common sense questions.
It claims to not be sentient. But @cajundiscordian might say otherwise.
It still can't add big numbers (=914229251437). OK humans, we're safe for now.
It's got a good poker face.
It seems to naturally incorporate chain-of-thought. text-davinci-003 got it wrong.
It's memory works and understand the sequential nature of our conversation.
It's still not good at algebra (answer x=-1 and x=2)
It seems to know when it doesn't know something but also seems to make some stuff up. I don't think @karpathy was ever an Assistant Professor. He wasn't a Research Scientist at OpenAI as of the training data for ChatGPT.
Overall, I'm really impressed by ChatGPT. Chat interfaces are going to get a big upgrade real soon.
BPE or whatever OpenAI is using currently still prevents spelling things backwards (as in 003).
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.