GPT-4.5

GPT-4.5 has been underwhelming to the say the least.

I have seen people get some interesting responses from it on Twitter/X, but I haven’t got any on my own account. It doesn’t appear any more empathic or qualitatively human like in my conversations with it. It’s possible I’m not using it correctly.

The jump from GPT-3 to GPT-3.5 was quite massively felt, the jump to GPT-4 was also qualitatively felt on the margins but in very meaningful ways.

I don’t post to youtube really anymore, but I’ve been meaning to put together some kind of supercut of all the false promises and hype coming from AI labs over the years. Basically, since ChatGPT was released in 2022 and was so viral, the false promises have been non stop from all the major AI labs.

Apparently, by sometime mid summer this year, we are supposed to see autonomous agents capable of doing most economically valuable jobs - or at least that’s what I recall hearing publicly from OpenAI and Anthropic leaders at the beginning of this year. The AI industry and its followers have the memory of a goldfish. I guess I remain pretty skeptical about the whole thing.

The language model space used to be research driven and more substantive, breakthroughs were severely understated, and there was a lot less hyperbole coming from companies like OpenAI.

Call me old fashioned, but the proof for AI products has always been in the pudding. If GPT-4.5 isn’t that great, that tells me everything I need to know about what is going on at OpenAI … whether we’re talking about their leadership, product development cycle, or research cadence. It’s the only thing I really look at and to be honest, I think you should too. Everything else is just noise.