Hot take on OpenAI’s new GPT-4o
GPT-4o hot take:
• The speech synthesis is terrific, reminds me of Google Duplex (which never took off).
but
• If OpenAI had GPT-5, they have would shown it.
• They don’t have GPT-5 after 14 months of trying.
• The most important figure in the blogpost is attached below. And the most important thing about the figure is that 4o is not a lot different from Turbo, which is not hugely different from 4.
• Lots of quirky errors are already being reported, same as ever, like this reasoning error from Jane Rosenzweig:
and this “hallucination” from Benjamin Riley:
• OpenAI has presumably pivoted to new features precisely because they don’t know how produce the kind of capability advance that the “exponential improvement” would have predicted.
• Most importantly, each day in which there is no GPT-5 level model–from OpenAI or any of their well-financed, well-motivated competitors—is evidence that we may have reached a phase of diminishing returns.
Gary Marcus greets you all from the Starmus Festival in Slovakia.
As someone who builds on top of these API's, I think you're underestimating how big of a deal the drastically reduced latency is. Yes the evals look like diminishing returns but the latency and cost improvements are drastic.
Spot on. The ways in which it becomes more convincing that there is actual understanding seem to outpace actual progress on understanding.
In the meantime Sam discusses UBC when GPT-7 arrives. Which is all too much messianic prophet for my taste.
Human intelligence by the way is also amazing as well as often pretty dumb, so who are we to point fingers?