AGI isn’t coming in 2025 — and GPT-5 may well not emerge this year, either.

Much was quietly revealed yesterday

Despite the daily shouts of “exponential progress” for nearly two years, GPT-5 hasn’t come. Kevin Scott, Microsoft’s CTO claimed last summer that it was imminent, and that it would be on a par with PhDs. So far it hasn’t actually arrived. Last March I predicted that it wouldn’t arrive in 2024.

It didn’t.

At a conference yesterday, someone with very good knowledge of OpenAI (Chatham House rules forbid me from naming the source, but allow me to repeat the remark)) said something fascinating, more for what was not said than what was said.

What was said is that we should expect to see GPT 4.5 soon.

What was not said by the well informed source was that we should expect GPT-5 anytime soon.

OpenAI has surely been racing toward GPT-5, for well over two years, presumably ever since August 2022 when they privately demoed GPT-4, hoping to beat competitors like Google from getting to similar technology before then. But it’s becoming increasingly clear that they are struggling in building a model worthy of the GPT-5 name.

We might see GPT-5 this year, but given bottlenecks on data and inherent limits in LLMs we may well not. All those fantasies a year ago about seeing GPT-5 drop after the 2024 Super Bowl are now far in the rear view mirror. The hypothesis I have been floating since 2022 that scaling data alone would not be enough for robust AI has become increasingly prevalent.

No release of 5 yet announced, nor obviously in sight. (o3 is a real advance in some domains beyond GPT-4, but it’s likely not the quantum leap forward people would expect of the name GPT-5, and at a rumored $1000/querry certainly not at the price point anyone might have hoped for).

§

While I was at the conference hearing implicitly that GPT-5 wasn’t coming, Elon Musk was giving an interview at CES. He made some of his usual overpromises (500,000 humanoid robots in three years, which seems to very unlikely) but also, importantly, and without clear acknowledgement, walked about his AGI predictions. Before, he had said that AI would exceed any human by the end of 2025, saying in April 2024, “I guess is that we'll have AI that is smarter than any one human probably around the end of next year?). His statement yesterday, was an unacknowledged retrenchment: “AI will do anything [cognitive] you want really within the few next years any cognitive task .”

Given that Musk is right in the middle of building what many think will be the largest language model to date, Grok 3, and that he has probably seen earlier returns on Grok 3, this is a notable retreat from his earlier predictions.

I will stand by my own predictions that we won’t see AGI in 2025, and I won’t be at all surprised if don’t even see anything worthy of the GPT-5 name (which I still expect not to be able to succeed at the kinds of criteria for AGI I have previously laid out here and here).

Gary Marcus wishes the media would hold those who make unrealistic promises to account. Because they really do, endless promises proliferate, and too many people, even policy makers, make key decisions based on unrealistic understandings of the field.