If you get the joke in the title, you probably saw the movie Airplane. In a classic scene near the end of the film, the plane (a jet full of passengers) finally lands at the airport, after multiple misadventures. But it can’t slow down, and hurtles out of control from one gate to the next. The terrified emergency substitute pilot desperately slams on the brakes, while the airport announcer says “now arriving Gate 8, Gate 9, 10 … Gate 13, Gate 14, Gate 15. ...Gate 23, …. Gate 25”, Loved ones waiting for the plane to land run across the airport as all this unfolds. It was truly great scene, mixing humor, drama, and terror — in a scene that I’ve remembered for decades.
In case you haven’t noticed, GPT-5 has come to seem to be a little like that. People keep on trying project its arrival, and its arrival keeps on slippin’ into the future.
§
Nobody to my knowledge has kept systematic track of the predictions, but I took a quick and somewhat random look at X and had no trouble finding many predictions, going back to 2023, almost always optimistic. The results of my informal and unscientific look are presented below, in the order in which they were made. A lot of them (though not the first) got tons of views. One, perhaps more of a hope than a prediction, from an NVidia employee, got almost a quarter million views.
All seem almost sad, in hindsight. What stands out the most, maybe, is the confidence with which a lot of them were presented:
On X, there’s even a whole account called countdown to GPT [screenshot from June 6]:
Alas, The Super Bowl has come and gone; so has April. Today is June 20 and I still don’t see squat. It would now appear that Business Insider’s sources were confused, or overstating what they knew. Sam just a few weeks ago officially announced that they had only just started training GPT-5. (Probably for the third or fourth time, if my hunch is correct, with GPT 4-Turbo and 4o perhaps being failed attempts that didn’t seem worthy of the name, with another earlier attempt halting before Sam’s time speaking to the Senate in May 2023.)
§
I’m clearly not the only one to have taken note.
By now there’s actually a new meme in town. This one’s got even more views:
§
As of today, I am more confident than ever that GPT-5 won’t land this year.
Why? OpenAI’s CTO Mira Murati all but said so.
In an interview with her alma Mater Dartmouth, Murati promised we’d someday see “PhD -level” models, the next big advance over today’s models, as she tells it … but not for another 18 months.
Now arriving Gate 2024, Gate 2025, … Gate 2026.
Gary Marcus is still betting that GPT-5 will continue to hallucinate and make a bunch of wacky errors, whenever it finally drops.
Whatever happened to Q*, which was hypothesized as the breakthrough AGI needed? Did anyone ever figure out what Q* actually was and whether it would figure prominently into GPT-5 and beyond?
Anyhow, the more I use GenAI, the more often I think to myself that "this stuff really isn't all that good." I'm currently paying the $20 a month for GPT-4 and Gemini, but I'm not sure for how much longer.
It's also a bit disconcerting how some GenAI pushers respond to my dilemma: They insist that I know nothing about prompting and that I need to find the right use cases. Well, I know enough about prompting to realize that I can often save time and effort by just doing the task myself instead of trial and error to find the right prompt. As for use cases, if I have to work on finding them, as opposed to handling the ones I have, GenAI would distinctly qualify as a solution looking for a problem.
"I think you wanna know what our chances are. The lives of everyone depends on someone back there who can not only fly this plane but who didn't have fish for dinner." You know it Gary. Everyone at OpenAI had fish for dinner... good thing you had the lasagna.