A 30-Page Explanation from Tim Apple: AI Is Not Too Good
June 9, 2025
I suppose I should use smart software. But, no, I prefer the inept, flawed, humanoid way. Go figure. Then say to yourself, “He’s a dinobaby.”
Gary Marcus, like other experts, are putting Apple into an old-fashioned peeler. You can get his insights in “A Knock Out Blow for LLMs.” I have a different angle on the Apple LLM explainer. Here we go:
Many years ago I had some minor role to play in the commercial online database sector. One of our products seemed to be reasonably good at summarizing business and technical journal articles, academic flights of fancy, and some just straight out crazy write ups from Harvard Business Review-type publications.
I read a 30-page “research” paper authored by what appear to be some of the “aw, shucks” folks at Apple. The write up is located on Apple’s content delivery network, of course. No run-of-the-mill service is up to those high Apple standards of Tim and his orchard keepers. The paper is authored by Parshin Shojaee (who is identified as an intern who made an equal contribution to the write up), Imam Mirzadeh (Apple), Keivan Alizadeh (Apple), Maxwell Horton (Apple), Samy Bengio (Apple), and Mehrdad Farajtabar (Apple). Okay, this seems to be a very academic line up with an intern who was doing “equal contribution” along with the high-powered horticulturists laboring on the write up.
The title is interesting: “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity.” In a nutshell, the paper tries to make clear that current large language models deliver inconsistent results and cannot reason in a reliable manner. When I read this title, my mind conjured up an image of AI products and services delivering on-point outputs to iPhone users. That was the “illusion” of a large, ageing company trying to keep pace with technology and applications from its competitors, the upstarts, and the nation-states doing interesting things with the admittedly-flawed large language models. But those outside the Apple orchard have been delivering something.
My reaction to this document and its easy-to-read pastel charts like this one from page 30:
One of my addled professors told me, “Also, end on a strong point. Be clear, concise, and issue a call to action.” Apple obviously believes that these charts deliver exactly what my professor told me.
I interpreted the paper differently; to wit:
- Apple announced “Apple intelligence” and failed to ship for what a year or more had been previously announced
- Siri still sucks from my point of view
- Apple reorganized its smart software team in a significant way. Why? See items 1 and 2.
- Apple runs the risk of having its many iPhone users just skip “Apple intelligence” and maybe not upgrade due to the dalliance with China, the tariff issue, and the reality of assuming that what worked in the past will be just super duper in the future.
Sorry, gardeners. A 30-page paper is not going to change reality. Apple is a big outfit. It seems to be struggling. No Apple car. An increasingly wonky browser. An approach to “settings” almost as bad as Microsoft’s. And much, much more. Coming soon will be a new iOS numbering system and more!
That’s what happens when interns contribute as much as full-time equivalents and employees. The result is a paper. Okay, good enough.
But, sorry, Tim Apple: Papers, pastel charts, and complaining about smart software will not change a failure to match marketing with what users can access.
Stephen E Arnold, June 9, 2025
Comments
2 Responses to “A 30-Page Explanation from Tim Apple: AI Is Not Too Good”
Got something to say?
[…] campo de la IA (y, previsiblemente, su ausencia de novedades en este campo en el propio WWDC). En su análisis, señala que mientras otras empresas ya ofrecen asistentes con razonamiento avanzado, Apple aún […]
[…] el campo de la IA (y, previsiblemente, su partida de novedades en este campo en el propio WWDC). En su análisis, señala que mientras otras empresas ya ofrecen asistentes con razonamiento progresista, Apple aún […]