vladimircaribb - stock.adobe.com
Apple AI paper likely self-serving but has some merit
Although the iPhone maker appears to be trying to draw attention away from the fact that it’s behind in LLMs, the point it makes might still be valid.
Apple's paper questioning the accuracy of reasoning models is getting mixed reviews as the AI market continues investing heavily in these models.
The consumer tech giant Apple released a new research paper on June 12 claiming that while many reasoning models show improved performance on benchmarks, the AI market is not considering some gaps.
According to the paper's authors, while the models show initial success with simple problems, their level of reasoning declines after a while. Apple also claims that what seems to be thinking is just pattern matching, which creates an illusion of thinking.
Apple released the document the same week it held its Worldwide Developers Conference, where it introduced the Foundation Models Framework for Developers, a new API that allows third-party developers to integrate Apple's on-device models into their apps and build on Apple Intelligence. Despite this move, Apple continues to face criticism for how its AI models underperform compared to AI models from OpenAI, Meta and others.
Apple’s conflicting motivations
Therefore, some see the research paper as self-serving and a way for Apple to cover up its being largely behind in generative AI (GenAI). However, other AI experts say two things can be true simultaneously.
"The motivation behind having this discussion is purely in Apple's interest," said Futurum Group's David Nicholson. "Shouldn't they be researching how to make Siri smarter? They should focus on that instead of throwing stones at everybody else."
On the other hand, Nicholson continued, "There is some merit to the discussion that Apple is opening up."
Investment in AI continues to grow. Meta, Amazon, Alphabet (Google's parent company) and Microsoft are expected to invest billions of dollars in AI technology in 2025. However, the ROI for the widely popular technology has not yet been proven.
"The thing that's driving the investment is the hype around the magic, and it's easy for us as humans to be fooled by a large language model because when these large language models statistically string words together, it feels like you're talking to another human being," Nicholson said. "It's thinking, but it isn't."
"It's a bit of Apple buying themselves time in the PR sphere," Nicholson continued.
Heeding the warning
However, Michael G. Bennett, associate vice chancellor for data science and AI strategy at the University of Illinois Chicago, said it's understandable that Apple's research received some backlash because it questioned some of the widely held beliefs that many hold about reasoning models.
However, the AI market should not dismiss Apple's research paper.
"I don't embrace the paper wholeheartedly or uncritically, but I do think that when a major force in the industry lays down an article like this ... it's still worthwhile to remember what they're saying," Bennett said.
He added that not paying attention could have serious implications, especially since the promise of reasoning models is that they can help solve problems that human minds might not be able to.
"The ambition and the hope are warranted, but if these systems can't perform even on moderately complex tasks, then it's better to understand it relatively early on," Bennett added.
Meanwhile, on June 16, Chinese vendor Alibaba revealed that its new Qwen 3 models will run on Apple devices, including iPhone, iPad, MacBook and Mac.
Esther Shittu is an Informa TechTarget news writer and podcast host covering AI software and systems.
