Apple Can Do AI Fast … for Text That Is
October 22, 2025
Wasn’t Apple supposed to infuse Siri with Apple Intelligence? Yeah, well, Apple has been working on smart software. Unlike the Google and Samsung, Apple is still working out some kinks in [a] its leadership, [b] innovation flow, [c] productization, and [d] double talk.
Nevertheless, I learned by reading “Apple’s New Language Model Can Write Long Texts Incredibly Fast.” That’s excellent. The cited source reports:
In the study, the researchers demonstrate that FS-DFM was able to write full-length passages with just eight quick refinement rounds, matching the quality of diffusion models that required over a thousand steps to achieve a similar result. To achieve that, the researchers take an interesting three-step approach: first, the model is trained to handle different budgets of refinement iterations. Then, they use a guiding “teacher” model to help it make larger, more accurate updates at each iteration without “overshooting” the intended text. And finally, they tweak how each iteration works so the model can reach the final result in fewer, steadier steps.
And if you want proof, just navigate to the archive of research and marketing documents. You can access for free the research document titled “FS-DFM: Fast and Accurate Long Text Generation with Few-Step Diffusion Language Models.” The write up contains equations and helpful illustrations like this one:

The research paper is in line with other “be more efficient”-type efforts. At some point, companies in the LLM game will run out of money, power, or improvements. Efforts like Apple’s are helpful. However, like its debunking of smart software, Apple is lagging in the AI game.
Net net: Like orange iPhones and branding plays like Apple TV, a bit more in the delivery of products might be helpful. Apple did produce a gold thing-a-ma-bob for a world leader. It also reorganizes. Progress of a sort I surmise.
Stephen E Arnold, October 21, 2025
Comments
Got something to say?