Barring unforeseen developments —which no one can bar or even assess the likelihood of, in general— we may now know the approximate scope of what LLMs can and cannot do. As such David and I discuss how he’s been using them and how to think about what they’re best for from the technical designer’s point of view. This point of view, perhaps needless to say, may soon be mandatory for all software designers, and at a minimum is worth considering deeply, as the trade-offs for “becoming technical” are one of the things LLMs have indeed changed radically. It may have been premature to do so when wags expected “superintelligence” any day, but if LLMs will indeed top-out at what they do now, we can start to evaluate how we want to respond as professionals.
That’s not to say more and exciting improvements aren’t coming, of course. These might included
expanded context windows; as David notes, LLMs fare much better working from data supplied for a given interaction than they do when asked to access knowledge inside themselves, so larger context windows are very meaningful (for e.g. personal data, history, codebases, etc.);
better multimodality; it feels as though we’re relatively early on the multimodal frontier, and designers may both benefit from improved multimodality (resulting in LLM capacities in the visual space, for example) and from seeing how UIs for LLMs must evolve to accommodate these capabilities;
and perhaps above all, integration into software environments, particularly operating systems. If, as discussed, LLMs are destined to commodify, “where they are” in the software stack is an important question. My suspicion that they’re destined to be device-level or OS-level (or even “ecosystem”-level) may or may not be justified, but if they are, it will change what they can do and how designers will use them.
But it seems unlikely, at the moment, that future LLMs will be so dramatically different or more-capable than they are today that designers should wait to see how they develop before making use of them. For myself, I had lazily hoped that I’d never need to configure a development environment to avail myself of their functionality, and that hope is fading for me; it may be time to do something I should have done a decade or more ago and get real!
As we mention, we may do another episode in which David attempts to teach me to use LLMs to prototype and accomplish other tasks. If you have any questions about LLMs for designers, or anything related, let us know, and thanks for subscribing!
