LLMs are not a panacea so stop hooking them up to your products and systems like they're a cheat code.
LLMs add a huge surface for unreliable, unsafe and untrustworthy experiences.
Can they be useful? Yes
Should they be the default? No
Are narrow models better for a lot of tasks? Yes
New little writings in thread (Actually written by me not Claude)
1/ In 1970 Marvin Minsky told Life magazine we'd have a machine with the intelligence of an average human within three to eight years.
His follow up: "If we're lucky, they might decide to keep us as pets."
55 years on, we're chasing the same goal.
@rorysutherland There is a point where models are all of a certain quality and it is going to be marginal gains… then the distinguishing factor is design.
I've started a little writing for my new newsletter, called https://t.co/D5Rhw7vRvP, focused on Human-Centered AI, which I've immersed myself in for the last 2 years and realised is the way if I/we don't want our kids to grow up in some strange Ridley Scott dystopia.
The first edition looks at some common mistakes I've seen in AI products and systems.
https://t.co/nW7buJYF7G
I see dozens of new AI products released every week and most of them seem to be answering the same question, "how can we automate what a human currently does?" This feels like a massive waste of a technology that could do good. value > efficiency.
Vector design tools like Figma, Sketch and Axure are great for communicating ideas but they're poor for designing products into production. UX'ers need better tools hence why @rivveo exists so product teams can work more efficiently and effectively.
https://t.co/4bmmkjShDt
Built this component with one prompt while building and testing @rivveo new AI Blocks... Still got lots of room for improvement... Design is about to change for the better. 😎