7
votes
Ask Tildes: What kind of non-mainstream AI are you using regularly?
Hey Tildes, I was researching about AI usage for a project and was curious -- do you folks use some non-mainstream AI (like outside of ChatGPT/Microsoft Copilot/Apple Intelligence), like perhaps a custom Chrome extensions or maybe a desktop application?
I wanted to understand what use cases for AI exist that the big companies are lacking in right now
I build the things (assuming we’re using the same definition of “AI ≈ big neural nets, probably transformers”), and a lot of the more compelling use cases I see are comparatively dull task-specific models for data handling: medical imaging, financial projections, internal document search and summarisation, that kind of thing.
The gains there compared to state of the art even two years ago are vast; if you’re interested in specific examples you might find it helpful to take a look at the competitions companies are sponsoring on Kaggle, that’ll give you an idea of what they’re publicly paying to solve right now.
The other major example that comes to mind is things like Photoshop, or your phone’s camera software (the actual internal postprocessing, even if you don’t edit the pictures): they’re mainstream tools, but not what most people think of as AI per se - the modern versions just have their functionality heavily augmented by neural net models.
The subtext to what I’m saying there is that the tech itself is often incredibly capable, but most of the time when a product is shouting “AI” as the headline it’s a sign that they don’t know what else to shout about in terms of user benefits. A lot of the actually really useful models for voice recognition, image search, OCR, machine translation, etc. are things we kind of take for granted already - because they’re not products per se, they’re a component of the wider ecosystem that’s there to solve a problem. It can control the home automation while your hands are covered in bread dough, or to find that cute picture you took three months ago, or give you an idea of what that menu says while you’re in Shanghai, or any number of other similar things.
I have a mobile app that uses a combination of internal and external AI (LLM and other) models to teach languages. Besides the commercial APIs, I use fine tunes of Llama and other foundation models that are customized for my purposes.