Condensely
Had a nice little 4-day streak going there, back to 0, no matter.
One of the most notable projects I am currently pursuing is Condensely.ai, an AI-powered learning tool aimed at neurodiverse learners. My entire future hinges on it - it’s also my BSc dissertation project.
At the stage of ideation, ChatGPT (or any other mainstream provider for that matter) did not yet have a search functionality and so generated content would consist of a heavy helping of vague, circumventing, ridiculously incorrect sludge, blithely presented as fact.
Like an overly enthusiastic waiter delivering a kick in the balls with an ear-to-ear smile, then looking to you on the floor asking “Anything else at all I can do for you, Sir?” - these responses made my (and every human being with a basic knowledge of just about anything at all) blood boil. How could it continuously spew out complete nonsense so blatantly?!
This issue persists even with search, though less often with the reasoning models, but alas, I digress.
Condensely
The premise was simple - build a platform that leverages the good bits of ChatGPT, but search-enable, and add features for neurodiverse learners.
Simple enough, right? Turns out, there is a rather loose definition of “Neurodiverse”, or indeed “features for neurodiverse learners”. With enough literature review though, the objectives and requirements became clearer. I was to build, essentially, an LLM model wrapper, with added features and utilities - a layer of abstraction for the user’s convenience. Like how the first GUI Windows 1.0 ran on top of DOS.
An idea with merit, in theory, and it has been continuously validated by the swarm of new search-enabled LLMs, but still I feel there’s something more.
With search, though the rate of hallucination is vastly reduced, there are still issues with stale content (due to web browser caching, which is outside my direct control). This issue in itself is relatively minor, and most of what I set out to achieve has been - the real rub is that I feel there’s something beyond this functionality. Something not yet invented, not yet conceptualised by me, but something I am sure will drastically change LLM capabilities, just as web search has.
What, then? I don’t yet know.