Tangentially about web-based generative AIs as products, I sense an opportunity to innovate around the pain of their non-determinism, perhaps an enhancement to Web Browser "History" functionality.
Q: Is this general usability issue being discussed more, due to the growth of new product teams working with generative AIs?
Some discussion: I am drawn to wonder if there is an innate value to include permalinks (or semi-permanent) ... to save each query's results. Or perhaps do something better which I've not thought of yet.
The issue in more detail, (I'm naive on the subject so forgive the redundancy:) Results aren't highly deterministic - randomness and date-sensitive models seem to generate, probabilistically, different results for the same query. Especially for Generative AI products that exist as webapps, where hitting 'reload' could provide inferior results and simultaneously eliminates superior prior results.
Non-determinism is a 'surprise' quality which can enhance and degrade the value of a generative AI tool, in my experience when using this category of tools. It's fun (endorphin rush) to spin the Roulette wheel and query again. And it's painful when the prior results ranked More Useful in one's mind.
Ideally for usability's sake, these products would include some affordance to the user, to manage that endorphin / pain cycle.
[Note, A complication, here with PHIND being the context. PHIND's HTML renderer itself is being iterated on .. this is a product-launch after all. So, results also look different not for data/algorithm reasons.]
This could be realized as a web-browser feature - an enhanced History. The current "History" model of browsers is a trivial response to the REQUEST/RESPONSE HTTP mode. The HTTP spec (last time i read it was in 2005) leaned towards everything-is-deterministic except for time.
Example: I input this query https://phind.com/search?q=linq+query+to+filter+every+other+... and it listed one code snippet. I re-queried and received a different snippet. Honestly, I like both results. As of this moment I'm writing, I can only generate one of the two results.
Q: Is this general usability issue being discussed more, due to the growth of new product teams working with generative AIs?
Some discussion: I am drawn to wonder if there is an innate value to include permalinks (or semi-permanent) ... to save each query's results. Or perhaps do something better which I've not thought of yet.
The issue in more detail, (I'm naive on the subject so forgive the redundancy:) Results aren't highly deterministic - randomness and date-sensitive models seem to generate, probabilistically, different results for the same query. Especially for Generative AI products that exist as webapps, where hitting 'reload' could provide inferior results and simultaneously eliminates superior prior results.
Non-determinism is a 'surprise' quality which can enhance and degrade the value of a generative AI tool, in my experience when using this category of tools. It's fun (endorphin rush) to spin the Roulette wheel and query again. And it's painful when the prior results ranked More Useful in one's mind.
Ideally for usability's sake, these products would include some affordance to the user, to manage that endorphin / pain cycle.
[Note, A complication, here with PHIND being the context. PHIND's HTML renderer itself is being iterated on .. this is a product-launch after all. So, results also look different not for data/algorithm reasons.]
This could be realized as a web-browser feature - an enhanced History. The current "History" model of browsers is a trivial response to the REQUEST/RESPONSE HTTP mode. The HTTP spec (last time i read it was in 2005) leaned towards everything-is-deterministic except for time.
Example: I input this query https://phind.com/search?q=linq+query+to+filter+every+other+... and it listed one code snippet. I re-queried and received a different snippet. Honestly, I like both results. As of this moment I'm writing, I can only generate one of the two results.