Playing (and working) with GenAI, I can see things it does really well:
Simulate an artificial intelligent person/human/thing
Understand grammar and words, in multiple languages
Summarizing or rewriting text
Answer questions and extend text
Adjust the response depending on the persona it’s supposed to be
Search, especially with support of RAG
However it also fails in other areas quite spectacularly:
Anything which requires logic
Anything which requires actual understanding
Exhibit A: The classical man with a wolf, a goat and a cabbage wants to cross a river. But with a twist: no goat.
The system recognized the pattern of the original story and it just adds a goat. Let’s ask where the goat comes from?
That really makes no sense.
And before someone thinks that this is a problem with Gemini, here’s Claude 3 Sonnet’s reply to the same question:
It magically added a goat too. And every animal now like cabbage.
Summary
When every response looks good, but some are plain wrong, how can you trust any response at all?
My personal solution:
Use GenAI for NLP and to recall things it has seen before. It’s very good at this.
Don’t use GenAI for anything which involves logic and understanding.
Because responses are not consistent and predictable, testing GenAI is very hard making it very difficult to use in scenarios where wrong answers can cause damage of any type.
BTW for testing I can highly recommend PartyRock as it allows you to test various LLMs and create text, chats and images with a very simple GUI.
{Categories} _Category: Takes{/Categories}
{URL}https://hkubota.wordpress.com/2024/07/12/genai-does-not-think-nor-understand/{/URL}
{Author}unknown{/Author}
{Image}https://hkubota.wordpress.com/wp-content/uploads/2024/07/image-5.png?w=640{/Image}
{Keywords}{/Keywords}
{Source}POV{/Source}
{Thumb}{/Thumb}