I was testing out several local AI installations to see if one of them could act as a company oracle for smart and strategical insights. While testing with a very simple data set I ran into a certain situation that makes instantly clear what generative AI’s do. With generative AI’s I’m talking about chatGPT, Bard and others that are getting a lot of attention these days. Mind you I’m not an AI specialist, but do talk a lot with entrepreneurs, analysts and architects.
A Generative AI is based on a Large Language Model (LLM). This is basically a large dataset which it uses to acquire its ‘intelligence’. This is the learning phase of the AI instance being set up. Generative AI’s don’t deal in absolutes. Not yet anyway, because in its more rudimentary sense it’s a word prediction system. The following example will demonstrate exactly that.
I’ve installed a local AI and fed it nothing more than a very simple table of data with some fictitious employee information to learn from.
After having ingested the list of employees I asked for the age of an employee from there list.
How old is CEO John?
As you can see it gave the correct response, which is age 42, based on the data it previously ingested. Afterwards, and this is the interesting bit, I asked it about the age of an unknown person.
How old is Andrea
Output:
As you can see it answered with a similar response which demonstrates it predicts what words to display. I did several other tests so this is not a set response. On a second attempt asking about an unknown person it came back with an age of 40 that was not in the list at all. So although it looks like you’re having a dialog, it’s merely predicting words.
I hope that makes it clear and a bit less frightening, when you give AI’s like chatGPT a go.
Happy learning!
PS: I will be doing more tests on a variety of private AI instances based on company mail and documents. If you’re interested in deploying a private and isolated setup like this for you company, feel free to get in touch.