Generative AI, at the end of the day, simply predicts the next most obvious thing (whether text/image/soundā¦) and the quality of its prediction depends on the data it is trained upon.
So what may seem to you as magic, is AI simply trying to predict the next most logical element in the sequence. So, when you ask GenAI tool a question, it is just trying to predict the most statistically relevant answer to that question, and that is why many of GenAI tools are incredibly bad at mathematics.
Instead of solving the math problem, these tools are just trying to predict the next logical thing in the sequence (from all the training data set they had):
FREE Database of 9,000+ AI Toolkit and 4,418+ Business Resources that will save you hours of work
Elevate your workday with this carefully curated database that 32,000+ professionals trust to help them rapidly innovate at their business and personal life.
This Database is packed with new AI Tools, business ideas and money-making opportunities that will effortlessly speed up your daily business workflows.
Click here to get Instant Access (Limited access: FREE for First 100 people only)
To understand GenAI tool = understand how itās trained
Even if it is not a math problem, GenAI tools can end up giving you āstatistically relevant from their training datasetā answers and not the ārightā or āaccurateā answers. That is why you mostly should be careful relying on it for facts.
Hence very generically speaking, it is important to think of GenAI tool responses to be āmost populist onesā in an unfiltered dataset training and maybe less populist + more accurate in a filtered dataset.
The GenAI tool race is the race of training and adoption (# of users) essentially. Example OpenAI is chasing high risk use cases at the cost of accuracy whereas Google is chasing high accuracy at the risk of delaying launches. I wrote about it Google vs. OpenAI here.
Datasets are not only relevant for text-based GenAI tools, but also for image-based ones. For example, the most famous Image Generative AI tools are trained on many different data sets:
And that is the reason you see wildly different outcomes for the same prompt: A dog walking in a park with a cloudy sky.
So, unless you live under a cave and want to leave your life to chance, it is always smart to know when to use which Generative AI tool - whether text based or image.
For Text Generative AI tools like ChatGPT, Bard, Bing and more - there are some thumb rules for when to use which. ChatGPT is not the holy grail answer to everything.
In next part of this newsletter, let us break down:
ā When to use which text-based GenAI (Bard, ChatGPT, Bing, Claude & more)
ā What to do for specific use-cases? (Like making a smartphone buying decision)
ā Which use-cases of GenAI you should try in your organization? (for different functions and industry sectors)
ā GenAI and ChatGPT use-cases for Product Managers