In today’s technologically driven world, AI is developing at a rapid pace. As we embrace the opportunity that this creates, it’s important to be aware of how these systems work.

Saurabh Shintre, an expert in artificial intelligence, machine learning, and cybersecurity joins Smarter Everything to discuss generative AI’s potential, natural language processing tools (such as ChatGPT), as well as exploring their current limitations and the challenges of fine-tuning these models. We will also discuss the accuracy of the responses generated, and how this new technology may affect the future of work and education.

Key Takeaways

• Information that becomes part of the training parameters for these models is not easily removed. They act like “black holes” of information, condensing data from all over the internet, and it’s difficult to see what is actually happening on the inside.
• The output of language models is probabilistic, and the models can generate irrelevant or non-existent information. These outputs are referred to as “hallucinations”.
• These systems cannot operate on their own quite yet and still require human oversight.

Time Stamps

• 02:46  Saurabh’s background
• 04:51  What is generative AI, and how is ChatGPT different from earlier ML models?
• 09:44 How ChatGPT learns through human feedback
• 14:28 How the models act like “black holes” of information and the issues this can cause
• 18:49 Thoughts on responses that could be patented or copyrighted information
• 19:55 Large language model temperature values explained
• 21:28 Generative AI and ChatGPT in education