ChatGPT, Generative AI and LLMs Explained: 2023 K-12 Panel on Prompts and Schools
At the 2023 AI Connect K-12 Data Science Conference, data leaders from UNC Health, Nike, Pella and others explained ChatGPT, generative AI and large language models. They covered prompt engineering, hallucinations, cheating in schools, coding with AI and the future of data careers.
By The AI Enterprise Magazine Team4 min read
Key takeaways
- ChatGPT is a drafting assistant. Rachini Moosavi likened it to having a hundred interns working side by side.
- Prompts matter. Clear prompts reduce errors, and models can still hallucinate because they predict statistically.
- Cite and verify. Panelists said AI answers can be wrong and sources should be cited.
- Protect company data. Nike's Taj said staff were encouraged to use ChatGPT but not with work email or work data.
- Generalists gain ground. Jessica Larson expected a shift toward people who can use tools across many tasks.
At the 2023 AI Connect K-12 Data Science Conference, a cross-industry panel explained what ChatGPT, generative AI and large language models are, how prompts shape their answers, and what the technology meant for schools and careers. Brian Mink, co-founder and president of AI Connect, moderated. Panelists were Jai Prakash of Pella Corporation, Rachini Moosavi, chief analytics officer at UNC Health, Jessica Larson, formerly of Instacart and Pinterest, Henry Clavo of the Consumer Financial Protection Bureau, and Taj, who said he ran AI for Nike North America. The panelists described the technology as it stood in 2023.
What are ChatGPT, generative AI and large language models?
Moosavi, who said she regularly briefs healthcare leaders on the subject, described ChatGPT as a chat-based application built on a large language model trained on a vast amount of internet text. At the time, she noted, its knowledge ended in September 2021, and not all of its training data was accurate.
She called it "like having a hundred interns working with you side by side," useful for first drafts. Generative AI, she said, differs from earlier AI such as risk models because it creates new content, whether images or bedtime stories she tells her daughter.
How does prompt engineering work?
Clavo said a prompt is turned into probabilities: the model predicts a response from the input and its training data, which can produce hallucinations. He advised writing clear prompts tied to the topic to reduce errors.
Larson added that models learn relationships between words across massive amounts of text. Very common words connect to so many others that results become vague, while very rare words have too few connections to help. Good prompts, she said, balance the two.
Will students use ChatGPT to cheat?
Larson compared the debate to the arrival of the internet and Google, and said tools to detect plagiarism would likely follow, as they did before. She recommended using the model as a starting point, writing independently and citing sources.
Taj pointed to an attorney who cited cases that did not exist, noting that language models do not fully understand what they say. At his company, he said, "we encourage people to use ChatGPT but don't use your work email and don't use your data from your work," because the publicly traded firm must protect its data.
Prakash said he used the tool while helping his daughter with math, found the first answers often wrong and had to guide it. He argued education should shift toward critical thinking and noted that memorizing multiplication tables was no longer required.
How will data scientists use these tools?
Prakash said he knows how to solve a problem but is not a strong programmer, so he types his logic into ChatGPT and uses the code it returns. It may not be the most efficient code, he said, but it speeds up his work.
Taj said his organization was looking at using AI to make engineers more productive without adding headcount. He described vendors offering to comment legacy code and make it searchable with semantic search, and models that can spot and correct coding errors. He also expected vendors to build products on top of GPT to address hallucinations, reliability and data privacy, a trend Mink agreed was coming.
What does the future of work look like?
Larson predicted a shift toward generalists, saying it was "not necessarily as important anymore to be an expert in one thing in particular." Mink summarized the key skill as literacy: knowing what you know, what you do not and what help you need.
On curriculum, Moosavi expected at least one data science course to become required, and Larson said elementary and junior high math already includes data science concepts, much of it statistics. Taj described data science as finding a function through data, "a squiggly line running through the data."
What did panelists see as the most impactful application?
In a closing lightning round, Clavo pointed to companies building their own financial prediction models. Taj voiced concern about deepfakes during an election cycle. Moosavi hoped generative AI could help transform healthcare, Larson saw lower barriers to entry for people excluded from computing, and Prakash expected digital-twin style applications in consumer products and manufacturing. Mink said generative AI would raise public awareness of data in fields such as law and medicine.
Where can you hear more from AI and data leaders?
AI Connect continues the conversation with 24 free live webinars on the 2027 schedule across four series: the AI × Enterprise Panel Series, AI Demo Days, Agentic AI Showcases and the Plug & Play Series. Details are at /webinars/, and registration for 2027 events is coming soon.
Frequently asked questions
What is a large language model?
Panelists described it as a model trained on enormous amounts of text that predicts likely responses to a prompt; ChatGPT is a chat application built on one.
What is prompt engineering?
Writing clear, on-topic prompts to get better output; Jessica Larson said it means balancing very common and very rare words.
Will students use ChatGPT to cheat?
Panelists compared it to the arrival of the internet and said schools would need detection, citation norms and more emphasis on critical thinking.
Should data science be taught in K-12?
Moosavi expected at least one data science course to become required, and Larson said existing math already includes data science concepts.
This article is based on remarks by Jai Prakash, Pella Corporation; Rachini Moosavi, Chief Analytics Officer, UNC Health; Jessica Larson, formerly Instacart and Pinterest; Henry Clavo, CFPB; Taj (surname not given), AI lead, Nike North America; moderator Brian Mink, Co-founder and President, AI Connect at AI Connect K-12 Data Science Conference 2023 on July 27, 2023. More coverage: More AI Connect Events.
Filed under: AI Use Cases & Applications, People, Culture & Leadership, AI Governance & Trust
