Expert author Mark Pesce’s masterclass on features and concerns of ChatGPT and other chatbots
ChatGPT is now more than a year old, and much has changed in the short time it’s been with us.
As author of Getting Started on ChatGPT and AI Chatbots Mark Pesce explained to this week’s masterclass there now a “Big 4” of AI Chatbots, and even Amazon has launched another called “Q,” its artificial intelligence (AI) helper for business users.
The Big 4 are:
· OpenAI ChatGPT which comes in different “flavours” of which the paid ChatGPT-4 is best
· Microsoft Copilot which is free and based on GPT-4 & 3.5. Every Windows 11 system has received an “always-on” upgrade to Copilot, with integration recently into Windows 10.
· Google Bard which is free and integrated into GSuite (an enhanced Bard Advanced may be released as a paid version)
· Anthropic Claude which is comparable to GPT-4 (and priced similarly) and is a Google-backed AI startup founded by former OpenAI engineers
Do AI chatbots think?
No, chatbots are not thinking, says Mark. At times they’re not truthful and they have no way of being able to know this. A completely wrong response is called a “hallucination” or a “confabulation.”
AI chatbots use “generative AI” which is recognising patterns in massive amounts of data that’s been fed into the models (including the false and nasty stuff) and then using them to create new data with similar characteristics.
How do I know if my chatbot is hallucinating or lying?
Chatbots work best with direct and specific questions. However, if you don’t know the subject matter yourself, don’t trust your chatbot, says Mark. It may have made a “best guess” as to the answer.
Mark suggests you can check if the chatbot is lying by asking the same question again but in a completely different way. Be less vague, says Mark, or give an example or two. Do you get the same response?
You could also try checking your chatbot’s answers against other those from other chatbots. You could do a normal Google search.
Or, you could ask a human who is expert in the subject.
What is a prompt and which prompts get the best results?
Prompt is the word given to the question, statement, or instruction you give the chatbot.
As Mark said, chatbots like specific and direct questions. It helps to tell them to keep the answer brief with “Be brief” at the end of each prompt. It also helps to give Character + Context + Challenge.
First, get your chatbot “into character.” Think of your prompt as a tiny story.
For example, your prompt could read:
“You are a world-class CFO. Over your career you’ve come up the ranks, managing the finances for a wide range of businesses in the service, manufacturing and retailing sectors….”
When you put the chatbot into character you help it sift through vast amounts of data.
Be aware that asking it to be characters such as a statistician, social media influencer, or cybersecurity expert may come with higher risks as they may offer answers that are wildly inaccurate or, as Mark says, “off the rails.”
Again, if it’s not your area of expertise, check answers.
You have a character, now give it context.
“Today you’re confronted with some hard choices. You need to cut costs by five percent over the remainder of the fiscal year – and you want to minimise impacts on both staff and customers….”
Finally, set your chatbot its challenge.
“Create three scenarios on cost-cutting alternatives and their impacts on the business. Explain your choices.”
Have a play, try it yourself, and read our article from Mark Pesce’s original masterclass in July on crafting compelling interactions with chatbots. on using ChatGPT
Chatbots are useful for reasoning and summarising
Mark says he uses this feature a lot himself, as chatbots are very useful for summarising documents in text, Word, PDF and other formats.
Upload the document to the chatbot and ask the chatbot to provide a summary with 5 key points, or a summary in 500 words, or in dot points, or in one line of text.
Chatbots can also analyse a spreadsheet, a presentation, and other kinds of documents. Mark suggests try it and see!
Take care what you share
Mark made this point throughout his presentation. If you don’t want it shouted from the rooftops, don’t type it into a chatbot.
Chatbots upload your prompts and documents into the cloud and this means they could end up in anyone’s hands.
Mark also mentioned something called a “prompt injection” which is a cyberattack tool that recently emerged as a security threat issue.
Anything you upload to an AI chatbot can contain hidden prompts for that chatbot. In essence, a hacker might insert hidden instructions into the AI so it becomes a “poisoned” search. There is currently no strategy around how to avoid that.
Turning mountains into molehills
One of the newest technologies is the Autonomous Agent. It converses with the AI chatbot, checking the answers (“Does this look right?”) and reflecting its “opinion” back and forth until it gets to “That looks like the answer.”
This is the cutting edge of what’s possible, says Mark, and it’s expected Autonomous Agents will turn a lot of mountains into molehills for humans. He says that with these tools a person will have more skills to complete a broader range of tasks.
Some chatbots now offer image generators, and Mark showed examples of its efforts to create an image that puts a Santa on a beach. He asked it to add snow and sparkles, which it did very effectively. Mark says such images can’t be copyrighted because they were not created by a human.
Working with AI
Given AI chatbots can lie, hallucinate, and may keep copies of everything you give them, it’s worth treating them with caution, and putting thought into how your business can and should work with AI chatbots like ChatGPT.
Mark suggests you should discuss as a team or as an organisation:
· Policies: When do we use it? What problems to we want to solve? Where do we want to use it. Where not?
· Procedures: How do we use it? What are the best prompts? What character and context do we use?
· Protocols: What happens if it goes wrong?
A thank you and closer from ChatGPT-4 itself
This article was written by a human (hi fellow human reader!!) without any help from ChatGPT, but I thought it would be fun to prompt ChatGPT to imagine it was a speaker who'd just done a presentation on this subject, and write a short thank you and summary to close out the piece.
Note, I did not feed this article into the chatbot before asking for this.
So, let me hand over to ChatGPT….take it away toddler bot…..
Ladies and gentlemen, I want to express my heartfelt gratitude to all of you for being a part of this insightful discussion on ChatGPT and AI chatbots today. It's been a pleasure sharing this journey into the world of artificial intelligence and its transformative potential.
In closing, let's remember that AI chatbots are not just about technology; they're about enhancing human experiences, improving efficiency, and opening up new possibilities. They can revolutionize customer support, streamline processes, and empower businesses to connect with their customers on a deeper level.
As we embrace this technology responsibly, let's keep in mind that the future of AI chatbots is in our hands, and together, we can shape it for the better. Thank you once again for your time and attention.
Thanks ChatGPT! Human signing off and thanking you also.
See Adam Grant in virtual livestream OR in person in 2024
There's an exciting event to catch in early 2024: Learning direct from one of the world's leading authorities on work culture and work-life. You have two options.
- Option #1: If you're in Australia don’t miss Adam Grant’s highly-anticipated Australian tour in February 2024 ADAM GRANT – LIVE: Unlock Hidden Potential & Transform WorkLife. Tickets are selling fast (Some categories SOLD OUT).
- Option #2: To watch a (live virtual) livestream of this event, join 8000 global business leaders who are professionally and personally developing with a Growth Faculty membership. The livestream is included in membership. See full list of membership inclusions at Growth Faculty annual membership:
· Unlimited access to dozens of live virtual masterclasses on trending topics around work, career, leadership, and culture; such as psychological safety, burnout, difficult conversations, strategy and execution, micromanaging, accountability and more, with opportunities to chat, ask questions of the speaker, and interact with your peers.
· Unlimited access to our multi-module leadership programs at both emerging and senior leader level.
· Discounts on tickets to be in audiences at our Global Headliner in-person events featuring top name speakers.
· Year-round leadership content on demand with replays of our live events, highlight reels, interviews with top business authors, and more.
Join a community of knowledge seekers who are inspired by the best. See our latest program of live virtual and in-person events.