#3  Making the most of ChatGPT (& chocolate brownies!)

#3 Making the most of ChatGPT (& chocolate brownies!)

ChatGPT has fired everyone up to the possibilities and threats of generating content quickly.?Let’s see how to get the best from this new tool, and some practical tips to leverage its power.

1. Access the paid version – if you can.

Apparently it costs around US$700,000 per day to run the servers powering ChatGPT, so the free tier often is overloaded and you can’t get access.?Bit of a problem if you’ve got a class to demonstrate it in.?But pay the 20 bucks a month for ChatGPT Plus and you’re in the full access group, and even when it’s full you get an instant email link that lets you jump the queue and start working. I know not everyone can afford to pay, and that’s fine, but if you can, or even better if you can persuade school or college to fund your subscription, then it does make things easier.?And you get access to the latest iteration of the ChatGPT engine – 4.0 as I write this – that has access to a much larger dataset and can give more detailed responses than the free 3.5 version.

2. Develop your prompts.

There’s been much made in recent months about “prompt engineering” and some people are making a LOT of money selling access to lists of prompts, helping others develop prompt-writing skills – I even saw a job post on LinkedIn for a Prompt Engineer paying over 300k!?So this is a thing.

When starting out, everyone typically enters questions like we’ve always done to Google.?Single line simple tasks - “Show me the recipe for chocolate brownies”. (Bonus tip for today - the Mary Berry one at https://thehappyfoodie.co.uk/recipes/chocolate-brownies/ is brilliant)

No alt text provided for this image

To get the best out of ChatGPT though, treat it like a personal assistant – a person you are actually talking to.?The more detail you give about what it is that you want, the better ChatGPT will normally perform in generating well-reasoned, complex and deeper responses.

No alt text provided for this image

Start by giving ChatGPT a role.

Then ask it a question or pose a problem you want it to solve.?Be as detailed as possible – don’t be afraid to use several sentences to specify just what it is you are after.

Next tell ChatGPT how you want your answer formatted – you can specify in table format, as a word-cloud, a Gantt chart – experiment and see what it will give you!

Finally, specify what tone you want – I often say “Give your response in a conversational, relaxed style, using UK English

Then see what comes back...

?

Now this is where ChatGPT really starts to shine.?Let’s say that there’s a point made in the first response that you want more details on.

Well you can just say “Expand on xxxxxx for me, giving me 5 ways that I could explain this to my class and providing examples” for instance.

ChatGPT remembers earlier responses, so you can build on them in a conversational manner to explore a subject more deeply, to go down tangential paths or just amplify a brief response into a more nuanced and comprehensive one.

It’s a new and very powerful way of using dialogue to build and develop your requests, deepening and complexifying the responses to end up with the material you are seeking.?Most of the time that material will then need editing and polishing, with your own style adding in, but it’s amazing how fast you can create outlines and fairly detailed breakdowns quickly using this technique.

A BIG caveat though – these AI chatbots do have a tendency to make things up – this is called “hallucinating”!?

So EVERYTHING needs to be checked and verified.?On a good day, links and sources may mostly be valid, but you do need to check everything for validity. This is one of the most important parts of any student based exercise in class, as the material in the responses may seem authoritative and well written, even though it has little basis in truth or fact.?Great learning for a class to go through such a response, fact-check it and then re-write the piece “properly”.

Here's a link to a recent article about ChatGPT making up articles supposedly created by the Guardian newspaper - https://www.theguardian.com/commentisfree/2023/apr/06/ai-chatgpt-guardian-technology-risks-fake-article

No alt text provided for this image

And Ethan and liliach Mollick’s excellent?HBR article at https://hbsp.harvard.edu/inspiring-minds/let-chatgpt-be-your-teaching-assistant is worth a read…


Joe Houghton?is an Assistant Professor at UCD Smurfit Graduate School of Business where he directs the MSc programmes in Project Management. After a career in IT in multinationals, Joe switched into a portfolio career of University teaching, management coaching and training.

He has authored 5 books to date including "Innovative teaching with AI: Creative approaches to enhancing learning in education", and "Project Management made easy...: the ECCSR approach".

Contact Joe on email at?[email protected]?for any requests for training, workshops or keynote speaking.

Elizabeth Hitches

PhD Candidate at the University of Queensland

1 年

Interesting! Learning lots from you!

要查看或添加评论,请登录

Joe Houghton的更多文章

  • #55 - Bloom & AI...

    #55 - Bloom & AI...

    An interesting take this week on Bloom's Taxonomy - a teaching model many educators were raised on. Here it's paired…

  • #54 - Creating a learning session - 3

    #54 - Creating a learning session - 3

    In issue 52 - "Creating a learning session - 1" & issue 53 - "Creating a learning session - 2" we began putting a…

    10 条评论
  • #53 - Creating a learning session - 2

    #53 - Creating a learning session - 2

    Engaged adult learners In issue 52 - "Creating a learning session - 1" we began putting a learning session together…

    2 条评论
  • #52 - Creating a learning session - 1

    #52 - Creating a learning session - 1

    OK, today I want to try something a little different. Earlier this week we were recording episode 7 of the ILOTA Things…

  • #51 - Comparing the paid AI's

    #51 - Comparing the paid AI's

    One issue with the explosion of all the different AI models is access to their most powerful & capable versions. There…

    6 条评论
  • #50 - The half century!

    #50 - The half century!

    Issue #50! Wow - I thought that this little newsletter probably wouldn't have legs when I started experimenting with…

    1 条评论
  • #49 Towards our own data...

    #49 Towards our own data...

    I think that the next stage in the evolution of AI for most users is going to be the emergence of tools that allow us…

    1 条评论
  • #48 - Europe's new AI Act

    #48 - Europe's new AI Act

    The new AI Act is set to become law across the EU in August this year, bringing tighter controls on AI usage across…

    2 条评论
  • #47 - Will AI Avatars replace us?

    #47 - Will AI Avatars replace us?

    AI avatars are fast becoming a thing. Whether it's social media influencers or - and of more interest to most of us -…

    5 条评论
  • #46 - Locally run AI's

    #46 - Locally run AI's

    One of the key issues for everyone who has been experimenting with the slew of AI tools we've been presented with over…

社区洞察

其他会员也浏览了