Let me start by saying I am completely biased - I am really not bought into or impressed by all the AI hype recently. I’m pretty sure ChatGPT and all the others are just a fancy interface for Google, so I have resisted playing around with them.
A lot of these Large Language Models (LLMs) have been released in the past few months. I think it’s safe to say the most notable has been OpenAI’s ChatGPT (3 AND 4). It feels like everyone is using ChatGPT to “supercharge” their days and their businesses (I was particularly impressed by the person who got ChatGPT to plan their weekly shopping), at least according to my Twitter and LinkedIn feeds.
Rather than define a LLM or ChatGPT, for now, here’s how OpenAI describe their new toy tool:
“We’ve trained a model called ChatGPT which interacts in a conversational way. The dialogue format makes it possible for ChatGPT to answer followup questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests.” OpenAI
Okay that sounds a bit more advanced than Google, so here goes nothing.
Thought #1 - these disclaimers are a bit scary
Before you can use the ChatGPT chat interface, there are three pages of disclaimers. They are short and easy to read - but they are a tad concerning. The main points are a reminder this is a free preview for the purposes of research to improve the tool itself (and make it safer); data will be collected on you (so don’t share personal details); they are keen to collect feedback.
The point which really caught by attention - and feels very important to call out is:
I understand this is perhaps needed for OpenAI’s protection, but the majority of what I’ve seen shared overlooks the possibility of incorrect, biased or misleading information.
So far, I remain deeply unconvinced.
Thought #2 - is this any better than Google?
The next screen has some examples, and a quick reminder of the capabilities (reminds what you’ve said and allows you to correct) and limitations (as above, but also limited knowledge of the world after 2021 - what??)
Now for the fun bit - chatting!
My best first attempt
In a terrible attempt to work out what is going on with this chat bot I tried:
What is the best TV show?To which ChatGPT responded with a long winded version of the best TV show is subjective, here are 6 well rated TV shows including Breaking Bad, Friends and Game of Thrones -
ChatGPT: Ultimately, the "best" TV show will depend on your personal taste and interest.
Google search actually returned some picks for me personally and popular shows from right now (full disclosure I have a Google Chromecast so Google gets to see what I watch all the time and I have three connected streaming services, a bit of an unfair advantage).
Teach me something
Trying to get a bit more specific (and remembering a thread I saw on how ChatGPT is going to replace all teachers…):
How can I learn data science?Should everyone learn about AI?Both of which gave long answers (who knew an AI could be so chatty) again sounding very detached and impersonal (and having 6 bullet points). But a useful starting point.
This was when I got thinking about Google search - how a good Google search is a skill in itself. This is actually something I try to include in all my tech education because of how invaluable this is. Maybe ChatGPT is helpful for people who aren’t good at Googling? I’m still very unconvinced of the benefit.
Third times the charm
I tried to remember the social media posts I’d seen which interested me, and I had a go of recreating these:
Optimize a daily schedule for an entrepreneurChatGPT: Optimizing a daily schedule for an entrepreneur can vary depending on individual preferences and the nature of their business. However, here's a sample schedule that incorporates important aspects for productivity, personal well-being, and business growth
How much money do you need to quit your job?ChatGPT: To determine the exact amount you need, it is advisable to create a detailed budget, consult with a financial advisor, and consider your specific circumstances and goals.
YAWN - the responses to these were so generic. All my hopes of finding something useful or inspiring seemed to be slipping away with every attempt.
Can I make my own version of you?ChatGPT: As an AI language model, I'm the result of extensive research, development, and training by a team of experts at OpenAI. Replicating or creating your own version of me would require access to the necessary infrastructure, data, and expertise in machine learning and natural language processing.
While OpenAI has released APIs that allow developers to build applications using the GPT-3.5 architecture (which powers me), creating a fully functional AI like me would still be a significant undertaking, involving substantial computational resources, access to large-scale datasets, and expertise in AI research and engineering.
If you're interested in working with AI and building your own applications, you can explore OpenAI's developer resources and APIs to leverage the power of language models like GPT-3.5. However, it's important to note that creating a full-scale AI like me would require a substantial investment of time, resources, and expertise.
The most interesting response so far and it’s basically a sales pitch for the ChatGPT API.
Thought #3 - am I the problem?
Following the advice of Taylor Swift, I started to think I am the problem. I'm not doing this right. I must be thinking about everything wrong. And then an idea hit me, if this thing in front of me is the amazingly smart brain it claims to be, why not ask it:
What is the best way to use you?The usual generic beginning,
ChatGPT: The best way to use me, as an AI language model, depends on your specific needs and goals
was followed by some actually useful tips.
How ChatGPT advises to use ChatGPT:
Ask specific and clear questions
Break down complex queries
Utilize context
Experiment and iterate
Verify information
Explore various use cases
ChatGPT: Overall, the best way to use me is to view me as a tool to augment your own knowledge and capabilities. By utilizing my strengths and complementing them with your expertise, you can make the most of our interactions.
Hmmmmmm, I'm going to take a break for now and try again soon supercharged by ChatGPT's advice.