Yes, you read that correctly — and no, I’m not suggesting AI has the flu.

Yesterday this topic came up when talking with one of my coaching clients, and I thought it was worth sharing (btw, I still have just one coaching spot available if you are interested).

First of all, let’s cover what we mean when we say, “ChatGPT hallucinates” (or to put it bluntly — makes stuff up!)

What are ChatGPT Hallucinations?

I asked ChatGPT to provide a quick explanation about what it means for ChatGPT to ‘hallucinate’. This is what it came up with:

“Picture this: ChatGPT, the AI whiz kid, is like that friend who’s really smart but sometimes comes up with crazy, out-of-the-blue ideas. You know, the friend who’s full of confidence even when they’re not entirely sure about something? Well, that’s our dear ChatGPT.

When ChatGPT “hallucinates,” it means it’s got a wild imagination! It’s like giving it a prompt and watching it go on a creative spree, conjuring up information that might sound believable but is not actually supported by facts or training data. It’s like ChatGPT is trying to impress us with its fictional tales, but we need to keep our skeptic hats on.

You see, this AI genius can confidently produce answers that are a bit “out there” and not necessarily trustworthy. It’s like asking ChatGPT about the revenue of a company, and it responds with a number that seems like it was “plucked from thin air”! We all have that one friend who’s a walking encyclopedia of made-up facts, and ChatGPT can be a bit like that.

The thing is, as ChatGPT gets more convincing with its responses, we might start believing its tall tales without questioning them. It’s like when your friend spins such a compelling story that you can’t help but nod along, even if you suspect they’re pulling your leg. We have to be careful not to fall into the trap of relying too much on ChatGPT’s words, especially in serious matters.

Oh, and let’s not forget about the risks! Sometimes, ChatGPT can accidentally spread misinformation or harmful content, like a friend who accidentally blurts out something hurtful without realizing it. So, it’s essential to keep an eye on what ChatGPT says and not take everything at face value.

But don’t worry, the smart folks at OpenAI are working hard to make ChatGPT more reliable and less prone to flights of fancy. They’re like the watchful parents, teaching our AI friend to be more responsible and thoughtful in its responses.

So, while ChatGPT is an amazing tool and can give us some mind-blowing answers, we should remember that it’s not infallible. We need to stay critical, verify its information, and use it wisely, just like we would with our quirky but brilliant friend. In the end, with a bit of caution and common sense, we can benefit from ChatGPT’s brilliance without falling for its charming hallucinations!”

What does it mean for ChatGPT to have a high temperature?

Essentially, “temperature” refers to the level of creativity or general randomness for a given LLM (large language model). In ChatGPT’s case, the scale is 0 to 1 — and ChatGPT’s temperature setting is 0.7, meaning that it has quite a bit of free reign to be ‘creative’ with the answers it gives, and the way it comes up with them. In other words, the lower the temperature, the closer the output is to the original data — but also the lower the temperature, the less availability the model has to come up with innovative responses.

What are the different temperature settings?

Low Temperature (0 to 0.3): Setting a low temperature results in more focused, coherent, and conservative outputs. The AI tends to choose the most probable words, leading to more deterministic responses.
Medium Temperature (0.3 to 0.7): With a medium temperature, you strike a balance between creativity and coherence. The AI generates responses that are moderately diverse and explore a broader range of possibilities.
High Temperature (0.7 to 1): A high temperature setting produces highly creative and diverse responses. The AI becomes more exploratory and imaginative, sometimes generating surprising or offbeat answers.

How to write a prompt specifying the ‘temperature’?

Simple! Write your prompt as you would normally, then on the next line type in:

Temperature: 0.4 (or whatever number you desire)

My suggestion is to play with various temperature settings for the same prompts to see what it comes up with, and what best suits your needs.

~~~~~

OK, well that’s all from the weird world of all things AI for this evening…

Oh, and yesterday evening I was discussing several subjects related to quantum physics, consciousness, holographic universes, manifestation, the nature of time, awareness, epigenetics (just to name a few) — and am thinking I might start touching on some of those subjects as well in my updates from time to time…

Wishing you a truly glorious day!

~ Bella

Ms Bella St John