PrecisionLearning and Forgetting
Welcome to the second edition of the Precision Weave newsletter.
In this edition we have a new blog post about why your AI sometimes forgets things and my thoughts how to cope with the incredible pace of change in AI and technology at the moment.
Also, the lovely people at the West Highland Chamber of Commerce have arranged for me to give a talk titled Practical AI Strategies For Small Business Owners at three different venues across the region this spring.
- Fort William – Tuesday 28th April
- Skye – Tuesday 5th May
- Oban – Tuesday 12th May
I will send out more information as soon as the sessions are open to booking so watch this space.
All the best,
Adrian
New on the Website
Why Fresh Conversations Beat Long Ones
Read ArticleEditorial
I used to think that the software industry re-invented itself every 18 months. Now, with AI, it’s less than every 18 weeks.
The pace of change is like nothing I have ever seen. New models, products and services spring up daily like mushrooms in October. So how do we cope and where should we focus our attention?
My advice is to make sure you understand the fundamentals because nothing else will make sense unless you understand the basics. With AI, that’s Models, Context and Prompts.
Prompts are how you interact with the AI. Give it the right prompt and you’ll get the right result – in theory – but sometimes they forget what you told them and sometimes they make things up. That’s the Context problem, and model makers like OpenAI and Anthropic are in a race to improve how much context their models can handle without making mistakes.
The trick is to give the model clear instructions (Prompt) with just enough information (Context) to fulfil a task. Do that, and your results will improve with every new Model that comes out.
Product vendors don’t want you to know that. They want you to believe that their product will save you all that icky prompting and give you perfect results every time. I disagree – if you don’t know the fundamentals how can you recover when something’s not quite right?
In other words, don’t outsource the thinking.
Happy prompting!