Welcome to Issue 3 of unfolding ai, we are delighted to have you as a reader, and welcome to our new readers!

In addition to the round up of news and things you need to know, this week we start our series of topics which you will need to know about as you get started in implementing AI in your business. Sorry, no actual aliens.

We hope you find this newsletter informative and valuable, and we look forward to hearing your feedback.

It’s in depth this week, make sure if you are a subscriber you read it all!

Best regards,

Paul, Co-founder (and newsletter editor)

p.s As this is a brand new newsletter, we would really appreciate it if you spread the word and did a social share. The icons are at the top of the newsletter, don’t forget you can also read our earlier issues.

Staying in context

Welcome to the world of token context, a central aspect of Large Language Models (LLMs) like chatGPT. Tokens are the building blocks of language, while context provides the surrounding information that brings those tokens to life. However, chatGPT can only access a limited token context due to length constraints.

This limited token context presents both challenges and opportunities. The main challenge is that lack of full context can lead to misunderstandings, loss of nuance, and frustration when we attempt complex queries or detailed analyses. Tokens taken out of context can lose their intended meaning.

Yet working within tight token limits can also stimulate creativity. It pushes us to find new ways to express ideas more concisely and capture the essence of what we aim to convey. Solving the puzzle of conveying meaning within tight token constraints requires artfully balancing brevity and depth.

Here is how to go about thriving in a limited context world (which is about 7000 words on chatGPT). Remember the context includes your prompt, and the responses.

1. Break Down Complex Queries: Think of it as a creative exercise. Break down complex questions into smaller parts, ensuring that the token limit doesn't cut off your creative flow.

2. Utilise Summary Techniques: Summarize key points to convey the essence without exceeding token limitations. It's about finding the core of your idea and expressing it succinctly. You can ask chatGPT to do the summary.

3. Be Aware of Token Count: While using chatGPT, keep the model's constraints in mind. It's like painting within a frame, guiding your interactions.

4. Iterative Conversations: Engage in a dialogue with the model, building the conversation iteratively. It's a process that allows you to explore complex topics without overwhelming the token context.

5. Use the edit prompt button: avoid adding in context to correct prompts with a new prompt, fix the error above the response.

Alternatively, Claude.ai has a 100,000 context (about 90,000 words) providing it produces the response quality you are looking for.

Apple and AI

Apple is the only major tech company to not have overtly talked about where AI fits into their strategy. With a predominately consumer product focus, there are numerous products which clearly have machine learning and AI within them, and have had so for a number of years. For example, heartbeat pattern detection, fall detection and photograph categorisation.

He also went on to highlight that Apple spends $22 billion on research, up $3 billion on last year.

Voice to text

Previously known as ‘audio transcription’, voice to text is a very well established technology, one of which ‘whisper’ is within the IOS/Android app version of chatGPT.

The small wave form icon at the end of the app input bar engages voice mode. Its is very good, much better than the current in-built translations. I use it frequently when on the go to take notes. Having taken a note my usual next step in chatGPT is a prompt along these lines.

please tidy the above up, do not summarise, correct grammar, and remove pauses or duplications. 

Transcribing is also a service and solution we are using at unfold:ai to streamline the ‘type this up’ process from our meetings and brainstorming sessions. These external tools are fast becoming integrated into the existing tools such as microsoft Teams. These tools have innovated quite quickly this year, we advise having a fresh look at them. I am currently using otter.ai, others we are testing are fireflies.ai. Both are subscription tools for any real use.

The changing terms of service risk.

All of the major technology firms, and websites, such as reddit, iStock are pretty upset about either side of the ‘you used our content’ debate. Some of the major platforms are taking steps via terms of service to future proof (via a ToS Update) how they will use User Generated Content and Data. I.e. everything you say, do, type, on the platform. You don’t think of a Zoom meeting as ‘user generated content’, but Zoom does.

Zoom is no example, and is an early indicator of what will be a common risk. Note they are still bound by privacy laws, this is a future usage risk.

Here is the link to the ToS (see section 10).

• Ownership of Data: Zoom owns data about how you use their services, like what features you access and any diagnostic information.

• Use of Your Content: They can use the things you say or share for different purposes, including marketing, improving their products, and developing new technologies like AI and chatbots.

• Potential Risks:

  • Privacy Concerns: Your conversations and usage patterns might be used in ways you might not expect.

  • AI Training: Your data might be used to train virtual avatars or voice-generating programs.

  • Control Over Your Data: You're giving Zoom a lot of control over your content, which could impact your digital identity rights.

It’s given me food for thought, as to which meeting services I would recommend.

Quick things to check out

2D to 3D and model generation will be a key activity as we move to more experience and geospatial solutions.

Fun Factory YouTube channel, an inspiring source of AI generated video content, they are pushing the boundaries on what is achievable.

The Wharton school is stepping into the AI in education conversation with a series of videos to help.

A prompt to try in chatGPT (and other LLMs)

Here is the corner of a typical product brainstorming session. Let’s take one of those to the next level

As an experience Business Analyst, you are refining the output from a product owner brainstorming session for a improvements to the email order management system and processes. Refine the following requirements into any identifiable child features.

Requirements

Ability to prioritise and categorise orders
Customisable email templates for correspondence
Automatic tagging and categorisation of emails

This use case continues in the premium section.

Our Events

Go Premium on us

Curious what a premium subscriber gets to discover. Try it for a month on us.

Subscribe to Premium to read the rest.