12 days of OpenAI, red lines and some photo fun

Welcome to the December edition of the AI: Actionable Insights newsletter, in this edition we're talking about OpenAI o1 and 01 Pro. What this means to you, and why this new tier at $200 a month (10x their current price) is actually very telling of their future plans.

We're also talking about red lines - sure there are some really useful things you could do with AI, but should you? Is that more of a business or brand decision?

And finally, some AI generated photo fun at the end...

ChatGPT o1 updated and o1 Pro, a taste of the future

OpenAI kicked off their "12 days of OpenAI", where they're promising to release or demo something every working day until Christmas, by taking their latest model o1 out of preview with a couple of updates.

o1 is their deeper thinking model, best suited to more complex tasks - it takes time to plan out it's answers similar to the chain of thought method I often share at talks.

They've made improvements to the speed of response for shorter queries and included the ability to upload images for it to interpret.

More notable is the launch of their new ChatGPT Pro tier - at a cool $200 per user, per month this is aimed at the really heavy users who need the absolute best model money can buy.

That's quite the step up from the standard $20 a month, and I think a shape of things to come as we move to bigger and better models.

Really fascinating that they set out so clearly on the launch page "in the future, we expect to add more powerful ... features to this plan".

Watch the launch video here if you'd like to know more:

Actionable Insight:

If OpenAI are making their latest models 10x more expensive to access, when they launch a much more significant model do you have the business case ready to justify that cost?

10x the cost per license is quite the jump, how much better would it need to be to justify that?

This is another great reason to document your use-cases and create proper pilots, with clear measurement of outcomes and their benefits.

If the model isn't quite there for a given use-case yet but when it is it's 10x more expensive, that's going to need a much more significant case to be made to senior management on the level of investment needed.

Document your use-cases, work out what success looks like and how you're going to measure it. When the price rises come, you're all set to make that case.

We share a simple framework for AI adoption that does exactly that as part of the Campfire Workshop we run with Dorset Chamber - join us on the next one on Jan 23rd if you'd like to learn more.

Red lines and an AI statement

I've done a lot of workshops recently, and something we talk about is having an AI Statement - a document that sets out what and how you will use AI.

We've started to be asked for this by some partners we work with, and I'm hearing it's starting to appear in the procurement process for others too.

It's also a great thing to set out your position for potential hires - having spoken to a lot of students recently, they're keen to use AI and want to know where businesses stand.

You can find the Intergage AI Statement here

Once you've worked out the things you could use AI for, it's equally important to decide if you should use AI for that task.

Let's look at three examples:

1) Customer phone calls: if I was a Broadband provider, I could probably implement a customer service AI system to handle a significant number of incoming calls - as a consumer, I'm probably ok with that if it means not spending 45 minutes on hold (true story, sigh).

But if I'm ringing a doctor's surgery, I'm probably less comfortable using a similar system - in both cases I'm sure the technology is there to book appointments and answer common queries but it's the expectation of the service user or customer that matters.

2) At a recent event I heard a speaker say proudly that you could generate content 70% faster using generative AI. Meaning you could post to LinkedIn every day, even multiple times a day!

Yes, you probably could...

...but it'd almost certainly be useless AI slop, and does anyone really want to hear from you THAT MUCH? Sometimes surely less is more.

3) Sending the virtual AI Luke would not be appropriate for a paid presentation, but it was fine sharing some information internally to my colleagues - and as a quick apology note to folks when I had to cancel going to an event as I'd lost my voice.

It's a decision if you choose to use AI - that's as much a brand, as it is a business and technical decision.

We're already starting to see some businesses promoting the fact they won't use AI for specific things - I'm sure that will be a point of difference that some wish to make as a selling point.

As we move out of the experimentation phase most businesses are in with their use of AI, we need to think deeper about what we choose to do.

Actionable Insight:

Once you've got use-cases and worked out things you could do with AI, take a step back. Is it what your customers would expect? Is it who you are as a brand? Who makes that decision in your organisation? Is it ethically appropriate?

Start to have these conversations in your business as you develop and experiment with AI - don't just have pockets of users disconnected across your organisation - this is a strategic direction, not just a tactical or operational one and senior leadership should set that direction.

And finally...

For a recent training session, I needed to send a headshot - I thought it'd be nice to create a version of me in a festive setting.

I trained a model on my likeness, which then allowed me to write a prompt to put me into any scene/outfit etc I wanted.

I used HeyGen with a selection of photos of me, as per their instructions, and then asked it to:

"Make my avatar look like one of Santa's helpers in an arctic and Christmas scene".

I was not prepared for the 4 results it generated.

1) Ok, that kinda looks like me, I could maybe use that.

2) Oh, so that is what I'd look like with long hair!

3) Ummm ok, long hair again and looking more feminine?

4) Yeah, that's me as a woman. What happened there???

All I can think is that the images it was trained on for "Santa's helpers" were mainly women? None of the other prompts I used created pictures like that!

Here are some others I tried, with me presenting in various forms, and as a jedi*.

*There was a semi-serious reason for the last one, it was suggested to me as a good test of the model to see if it got lighting right.

Actionable Insight:

Apologies for being a broken record but... have a process to check the output of any AI system.

We have a process for reviewing and editing content generated by AI, using a 15-step checklist for things like bias and stereotypes, spellings, grammar, common GPT-isms, accuracy etc - do you?

If not, consider making one as an extension of the AI Statement we talked about above.

"The AI made a mistake" should not be a valid excuse.

Upcoming training workshops and talks:

Campfire Workshop Jan 2025 - Dorset Chamber (paid workshop) - Jan 23rd 0900-1230 - Learn about a simple AI adoption framework, how to develop a roadmap for piloting and using AI in your business and see real world use-cases and examples.

AI Trends for 2025 (free webinar) - Jan 16th 1000-1130 (Webinar) - Looking at the current rate of change, new tools and capabilities coming to market, and what to expect as a business from AI in 2025.

Thanks for reading in 2024 - we'll be back in 2025, kicking off with some future gazing and much more.

If you have any feedback or topics that you’d like us to cover, please do let me know!

For more tailored help with anything we've talked about in these newsletters, please do get in touch.

Wishing you a Merry Christmas, and a happy, healthy and wealthy 2025.

Cheers,

Luke