Alex Calder, Staying informed of AI advancements is not just an advantage but a necessity. Our mission is to equip you and your organisation with timely insights, critical updates and expert analysis on the latest in AI so you can navigate the future with confidence and precision.
In this weekly, we’ll bring together the latest news to help you plot the course you should take to navigate safely through the choppy waters of AI change.
For the first edition, there’s only really one big news story in AI over the last week and that’s the release of OpenAI’s o1 model so we’ll concentrate on that today. In future issues, I expect we’ll be a bit more discursive and cover a range of stories from across AI and marketing technology.
OpenAI o1 Release
Have you ever tried to ask ChatGPT how many times the letter ‘r’ appears in the word ‘strawberry’?
If you have, you’ll know that it often falls down at this (to us) simple task. This is a great example of the idea of a jagged edge when it comes to the abilities of AI. To the human mind, knowing how many ‘r’s there are in ‘strawberry’ is a much easier task than, say, creating a SWOT analysis for your organisation. But AI fails at the former while succeeding at the latter, showing that there is no linear border between what AI can do, and what it can’t.
The reason for this is at the core of how generative AI tools like ChatGPT work, the reliance of patterns and the outputting of tokens, i.e. delivering a response token-by-token could result in stra then wber then ry, and that lack of context of what came before means it can’t answer something as simple as the number of letters in a word.
OpenAI o1 seeks to solve this challenge, at the expense of increased time and cost. This model takes the time to think out it’s answer, sometimes taking 10/20 seconds before responding, in order to give a more considered answer and to review it’s response.
Its ability to think through complex problems is astonishing and well worth experimenting with if you’re a Pro ChatGPT user.
What are the limitations of the o1 model?
- Only paying ChatGPT users can access o1, meaning it’s limited to Plus and Teams members, with Enterprise rolling out soon.
- It’s limited in it’s usage, you won’t be able to ask questions all day long, eventually you’ll hit a limit.
- It’s very costly to use via API. Don’t expect to see it rolled out to the tools you use that are built on ChatGPT, the cost vastly outweighs it’s usefulness.
- It’s text-only, meaning you can’t use it with voice, file uploads or web browsing. It’s focused purely on responding to your text prompts.
We’re only at the beginning of this implementation however, the CEO of OpenAI, Sam Altman, commented at a recent event that ‘even in the coming months you’ll see it get a lot better as we move from o1-preview to o1.’
It’s important to note o1 is not a replacement for GPT-4o which still has much more practicality for most users due to it’s multi-modal nature and speed.
Presumably, we’ll see GPT-5 in the not too distant future, which will see even more advances.
And finally…
I visited a couple of AI-focused events last week - or perhaps, events which should have been more AI focused.