Teaser
Wondering what AI developments we can expect in 2024?
In this episode, I'll explore three trends that I believe will make AI even more useful and relevant than it has been in Twenty twenty three.
Stay tuned until the end where I'll debate whether one of the AI innovations threatens the virtual assistant industry or not.
I'll unpack all of this after the intro.
Welcome
If this is the first episode that you're listening to, welcome to the Macpreneur tribe, and if you're a longtime Macpreneur listener, thank you for tuning back in. As a fellow solopreneur, I appreciate that you dedicate these 15-ish minutes with me every week. Over the past few months, I've had a chance to interact with other Macpreneurs, and some of them expressed interest in being able to connect and discuss with other solopreneurs who run their businesses on their Mac.
Now, before launching a Macpreneur community, I would like to be sure that enough of you actually want that. So if this idea sounds interesting, then head on to macpreneur.com/community, where you will be able to join the waitlist.
Just enter your name, email address, and pick all the online platforms that you prefer using.
And so to get the chance to become one of the founding members of the Macpreneur community, visit macpreneur.com/community.
Let's talk about AI developments that I see for 2024. There are three main areas. The first one is multimodality, the second one is system 2 thinking, and the third one is AI agents.
Multimodality
So multi modality, what does it mean?
It means that a model would be able to accept multiple types of input and then generate multiple types of output.
So for instance, ChatGPT 4 can already do that.
You can speak, you can give it an image, you can type, and then as the output, it can produce text, it can produce images, and it can respond via a synthetic voice.
Runway Gen 2 can also do that.
So with Runway Gen 2, you can create a video, but with different kinds of inputs. So text only, text plus image, text plus video, and so on.
My prediction is that most commercial AI systems will need to be multi modal if they want to survive.
System 2 thinking
The second one is System 2 Thinking.
So in 2011, a psychologist named Daniel Kahneman published a book entitled "Thinking Fast and Slow".
It's really a fascinating book. If you have never read it, I can only encourage you to do so.
So it's a book that explores how our brain works. And it is based on scientific evidence.
So in a nutshell, our brain has two modes of operation, system one and system two.
System one is fast, automatic, intuitive. So imagine you're driving on a completely empty motorway. You don't even need to think about driving. You can listen to podcasts, you can brainstorm.
Now, System 2 is much slower, logical, deliberative, and this is what you will engage if you now drive through a very busy intersection during rush hours. I know that myself, when I'm in that situation, Even though the podcast continues to play, I absolutely cannot remember whatever was, uh, said in the podcast because in that intersection, my attention is fully, on the traffic.
So now if, if we go back to AI and especially large language models,
Up to now, those have been operating in a state equivalent to system 1. So they predict the next token or the next fraction of a word as fast as possible.
But some researchers have been working on special prompting techniques in such a way that the large language model can think its responses step by step rather than in one shot and what they've noticed is that the large language model was producing much more accurate responses. And they call that technique System 2 Attention.
And so my prediction is that future large language models will be trained or programmed to use System two attention, maybe not all the time, but maybe as a switch or as a parameter that we will be able to configure.
AI Agents
And then the last development that I see that will mature during 2024 are AI agents.
So the way that we interact with AI right now. is either through a single browser window or an application, but all the interactions stay within that environment. And even though some large language models, they can search the web, they can't interact with other app windows, whether it's a browser tab or a separate application.
Now, imagine asking an AI system to go on LinkedIn. Search for 10 people who would fit your ideal customer avatar profile, follow them, and then fill a Google spreadsheet with their name, their title, and a link to their profile.
Well, that's exactly the type of task that an AI agent would do. And the thing is, such AI agents These already exist today, however, they're limited to interacting only with browser windows at the moment.
There are three AI agents that I discovered, and two of which that I've already played with a little bit.
So the first one is called HyperWrite Personal Assistant. Then the second one is AI Employee, and the third one is Adept ACT-1.
HyperWrite Personal Assistant
So, HyperWrite Personal Assistant works as a Chrome extension, and it can be triggered from any web page once the extension is installed in Chrome. It can be used to book a flight, to organize the Gmail inbox, and any other task that can be done through a browser. The free plan is limited in the number of actions that an AI agent can perform for a given task.
Now, I asked HyperWrite to find 5 online communities for solopreneurs, and it ran out of 3 actions before even completing the task.
What I noticed is that it's still quite slow, and also HyperWrite warned me that I'm responsible for all the actions. taken by the AI agent, meaning that I couldn't really give it a task than go to the kitchen and refill my cup of tea.
Instead, I stayed at my computer to double check what it was doing.
So it's still early days and I'm hopeful that it will get better in the future.
Now, on a side note, HyperWrite is developing another AI agent that can run on a Mac and can interact with any application window, promising some crazy automation that currently requires a human to do.
AI Employe
Another one that I recently discovered is AIEmployee. It's another Chrome extension that leverages GPT 4 Vision, the model from OpenAI that can recognize objects inside images. Now, AIEmployee works differently than HyperWrite, so rather than giving it a prompt, you need to configure workflows. step by step.
And so like a human, the first time that it needs to do a task, you will have to train it. But once it has been trained, you just click on the workflow that you want to trigger and it will execute.
Now some of the examples that the developer uses to demonstrate the capabilities of AI Employee could actually be done with more traditional methods like Zapier, for instance, with the advantage with Zapier that everything happens automatically and in the background.
And yeah, AI employee is pretty slow, but it's even worse than that because once a workflow has been initiated, you can't really use your computer because the AI employee needs to be able to see the browser window at all times.
In other words, I'm not super impressed, but again, it's early days and it's worth monitoring in my opinion.
Adept ACT-1
The last AI agent tool that I found is ACT1 by the company called ADEPT.
It's similar to HyperWrite Personal Assistant in a way that we prompt it, and it seems to use the same kind of Chrome technology that AI employee uses. I haven't been able to test it yet, but I joined a waitlist and fingers crossed I will be able to evaluate it in the coming month.
AI Agents: Threat or opportunity for VAs?
Now, I'm not entirely sure if AI agents are a threat or an opportunity to Virtual Assistants.
At this point, I would say both because AI agents will be a much cheaper alternative than VAs. So for us, solopreneurs could be a good alternative, but at the same time, imagine a virtual assistant leveraging AI agents, that virtual assistant will have superpowers and will be able to do much more in less time.
Recap
To recap, I've shared with you the most promising AI capabilities that I believe will be further developed in the next 12 months.
So, multimodality, system 2 thinking, and AI agents.
If you found this episode useful, I'd be super grateful if you could share it and if you do that on Instagram, you can tag me, my handle is MacpreneurFM.
Next and outro
In the next episode, I will explain the three reasons why, contrary to popular belief, Apple is extremely well positioned with regard to the AI race.
So that's it for today. If you haven't done it yet, visit macpreneur.com/ai to grab your own copy of the top 10 AI tools cheat sheet.
This PDF will give you the edge that you need to boost your solo business in this fast-paced world.
Once again, it's macpreneur.com/ai
And until next time, I'm Damien Schreurs, wishing you a great day.