Written by Signal Theory Technology & Innovation Director Elijah Kleinsmith
In 2023, I showcased a new AI tool every weekday in May. Since then, we have been integrating the most powerful tools from that series into our workflow here at Signal Theory. Here are the top three things I have learned along the way and plan to leverage in 2024.
1. Multimodal AI will be behind the most disruptive tech of 2024.
Multimodal AI is a type of AI that can input and output multiple data formats, including photo, video, audio and text. I like to think of it as akin to our human senses – I can touch and feel, speak and listen, read and write, and so on. Similarly, AI can now understand photos as well as create them, listen and respond audibly – input any modality, output any modality.
Perhaps the most visceral experience I have had with multimodal AI was toward the tail end of 2023 when I decided to pick up some Ray-Ban Meta smart glasses as an early Christmas present to myself. I wore them on a road trip across Kansas with a buddy of mine and the experience was nothing short of existential.
We were driving his Jeep near the Colorado border when we encountered snow. We weren’t sure if you could change the traction control system while the vehicle was in motion, so we simply asked Meta. Having so much knowledge, this readily available, got us thinking a bit deeper about the technology.
“Hey Meta, explain the theory of mind to me.”
“The theory of mind, a fundamental aspect of social cognition, is our capacity to understand others as thinking beings with subjective experiences and mental states different from our own.”
I think back to this January 2023 tweet from tech entrepreneur Naval Ravikant. Now, nearly a year later to the day, imagine a “calculator” for not just reading and writing, but also audio, code, image and video production. The Ray-Ban Meta smart glasses are now testing multimodality, allowing the AI to view through the on-board camera and assist with whatever the person wearing them happens to be looking at.
When I was in high school, there was a lively debate about the value of everyone learning math if we all carry a calculator in our pocket. Now, one has to wonder what the value of human intelligence is if we all carry the sum of all human knowledge on our sunglasses or headphones. It reminds me of this ominous tweet from the Chief Scientist at OpenAI.
In 2024, it will be critical to take a deep look at “what are our outputs and what are our inputs” across all segments of your organization to remain competitive in a world that is working to automate cognition.
2. AI is better at creativity than math.
Perhaps the most surprising thing to me about the past year has been discovering what kinds of tasks large language models (LLMs) excel at. It is also true that this new class of AI fails certain kinds of tasks more than others. But the fascinating part is that LLMs are better equipped to be creative than factual.
Traditional sci-fi predictions about AI would have it such that machines would excel at deterministic tasks far before they uncover the ability to reason with humans on an emotional level. Turns out, this couldn’t be further from the truth. When looking at which task automations would benefit from the integration of an AI sublayer, we have found far more success utilizing AI for non-deterministic tasks (analyzing qualitative data, generating proof of concepts, brainstorming and providing strategic insight, for example) than we have for deterministic tasks (time tracking, financial analysis and proofreading).
In fact, a new study from the University of Montana echoes this experience. This study used the Torrance Tests of Creative Thinking, a gold standard for measuring creative talent. The big takeaway? AI is holding its own against the top 1% of human creative minds. This revelation isn’t just a technological leap; it’s a profound shift in understanding the potential of AI in fields driven by creativity. Meanwhile, GPT-4 only scores in the 89th percentile in SAT Math.
I believe the importance of this point is undervalued at the moment. Many people still seek AI solutions as a sort of nebulous catchall to the mundane work we don’t want to do when in fact, the opposite is true. AI promises to, at the very least, disrupt how we approach our most creative, human-feeling work.
3. Looking for the most bang for your buck? Start with the basics.
Admittedly, I have a bias toward experimentation – especially when it comes to technology adoption. But what has surprised me above all else when it comes to AI adoption is how many people did not try ChatGPT Plus in 2023. If that’s you, I plead with you to try it in 2024. For the cost of a Netflix subscription ($20/month at the time of publication), you can have the most powerful AI assistant available to humanity at your fingertips.
The most common question I get is, “What does ChatGPT Plus have that the free tier does not?”
- Plugins, which allow ChatGPT to use 3rd party software.
- File upload, which allows you to upload files for analysis.
- DALL•E, which allows you to create incredible images with ChatGPT.
- Custom GPTs, which is OpenAI’s first step toward introducing AI agents.
Even ignoring those benefits, the most compelling reason to me is that GPT-4 is much smarter than GPT-3.5. Wouldn’t you pay $20/month for a smarter assistant if given the option?
The other AI tools I experimented with in May 2023 were interesting, but none as useful as ChatGPT Plus and the OpenAI API. Lost in the AI hype is an underlying truth behind what makes these tools work. Conceptually, I like to think of this as a triangle, each interconnected point representing an important piece:
- The LLM: Want to ensure your AI tool is excellent at reasoning and creativity? Use a top tier LLM like OpenAI’s GPT-4 or invest in fine-tuning your own open source model like Mistral.
- Embeddings: Want your AI tool to reference your other work or a knowledge base of some kind? Ensure your LLM has access to the knowledge and tools you need through embeddings.
- Prompting: Want to refine how your AI responds to the audience it is intended to help? Work on your prompt engineering.
At Signal Theory, we’ve been experimenting with every part of this triangle to produce some really powerful tools that assist with creativity, content creation and building strategic bridges across disciplines. If you have an interest in learning what AI can do for your business’s marketing efforts, reach out to me at elijah.kleinsmith@signaltheory.com – I’d love to show you a demo of some of our existing tools and talk about what the future holds for advertising technology.
