The explosive growth of Generative AI in 2024 has transformed it from a specialised tool into a mainstream force. While ChatGPT may have captured public attention, the landscape has evolved far beyond simple text generation. This piece explores my journey from basic AI usage to understanding its broader implications for technology professionals.

Breadth of Field

While OpenAI's ChatGPT may have the greatest name recognition, I was surprised by the breadth of organisations investing heavily in this area.

According to Simon Willison's comprehensive "Things we learned about LLMs in 2024", while OpenAI's GPT-4 entered 2024 as the stand-out leader in standard industry benchmarks, that dominance proved short-lived. By year's end, 18 models from 11 different organisations had surpassed GPT-4's performance metrics.

As the field expands and models become more capable, the art of effectively communicating with these systems becomes increasingly crucial. My experience reflects a common evolution in approach.

This evolution in communication strategy is particularly important given that model performance isn't just about raw capabilities - it's about how effectively we can harness them. The quality of outputs directly correlates with the sophistication of our inputs, leading to a fundamental principle: "Garbage In, Garbage Out."

Garbage In, Garbage Out

To get the most out of any model, prompts need to be as detailed as possible. Previously, I had erred on the side of caution and kept them brief to avoid misunderstandings. However, take a look at this example of a ‘Good’ prompt from Anthropic’s Claude prompting guide.

You are a fabric supplier for my backpack manufacturing company. I'm preparing for a negotiation with this supplier to reduce prices by 10%. As the supplier, please provide:

1. Three potential objections to our request for a price reduction
2. For each objection, suggest a counterargument from my perspective
3. Two alternative proposals the supplier might offer instead of a straight price cut

Then, switch roles and provide advice on how I, as the buyer, can best approach this negotiation to achieve our goal.

Anthropic's Prompt Engineering documentation is a great place to start asking better questions of the these models.

While mastering prompt engineering is crucial for text-based interactions, the landscape of AI capabilities has expanded far beyond pure text which requires us to think about prompts in increasingly sophisticated ways.

Moving beyond text

2024 marked the tipping point for multi-modal AI (systems capable of processing multiple types of input like text, images, and sound) as these systems evolved from interesting experiments into practical tools that could seamlessly work with different types of content simultaneously.

This came to the masses with the introduction of Visual Intelligence as part of Apple Intelligence; users can take a photo of an object, ask questions about it verbally, and receive information drawing from both visual and contextual understanding. This is a big step in broadening the adoption of AI.

Awareness

Ben Evan’s “AI eats the world” neatly summarises how there’s fairly widespread awareness of these tools but a much smaller proportion of people have actually experimented with the tool and no country has more than 10% of people where this has become a core tool used each day. As these features become tighter integrated into the platforms where people spend their time, be it Gemini for Google Workspace, Co-pilot for Microsoft Office, GitHub Copilot or the iOS Camera app, this proportion will only increase and it will be fascinating to see what emerges as the ‘must have’ use cases.

As we observe these patterns of innovation, adoption, and integration, several key themes emerge that help us understand both the current state and future trajectory of AI technology.

Conclusion

The landscape of Generative AI in 2024 reveals a fascinating paradox: unprecedented investment coupled with uncertain practical applications. The comprehensive displacement of GPT-4 from its leadership position demonstrates both the intensity of competition and the accelerating pace of innovation. This proliferation of capable models, combined with declining training costs and the democratisation of access through integrated platforms, creates fertile ground for experimentation and discovery.

Looking ahead to 2025, the combination of improved multi-modal capabilities and broader platform integration creates conditions for rapid innovation. While the "killer app" remains elusive, the sheer scale of investment and increasing accessibility suggests that breakthrough applications will likely emerge through practical experimentation rather than theoretical planning.