The Evolving AI Landscape: Which Model Should You Use in (July) 2024?

For the past year, one of the most common questions I've received is, "Which AI should I use?" My answer has consistently been to use the most intelligent model available, as I believe the intelligence of the model is the key feature for AI applications. Until recently, this meant recommending OpenAI's state-of-the-art models, whether it was GPT-4, GPT-4 Turbo, or the latest GPT-4o.

While Gemini Pro briefly challenged this recommendation, OpenAI consistently managed to regain the top spot, maintaining their competitive edge. However, the AI landscape is rapidly evolving, and there's a new contender for the title of the most advanced AI model.

Anthropic's here to say hi!

Anthropic, founded by former OpenAI employees, has been quietly and steadily developing their own AI models under the name Claude. Their focus has been on creating more professional and "cleaner" tools, with incremental updates that have kept pace with OpenAI's offerings.

A week ago, Anthropic made a significant move by releasing Claude 3.5 Sonnet. For those unfamiliar with Anthropic's lineup, they offer three tiers of models:

1. Haiku: Fast and cost-effective
2. Sonnet: Mid-range capabilities
3. Opus: High intelligence, premium pricing

The release of Claude 3.5 Sonnet marked a serious leap in intelligence and capability, with Anthropic claiming it surpasses OpenAI's GPT-4o, the previous state-of-the-art model.

Hands on with Claude 3.5 Sonnet

I've been using Claude 3.5 Sonnet alongside GPT-4o, and I can confirm that it does seem to be a better tool for professional applications. While many in the industry are praising its strength as a code development tool, I've primarily been using it to support cognitive work, including:

• Reading and analysing transcripts
• Developing documents
• Checking work
• Building professional tools

Claude has always boasted a very large context window, which is really useful when working with big documents and complex contexts. Previously, it struggled to understand instructions as well as GPT-4. However, this limitation has been addressed, and Sonnet has now become my go-to tool for cognitive work.

Oh no, things have changed again!

It's important to note that in the world of AI, things can change overnight. This underscores the necessity of continually testing various tools to evaluate how they perform for your specific use cases if you want to leverage the best artificial intelligence available.

Can I kick it, what features does it have?

When choosing an AI model, it's crucial to consider additional usability features:

1. ChatGPT's Mac app remains my preferred interface, offering:
   • Built-in transcription capture
   • Ability to import screenshots
   • Web search functionality
2. Anthropic has introduced some compelling features:
   • Projects (Claude's version of GPTs, allowing customisable single-purpose chats)
   • New artefacts window (enabling AI to generate outputs outside the chat window – a brilliant feature likely to be emulated by other models soon)

These advancements have made the choice of AI more complex and nuanced.

Recommendations (Today)

As of now, I would still recommend GPT-4o for most users due to its broader skill set and the likelihood of an imminent model update. However, for those with the budget, I strongly encourage trying Claude 3.5 Sonnet.

For users keen on staying up-to-date with the latest models, Poe.com offers access to various AI models, although it may lack some interface features found in dedicated platforms.

So this is a good thing?

The intense competition in the AI space is driving rapid improvements in intelligence, usefulness, and features – a trend that ultimately benefits us as consumers. While the current leader may change, the overall trajectory is towards more capable and versatile AI assistants.

As we feel our way in the evolving AI landscape, it's important to remain adaptable and open to exploring new tools. The "best" AI for you will depend on your specific needs, budget, and use cases. Stay informed, keep testing, and don't hesitate to switch models as the technology advances.