Claude and Llama: Why should you care about another AI chatbot?
John Swaffield
Digital & Multi-Channel Marketing Manager at Taking Care, part of AXA Health
In July 2023, AI startup Anthropic released Claude 2 in the UK and US and Meta and Microsoft announced Llama 2.
So why should you care if you already have access to ChatGPT or Google Bard??
At first glance, Claude 2 is similar to OpenAI’s ChatGPT and Google Bard. All three respond to natural language prompts. However, Claude 2 can process the equivalent of 75,000 words in a prompt (100,000 tokens in AI-speak), meaning you could upload the entire Stephen King novel Carrie as a prompt. In comparison, Bard’s token limit is around 1,000 whilst ChatGPT 3.5’s is 4,096.?
That alone means Claude 2 is potentially incredibly powerful and indeed does well for document analysis and summarising long content (more of that later).?
Zuckenberg sees AI as critical to Meta’s growth and by releasing Llama 2 as open source, is weakening the competitive advantage of OpenAI and Anthropic. Any organisation can freely download Llama, build upon it and develop it to improve or extend its functionality. Llama could be the disruptor and expect to see it integrated in devices and services in the future. Llama 2 is trained on 2,048 tokens but as it is open source, it could be developed to work on more or less.
You can try out Llama 2 at Labs.Perplexity.ai without having to be a developer and download the source code.
Claude and Llama matter because they drive innovation and competition at Google and OpenAI. With Microsoft the main investor in OpenAI and Meta’s preferred partner, Microsoft are hedging their bets by working with both.
Here’s how I tested Claude 2 and Llama 2’s ability to analyse documents as a comparison test against Bard and ChatGPT.
Summarising long content
Semrush have released its “Future of Ecommerce 2023” report. It's full of insights but is 53 pages long. Why not use AI to summarise the key points to save time reading the whole thing and test the three main AI platforms at the same time?
Here’s how the three AI platforms performed:
Bard...epic fail
Bard is not able to process PDFs (either uploaded or the hosted URL). When asked to summarise the key points from a URL of the PDF, the result was: “I'm not programmed to assist with that.”
ChatGPT 2.0 with Code Interpreter plugin...disappointing
ChatGPT performed surprisingly poorly, particularly as I used the Code Interpreter plugin with the paid for 2.0 version. It stated “Unfortunately, the text extraction only covers the first part of the report due to the limitations of the current text extraction method. To generate a comprehensive summary, I would need to be able to parse the whole document.”?
It then attempted to summarise the report but only managed 5 points (and one of those was an assumption):
Claude 2...easily the best result so far
Claude parsed the whole report, understood it was on the future of ecommerce and summarised the document in 8 points. Pretty impressive!?
Here's the summary from Claude 2:
Llama 2 - the best performer
Llama quickly and accurately summarised the report as providing “insights into the current state of e-commerce, including trends, consumer behavior, and industry performance”.
It listed 10 key takeaway and pulled out the most relevant and insightful points better than Claude:
Overall, the report suggests that e-commerce continues to experience rapid growth and evolution, driven by changing consumer behaviors, advances in technology, and the rise of new markets.
Free to use - so why not try Claude 2 or Llama 2?
In my simple test, Llama came out on top when asked to summarise long content. Other reports suggest Llama is as good as ChatGTP 3.5 for content and idea generation. However, it has been released as an open source model for developers so unless you are a developer, you will be relying on free websites that have deployed it. Claude, ChatGPT and Bard are easier to try out.
Like Bard and ChatGPT 3.5, Claude 2 is free to use. All you need to do is sign-up with an email address or Google account. You don't need to be technical or understand AI to use these tools. You just need to be curious!
For the full power of ChatGPT, version 4 costs $20 per month. ChatGPT 3.5 is free but doesn’t have the power of plugins and the code interpreter.
They are all better than Bard - unless you need up-to-date research
Bard can access the internet to retrieve information. This gives it a big advantage over Claude 2, Llama 2 and ChatGPT which cannot and therefore their knowledge is limited to the data they have been trained on (ChatGPT 4 was able to access the internet until recently, when OpenAI removed the featured, apparently because people were using it to access information behind paywalls).?
However, in my experience Bard is more prone to making up responses (“AI hallucinations”) than Claude 2 or ChatGPT and less likely to give you the response you really want. Bard feels like work in progress, whilst Claude 2 is more complete and therefore a good alternative to ChatGPT for some tasks.?
What's good for today may not be good for tomorrow
OpenAI, Anthropic, Meta and Google are constantly working on their AI solutions. Claude and Llama matter if for no other reason than it offers competition to Open AI and Google.
With each new release, the features and capabilities of AI chatbots change and improve. Whilst Llama and Claude outperformed ChatGPT and Bard at summarising long documents today, that may change tomorrow.
Who knows, maybe Google will finally come good with Bard so its worth experimenting with a couple of platforms rather than sticking to one.