Anthropic’s Claude launches a new AI model, Claude 3.5. Perplexity.ai reports:

Claude 3.5 Sonnet sets new industry benchmarks, outperforming competitor models like OpenAI’s GPT-4o and Google’s Gemini 1.5 Pro, as well as Anthropic’s own Claude 3 Opus, across various evaluations. source

Update 6/25/2024

You may want to check out Claude.ai’s latest announcement on this date about Claude.ai Projects, their version of ChatGPT’s Custom GPT.

Perplexity, You, and other AI middlewares rapidly reported they can channel the new model. But why is that important? Claude is no clod, it is a lot faster and has more capabilities now.

You can see the new Artifacts feature in this post. After you enable it, you'll see an experimental feature has been enabled.

AI Can’t Read Handwriting?

Someone on Facebook, lauding how wonderful cursive writing is for the brain and how AI cannot read it (which is inaccurate) had someone else ask if a “nerdy” person could test that assertion about AI being unable to read handwriting.

As much as I love handwriting, and using cursive now that I had time to teach myself it again, I didn’t think that assertion about AI was right. Two seconds later, I knew it was not.

An Experiment

For fun, I took a page of my handwritten notes (shown above) then dropped them into Claude 3.5 with the prompt appearing below:

PROMPT: Provide the text in the handwritten document attached verbatim. Create a mermaid chart showing key ideas.

The result? Pretty cool. Before when Claude was asked to provide diagrams, it could not. Now, it’s able to render the code in a snap:

What’s more, you can see the code, too:

To edit the code, you’ll have to tell Claude what you want change, such as:

Make graph LR (left to right) rather than TD (top down)

Here’s what that looks like:

Pretty amazing. I’ll have to try a few other things out with Claude.ai, and check to see if Perplexity.ai CAN really channel Claude 3.5 or not. Oh, and I forgot to attribute what book my notes are from…

References