Can Claude Writing Styles Write Like Me? (I Tried It)
Anthropic also released a new feature yesterday, Writing Styles. Can it write like me?
I wrote a newsletter earlier this week that was centered around my frustration with Claude. Most days, the web version either limits me to a few messages or craps out completely after one prompt.
It's interesting that, according to Reddit users, Anthropic has just removed support for Claude 3.5 Sonnet on its free plan. This hasn't been announced as far as I can tell, but it seems free users can only access Haiku now.
Perhaps this will reduce the number of outages and message limits that have been making Claude so difficult to use.
Around the same time, Anthropic released a new feature called Writing Styles.
You can give Claude an example of something you've written and it will try to write like you.
I would never try to use AI to write for me. Where's the fun in that?
But Claude has a good understanding of language, so this sounded like it might be a good shortcut to a first draft.
We started off well, and Claude was very complimentary about my latest Substack newsletter.
But when I used the option to generate a blog post, the output was nothing like the newsletter.
And what is it with LLMs and the word "landscape"?! Argh.
The result is a real word salad that doesn't sound anything like my newsletter.
You can fine-tune the output in a very basic fashion by clicking Edit With Claude. But it's just a prompt. You can't upload additional samples.
I guess I'm more fussy about language than the average user, but I won't be using this feature. It would have been easy to make it better. If we could upload a few examples, I'm certain the results would be closer to the style in the source files.
Fortunately, you can do something like this with a Typing Mind agent.
I recently set up an agent that's trained on multiple examples of content written by the same person.
You can also adjust the system prompt, enable web search, and turn on artifacts for your agent.
I trained it on about 10 examples and gave it a prompt that clearly outlined what I wanted it to focus on. I had to adjust the system prompt quite a bit to avoid over-compensation on things like structure.
The results are not good enough to publish as-is. But it can be helpful. For example, I have used this agent to expand a content brief and flesh it out.
A more detailed brief gives writers more context and support than a traditional brief, which is usually a bullet-point list.
Reminder: Typing Mind is having a Black Friday sale. I recommend that you check it out, and I'm not getting any kind of kickback or commission. It's genuinely great, and it's a lifetime deal, so no ongoing costs except for the API(s) you want to use.