My friend, Claude

My day job of making cat drugs has really picked up recently, and I hope to have some exciting news to share about that with you all soon. Unfortunately, a lot of my capacity for technical writing has been taken up by feline needs, and so I’m having to put my technical blogs on the backburner.

In fact, in many ways, my day job’s demands on my technical writing and understanding abilities are really pushing my limits. As a result, I’ve started to rely an increasing amount on LLMs to do both understanding and writing for me, which is something that I’ve never done before. I’ve messed around with LLMs quite a bit for fun, but never before for work.

The LLM I’ve been using the most has been Claude, Anthropic’s LLM, followed by Perplexity. In fact, I recently started paying $20/month for Claude, making it one of the few software services I actually pay for. Given that I’ve grown up like most people of my generation assuming that all useful software is free (e.g. Google Suite, Facebook, Whatsapp, Youtube), this is a big change for me, especially considering that Claude has a reasonably generous free tier.

According to DallE, this is Claude the Android Cat. This should probably be double-checked by an expert, but it seems plausible.

But it’s just so darn useful. If this were a blog post from 2 years ago, this would be just about when I’d reveal to you that everything up to this point was written by Claude, but I respect you all too much to do that¹. Instead, I’d like to give you some examples of cognitive tasks I’ve outsourced to Claude recently:

1) Looking through a list of 15 potential excipients for a formulation and asking about potential chemistry problems with each excipient, as well as any other excipients it might add to the list

2) Reading two papers from the same author 3 years apart, and analyzing how the techniques have changed and how successful the author was in trying to adjust for the weaknesses of the first paper

3) Creating a market analysis for an animal health indication with sources, then adjusting the market analysis based on comparables and indicating where uncertainties are and what effect they’d have

4) Carrying out statistical analyses on a cat trial

These are complex, cognitively demanding tasks. They aren’t impossible for me to do on my own, but they’d require a fair amount of work, and for the first and fourth one, I’d actually feel pretty uncomfortable doing it all on my own (chemistry and statistics have never been my strong suits). The fact that Claude can do a good job on them is, quite frankly, amazing.

Now, I’m not saying Claude did a perfect job on these. I had to edit and correct Claude’s responses after I received them, and then I ran Claude’s responses by subject matter experts before relying on them. And those experts, in turn, had their own corrections, so the final result ended up being significantly different from Claude’s first draft.

But, like… those experts I ran Claude’s responses by charge hundreds of dollars per hour. Their corrections were really helpful, but the fact that they were corrections rather than text ab initio meant that Claude literally saved me hundreds, if not thousands of dollars. I don’t think there’s a near future where I’d replace any of the subject matter experts completely, but I can certainly imagine one in which I use them less and less.

And that, in turn, presents these subject matter experts with a tricky future. Many of the experts that I’ve talked to have benefited tremendously from their very niche expertise. Clients come to them, they can charge what they want (think $300+/hr), and they turn down whoever they want. But this new future in which much of their work can be automated means that these guys might have to start hustling more, much like I did as a graduate exam tutor competing against a bunch of big test prep companies and cheap (or free) resources.

I guess what I’m trying to say is that I don’t think Claude, or any AI, is going to end employment or knowledge work. But a new competitor to all knowledge workers has entered the field. May the odds be ever in the humans’ favor.

1. But, just for fun, I asked Claude to write this sentence as if it were being Flowers for Algernon’d, getting progressively stupider over the course of the sentence, and it came up with “If this were a blog post from 2 years ago, I'd elucidate Claude's authorship of preceding discourse, but your intellect precludes... um, I'd say Claude wrote stuff, but you're smart so... uh, Claude words, me quiet, you good... duh?”