The number of screenshots of Claude conversations is going up in my life and it’s beginning to have an impact on my general mood. Most of the time it’s well-intended; coworkers working through a problem with a chatbot before bothering me or someone exploring unconventional ideas before bringing it to a broader audience. Both of those situations seem considerate. A tool for thought, as it were.

It’s probably good practice to do your own research before you rope in another person on a cognitive task. But a small thread in the reliability of these tools-for-thought begins to unwind due to the well-documented sycophantic nature of engagement-thirsty language models. One of my favorite studies is Anthropic’s own study (2023), when asked to review an argument with I wrote this [...], the LLM gave positive feedback. But if you start a new chat with I didn't write this [...] or give any hint of your own personal bias, it provides much more critical feedback. And the ultimate twist is, we prefer models that are super nice to us.

I was telling my son’s friend about this phenomenon and he responded with the perfect GenAlpha summation:

AI is a D1 glazer, bro.

We need to acknowledge that we’re probably getting the answer we want rather than a cold-hard fact. Not to get too serious but when I read about AI psychosis, I think the overly-confident “You’re a genius” style of reply is the point where it all starts to go wrong.

Awhile back Hidde De Vries identified a pain point around LLM-usage in standards work which leads to something I call an asymmetry of thought. In a conversation where one person is a domain expert and one person is copy-pasting ChatGPT responses, it creates an imbalance of effort in the discussion. A second-hand burdening, like Brandolini’s Law, where debunking inaccuracies and subtly-wrongs takes more effort than creating the inaccuracies. We are, in effect, taxing experts to do quality assurance on the model’s responses. When that work is unpaid I believe this is immoral… or at least it’s a breach in social etiquette.

In situations like that where people are using models as a form of appealing to authority, unless both parties have agreed that the LLM is a neutral arbiter, a random noise generator has the same amount of social authority.

So, when a screenshot or copy-pasted block of “Here’s what Claude said…” comes across my screen… I don’t care about the screenshot. I don’t care about the screenshot because I want your thoughts, not Claude’s. I want your unfiltered and half-baked ideas, not Claude’s synthesized extrusions. Instead of a screenshot of a reply, I’d rather have the original prompt so I can just ask the machine myself. At least then I’d know the context you provided the machine and your understanding of the problem, because that matters immensely.

A trite example that I deal with from time to time, if you ask an LLM “React or Web Components?” it will say “React” because that’s what was popular in the 2023 training data. But did you mention anything about different teams on different tech stacks? Anything about hitting memory ceilings? Context is everything to these machines and –as the disclaimers say– they can make mistakes. If we need anything more than an approximation, a language model might not be the right tool for the job.

Anyways, that’s my issue with other people’s chatbots weaving their way into my daily conversations. I’d rather have a human-to-human conversation with you, not a chat with Claude by proxy. What Claude said is an okay chunk of “anecdata”, but it’s not a substitute for our working relationship.