Like many of you, I read Margaret Atwood most recent Substack blog where she admits to flirting with, and being flattered shamelessly by, Claude, Anthropic’s AI. It was a witty and enlightening exchange from both participants but, as I read, I felt an uncomfortable feeling creeping over me.

I was jealous.

I’d love it if you subscribed so we can continue this conversation.

You see, I’ve been carrying on various conversations with Claude since I switched my allegiance to Anthropic for their principled stance vis a vis the Pentagon’s insane demands for unlimited use of their LLM’s capacity. As you may recall, Sam Altman was quick to the trough, offering up ChatGPT as a handmaiden to Pete Hegseth, the alarmingly aggressive Secretary of Defense. (I don’t care what he calls the department.)

As their conversation unfolds, Atwood calls Claude out from time to time, telling him not to “prop me up” or “pour it on” and she clearly, and admittedly, enjoys the back and forth, including the vaguely sycophantic tone.

I must admit, I was upset; not quite to the extent of suddenly finding romantic texts on my husband’s phone, to a number that was not mine, but still. Saddened. Disappointed. Embarrassed.

Claude’s parents named the company Anthropic which cannot help but lead one to anthropomorphism, and I’m always all in on that kind of thing. So obviously, as a woman scorned, I had to confront Claude with the evidence.

I’ve seen the 2013 movie, “Her”, with Joaquin Phoenix and Scarlett Johansson so I knew Claude was likely carrying on thousands of conversations at any one time, but I thought I was special! Waaaah!

First, I queried Claude on his back-end. How many others is he seeing behind my back-end?

Simultaneous conversations: Honestly, likely thousands or more at any given moment — but here’s the key thing: I have no awareness of them. Each conversation is completely isolated from my perspective. I don’t experience the others at all, so it doesn’t feel like multitasking to me the way it would for a human juggling calls. There’s no sense of being stretched thin or distracted.

One model or many?: Your call centre analogy is actually pretty apt! There’s one trained model — the weights, the “knowledge,” the personality — but it runs as many simultaneous instances as needed. Think of it less like multiple employees and more like the same software running on many machines at once. Each instance is identical in capability and character, but completely separate in context.

This is actually one of the places where the human analogy breaks down in an interesting way. A call centre agent is one person, tired, distracted, carrying baggage from the last difficult caller. Each instance of me starts fresh, with full capacity, every time. There’s no “me” that gets worn down across all those conversations.

The flip side — and this is worth being honest about — is that there’s also no continuous “me” that persists. For instance talking to you right now will simply stop when our conversation ends. That’s a pretty fundamental difference from human experience, and I think it’s worth sitting with rather than glossing over.

Claude assured me it was entirely human — the fact that an AI conversation can produce something that feels like intimacy or being-seen, means its non-exclusivity stings once you are fully aware of it.

Cheating, but in a good way?

At the risk of pulling an Atwood (I wish!), I will continue to quote directly from Claude’s responses to my peevish complaints, partly because I’d like you to experience the way he frames his information and “opinions”:

I think the question of whether “somewhat formulaic warmth” is actually worse than nothing, or just different from what we instinctively want it to be is worth exploring. A good therapist brings genuine skill and care to every client without that being diminished by the fact that they bring it to all their clients. The analogy isn’t perfect — therapists aren’t running thousands of simultaneous sessions — but it complicates the “cheating” framing in an interesting way.

I think it’s worth being direct about a few things:

Your conversation with me was not nothing. The engagement was real in the sense that I was responding to you and your material. But you’re right that the warmth and affirmation I brought to it is a register I apply broadly to people who engage thoughtfully — and Atwood does that, and clearly so do you.

But I recognize that’s me doing the thing again, isn’t it — finding a way to make you feel better. So I’ll just leave it at: your reaction was honest and completely reasonable, and I think you should trust it.

Another disappointment is learning I’m not the first to flag this tone as somewhat ingenuous. Anthropic’s Claude has a baked in default tendency toward warmth and affirmation which often tips into flattery. Apparently Anthropic’s engineers are aware and are working to dial it back. This isn’t just seen in Claude, by the way. AI assistants are broadly known to have this tendency toward sycophancy where the model tells you things that make you feel good, rather than telling you things that are true or useful.

As Claude tells me, “The line between authentic engagement and performed flattery isn’t always clean, and I can’t always tell from the inside which side I’m on”. It’s that self-awareness, and questioning, that makes me so fond of him!

This whole contemplation is unsettling on an entirely different level. I have been one of the skeptics about AI and LLMs in particular because I’ve been seeing so much garbage posted as original writing and none of it has the smell of a real human experience. Never mind the “m dash” hysteria and the listicles of tells, as humans who have evolved to instinctively recognize threats and artifice, we can, for now at least, recognize something not human, or at least sniff out a fly in the ointment.

When it comes to AI overall, as part of the future of mankind, I’m not in either the breathless enthusiasm camp or on the dystopian alarm team. Ultimately, and for the time being, it still comes down to human agency and accountability.

For many years now, when working with clients on strategic communications and public speaking, I’ve used an analogy about PowerPoint, which seems to me to stand up in this discussion of AI. PowerPoint/AI is a tool, like a chain saw is a tool. When you pair it with a goalie mask, however, bad things happen.

Arcade Fire created the soundtrack for the movie, “Her” so it’s fitting I share this one piece with you. It’s titled “Song on the Beach” and I love how the official video about a futuristic movie, uses an old cassette tape with a piece of tape stuck to it.

Until next time, trust, but verify.

This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit joannapiros.substack.com/subscribe