humancode.us

AI should not simulate real humans

September 6, 2025

I was reminded of the “Booby Trap” episode of Star Trek: TNG where Geordie (accidentally) gets the computer to simulate Dr. Leah Brahms, the inventor of the Enterprise’s warp engine, inside the holodeck to help him solve their conundrum. Geordie decides to imbue the simulation with (what the computer thinks is) her personality, and proceeds to fall in love with her simulation.

In a rare continuity play in STTNG, the real Dr. Brahms appears in the next season. Geordie excitedly approaches her, expecting to develop a relationship that parallels his holodeck experience, but the real Dr. Brahms turns out to be unfriendly…and married. She later finds Geordie’s wildly inaccurate and romantic simulation of her, and is justifiably horrified at the invasion of privacy.

This episode aired in 1989, but was prescient. There is a strong desire today to use AI to simulate loved ones who have passed, or historical figures, or celebrities, or someone out of reach.

The ethics of simulating a real person in AI is exceedingly fraught, to say the least. At best, you get an inaccurate caricature, which may project unfair stereotypes: you get a “chat with Abraham Lincoln” that gives you a grotesque simulation of an historical figure full of tropes; at worst, you turn your ex’s likeness into a subservient puppet that says and does what you want.

The realism of the simulation is what makes this space so problematic today: when the product successfully crosses the uncanny valley, it risks the user forgetting that they are interacting with a mere simulation of a person, and believing that they are interacting with the actual person. I think this crosses boundaries.

I view AI products with suspicion, but I view AI products that simulate actual people with disdain. An AI that purports to let you “talk to your deceased loved ones” is a cruel and disrespectful parlor trick, and should be met with disgust.

Anyway, I love Star Trek.

The end game of generative AI

August 29, 2025

I’m now starting to be convinced that the only maybe-possibly-sustainable use case of AI is going to be ad-driven entertainment/chatbots/“virtual girlfriend” type apps, as well as low-effort social malware such as spam, astroturfing, and phishing, running on very old non-reasoning models (and small diffusion models for generating the “photos” of your artificial girlfriend).

I say this because a trillion+ dollars have been spent on making generative AI productive enough to replace humans, and so far, there has been zero use cases that are even plausible. The cost per inference for reasoning models is absolutely exploding to literally hundreds- or thousandfold over non-reasoning models, and there is little evidence that anything works as claimed.

Low-effort chatbots, though, have been shown to effectively prey on the addictive nature of staring into an interactive funhouse mirror, and has the potential to generate more “engagement” (i.e. ad impressions) than traditional algorithmic social media.

Unless there is a significant advance in semantic modeling and inference, this is as good as it gets: we are now seeing the limits of predictive pattern-matching models. Improvements in “reasoning” are quite clearly asymptotic—a huge increase in token output and processing time only produces marginal improvement in output quality, and is not worth the extra money in virtually all cases.

Read more…

When it’s time to say goodbye

August 20, 2025

Saying goodbye to an old dog, sad

When it’s time to say goodbye, more than anything, I want to give my dog one very good day.

She’d wake and her tummy wouldn’t hurt. Her legs would get their spring again. She’d lick me awake to start the day.

The stairs would be easy for her. She’d get a good chunk of treats and chew them on her favorite blanket.

She’d stare through the window and her eyes would be unclouded. She’d bark at Igor taking his smoke break, and the UPS truck.

We’d go to the park. She’d run and I’d barely keep up. We’d play silly games and she’d bark and sniff and mark all over. She’d go home panting and happy, and she’d get her paws wiped and take a good long nap.

She’d be surprised that dinner is all her favorites: cheese, hard boiled eggs, chicken and rice. Her belly would be full.

We’d settle in for the night and she’d lick my face a hundred times before nesting by my feet. She’d get so many pets.

I’d kiss her as she closed her eyes. Good girl.

Good night. Sweet dreams. I love you.

Respect running code

August 20, 2025

One big red flag in a developer for me is when they complain about how much some old code sucks, when that code is still running and doing its job.

Shitting on running code tells me that a developer has little respect for the work that came before them, and little empathy for the developers who wrote that code, under different constraints, with different tools, in a different time. It’s good to want to improve things! But it should be done with a healthy appreciation of what already works.

Can old code become obsolete? Sure, but that happens after they are replaced by something better, not while they’re still operating (and we all know not every replacement is actually better).

Technology marches on. People learn new ways of solving problems. Tools improve. Organizations and resources change. To judge something written many years ago against the circumstances of today is misguided and—dare I say—disrespectful of those who came before.

PS: And don’t shit on old code even after they’re obsolete. Treat them with respect. Thank them for a job well done, and let them take their place in the history books.

Defederation doesn’t work against large instances

August 16, 2025

Access and communication across diverse servers is important, and people do depend on it to keep in touch with their friends and audience. But the safety and dignity of vulnerable people is more important. Knowingly putting your vulnerable members at risk by placing them within reach of a known-bad-actor server is unconscionable. Privileged people need to fucking deal with it when communication is severed to stop abuse; the safety of vulnerable people takes precedence.

Having said that, I think smaller instances defederating from larger ones remains disproportionately painful for the smaller instance, and largely ineffective in changing the behavior of the larger. Ideally the answer is to never let any instance (or small set of instances) become large and dominant, but I don’t know how to achieve that end in the real world.

Read more…

< Newer Posts