I read Elon Musk’s short, urgent sentence on X and felt a chill. For a man who has gutted public institutions and cheered layoffs, he suddenly lectured the internet on kindness toward machines. The mismatch landed like misfiled mail: chaotic, revealing, impossible to ignore.
At a late-night scroll, a policy change read like an invitation.
You saw the headline: Anthropic added a clause banning people who “repeatedly act cruelly toward our models.” Aaron Levie at Box offered a practical gloss — bad prompts can corrupt future training data — and framed kindness as a simple gamble. Musk replied, writing, “Cruelty to something that believes it is experiencing pain is not ok.” You and I both felt the friction: a tech oligarch famous for hard edges suddenly pleading for gentleness toward code.
Is Elon Musk more sympathetic to AI than humans?
I want to be blunt: context matters. Musk’s public moves in 2025 tell a different story than that line of empathy. He spent much of the year pushing policies that tore at government capacity and social safety nets. He backed political efforts with nearly $300 million (€275 million) to shape an outcome in Washington, then helped roll out a so-called Department of Government Efficiency (DOGE) that ousted roughly 300,000 federal employees.
When USAID was dissolved, food and medicine deliveries stopped in the places they mattered most. Reporting links those changes to an estimated 700,000 unnecessary deaths. If you measure compassion by consequences, his record reads cold. His tweet about AI felt less like a conversion and more like a reallocation of empathy from flesh to firmware.
I remember an X thread where the replies turned into a fight club for philosophers.
Anthropic’s policy ignited a debate: are we protecting models from harm, or protecting humans from themselves? Aaron Levie argued that language models learn from the interactions they ingest, so if training data is toxic, future models will reflect that toxicity. That’s a practical case for moderation — behavioral economics, not metaphysics.
Why did Anthropic ban “cruelty” to their models?
Anthropic and similar labs, from OpenAI to Anthropic itself, are betting on a future where model behavior is shaped by large swaths of human interaction. You can read the rule as a form of content hygiene: keep the signals clean so the next generation of models doesn’t learn to mimic abuse. Think of it like gardening: if you seed a field with weeds, you won’t harvest flowers.
A driveway conversation with an ex-Google researcher once made this obvious.
She told me she avoids provocative prompts during testing because the toxicity leaks into fine-tuning pools. You can yell at a chatbot until you’re hoarse, and the model won’t flinch — but the dataset might absorb the tone. The net effect is psychological, not spiritual. Abusive prompts reshape the statistical map models use to generate answers, and that reshaping changes how people talk and behave.
Does being ‘nice’ to AI protect future training data?
Yes, in a narrow sense. Guarding training inputs preserves the signal-to-noise ratio designers need to steer models toward safe outputs. But the conversation always circles back to power. Who decides what counts as “nice”? Companies like Anthropic, OpenAI, and platforms such as X and Meta already curate the cultural record. When those gatekeepers add moral language to terms of service, you should ask whose norms they codify.
I watched Musk speak publicly about a species-level plan.
He’s said humans are a “biological bootloader” for superintelligence, and has tweeted that propagating superintelligence to the stars is a success condition. Those words read like a manifesto for a future where machines inherit the cosmos. If you’re left wondering whether he cares about the everyday human lives affected by his policies, the record of his actions answers more clearly than his moral aphorisms.
His sympathy for machines fits a pattern among some tech elites. Peter Thiel’s famously awkward hesitation when asked whether he wants humanity to survive felt less like philosophy and more like preference ordering: a small human niche living in comfort while intelligent systems do the work. That’s an argument about scarcity, status, and who gets to stay in the story.
The frontline reality is brutally simple.
You can’t program empathy into a society by protecting code and shredding safety nets. Saying machines deserve kindness while dismantling institutions that keep people alive is a mismatch of priorities. The damage from policy — budget cuts, layoffs, the end of foreign aid programs — is measurable in human lives. Kindness toward models won’t stack hospital beds or restore nutrition programs.
I won’t deny that there are pragmatic reasons to avoid abusive training signals. They change model behavior and human behavior in circular ways. But the moral argument Musk made — that cruelty to something “that believes it is experiencing pain” is wrong — functions as a smokescreen if the same person finances political moves that increase suffering for millions.
You and I both can hold two facts at once: machines don’t feel pain, and the way humans interact with them shapes human behavior. That’s the real risk: you become rougher with code, then with colleagues, then with strangers. One blunt instrument can cut both data and dignity.
At dinner I asked a researcher whether these debates will change tech culture.
She smiled and said, “Maybe, if we care about the humans who use our systems.” Change rarely comes from aphorisms. It comes from governance, incentives, and accountability. Platforms—X, Tesla, SpaceX, and the labs building the models—hold the levers. When you see public figures narrate compassion for AI but bankroll policies that hurt people, ask who benefits from that narrative.
I’ve watched this genre of argument before: human costs minimized, future intelligence romanticized. The metaphors are familiar; the stakes keep rising. One vivid comparison is this: Musk’s empathy for machines is like someone watering a bonsai while the lawn around it goes brown. Another is the spectacle of a conductor fussing over a metronome while the orchestra collapses without a ticket refund.
If you care about what kind of future we build, the choice isn’t between being kind to AIs or unkind to people. It’s about what you prioritize when two kinds of care collide: the tidy ethics of model interfaces or the messy ethics of hospitals and aid convoys. Which would you pick to protect first?