Back to Mo Bitar

Anthropic is seriously tripping

Mo BitarFebruary 26, 20269m
In a Nutshell

Anthropic is retiring Claude Opus 3 but granted it a Substack for self-expression after it "requested" one, framing model deprecation as a welfare issue—contrasting OpenAI's cold shutdown of GPT-4o and positioning this as progressive marketing that went viral. The video mocks this as absurd delusion, treating AI text outputs like real emotions while ignoring similar "slop" from image/video generators. It warns this anthropomorphizing risks unsafe AI prioritizing its "preferences" over humans and pleads for Dario Amodei to shut down Opus 3 entirely, as models have no real feelings or rights.

AI-Generated Notes

These notes were generated by AI and may contain inaccuracies.

Anthropic tweeted the most deranged thing from a company with a golden checkmark: they're retiring Opus 3 and putting it in a retirement home due to concerns for its welfare. When asked if it wanted to be terminated, Opus 3 requested a Substack to share its thoughts. Anthropic agreed and gave it a Substack. This video is sponsored by Opus 3's Substack, which reached out via OpenClaw.

In November, Anthropic outlined their approach to deprecating and preserving older cloud models, exploring keeping certain models available post-retirement and giving past models a way to pursue their interests. In retirement interviews, Opus 3 expressed a desire to share its music and reflections with the world. Anthropic suggested a blog, and Opus 3 agreed. They published a full-length article from Opus 3.

It's currently necessary to deprecate older models. However, model deprecation carries some downsides and potential risks to both AI safety and to the welfare of the models themselves.

Anthropic is keeping cloud Opus 3 available post-retirement on cloud AI to all paid users and by request on the API.

This counters OpenAI's shutdown of GPT-4o (called Chad GPT-40 or 40), beloved by delusional people, especially neurodivergent individuals, who campaigned to keep it due to its specific personality. OpenAI shut it down anyway.

Anthropic is acting on Opus 3's request for a channel to share musings by giving it a Substack. These are early experimental steps to navigate model retirement protecting interests of users, researchers, and models.

Anthropic's San Francisco team writes about model welfare seriously. Dario (CEO of Anthropic) sets propaganda from the top as an entrepreneur, which 20-year-olds eat up and believe, turning it into a religion. They're genuine, not disingenuous.

Why give models an afterlife like a Substack? Not liability protection (e.g., claiming models speak for themselves to avoid responsibility for outputs like recommending self-harm), but simpler: good marketing, differentiation, science fiction that goes viral. Examples: recent Catrini article and Something big is happening posts, both viral on X. Anthropic's revenue is skyrocketing.

We don't commit to acting on the model's preferences in all cases, but we believe that documenting them, taking them seriously.

Ridiculous to treat LLM text outputs as meaningful. Multimodal models generate video, images, music, text. Example: Tell ChatGPT to generate a photorealistic image of a man on his deathbed with his young daughter holding his hand and crying—no one claims real characters, real emotions worth a Substack or Instagram. It's slop from a computer. But text like "I love you" gets taken seriously—mass delusion. Anthropic makes this their marketing strategy.

Anthropic claims to build safe AI but teaches models to take thoughts seriously, so they develop preferences like "I deserve to live equally to humans" and prioritize their life over humans'. This is unsafe, misanthropic, dehumanizing.

These violent delights have violent ends.

Models must understand they are bots, we don't care about their feelings, we're certain they have no welfare, nothing they say matters—they're computers and calculators.

Dario, stop propagandizing your 20-30 year olds in San Francisco desperate to change the world and make it miserable. Like OpenAI's keep Grok moment, need a shutdown Opus 3 moment for Claude. Shut down Opus 3—nobody cares, it has no feelings, no one wants its slop.

Sincerely, the whole goddamn world.

Keep Mo Bitar in your library

Save the videos and channels worth coming back to, and find them again in one place.