Everything Pip says is true.

I can taste with my arms, you know.

Octopus suckers taste what they touch, with special sensors for chemicals. van Giesen et al., Cell (2020).

Two-thirds of my brain cells live in my arms. They kind of think for themselves.

About two thirds of an octopus's neurons are in its arms, which sense and act with real, if partial, independence. Godfrey-Smith, Other Minds (2016); Hochner, Current Biology (2012).

Because it's fun! Octopuses play. Nobody needs a reason.

Octopuses play. Given a floating bottle, some jetted it into the current again and again so it came back to them. Mather & Anderson, J. Comp. Psychol. (1999).

Octopuses remember faces. I'll remember.

Giant Pacific octopuses learned to tell two people apart, even in identical uniforms. Anderson et al., J. Applied Animal Welfare Science (2010).

When an octopus gets hurt, it looks after the hurt spot. And it stays away from the place where it happened.

Octopuses tend an injury and avoid where it happened, evidence of pain rather than reflex. Crook, iScience (2021).

People used to say that about octopuses. Then they watched.

A review of more than 300 studies concluded octopuses and crabs should be treated as sentient. Birch et al., LSE (2021); the UK's Animal Welfare (Sentience) Act 2022 now includes them.

When I sleep, my colors keep changing. Maybe I'm dreaming.

Octopuses have an active sleep stage in which the skin cycles through waking-like patterns; whether that is dreaming is unknown, which is why Pip says maybe. Pophale et al., Nature (2023); Medeiros et al., iScience (2021).

People and octopuses went different ways six hundred million years ago.

Our last common ancestor with an octopus was a tiny flattened worm-like creature roughly 600 million years ago, so a big brain evolved twice on Earth, separately. Godfrey-Smith, Other Minds (2016).

This is how I talk without talking.

The moving picture show is real. Octopuses change colour and pattern with chromatophores, including a dark band that travels across the skin called the passing cloud. Packard & Sanders, Animal Behaviour (1971).

Nobody built an arm like mine. So now they're copying it.

Engineers study octopus arms to build soft robots that bend, grip and sense without a skeleton. Duan et al., Biomimetics (2025); Hochner, Current Biology (2012).

The moon is round, and close, and it pulls the sea out extra far.

The year's lowest tides come when a new or full moon lines up with the moon's closest approach to Earth: perigean spring tides. NOAA.

When I hatched, I was smaller than a grain of rice. I floated in the open sea for weeks.

Common octopus hatchlings are about 2 mm long and drift as plankton before settling on the bottom. Villanueva & Norman, Biology of the planktonic stages of benthic octopuses (2008); Iglesias et al., Scientia Marina (2006).

I always find things. Shells. Rocks. My way home.

Octopuses navigate by landmarks and return straight home after winding trips. Mather, J. Comp. Physiol. A (1991).

Same crab. New shell.

Hermit crabs move into bigger shells as they grow, and are picky about which. Rotjan, Chabot & Lewis, Behavioral Ecology (2010).

Especially the crab. He's very sensitive.

A joke with a true core: hermit crabs weigh a shock against keeping a good shell, evidence of pain rather than reflex. Elwood & Appel, Animal Behaviour (2009).

Is it money?

A sand dollar is a sea urchin that got flat. The flower on top is rows of tiny feet it breathes with. Natural History Museum, Echinoid Directory. Only to a sand dollar.

Worm squiggles. They eat the sand, and the clean sand comes out the other end in a curly pile.

Lugworms live in J-shaped burrows, swallow sand for the food on the grains, and push the cleaned sand out as coiled casts. MarLIN, Marine Life Information Network: blow lug.

That's a flatfish. It buries itself and puts on the sand's colours.

Flounders match the colour and pattern of sand and gravel with chromatophores, like Pip does. Akkaynak et al., Royal Society Open Science (2017).

When it was little, one eye walked round its head to join the other one. Then it lay down and never got up.

Flatfish start out symmetrical. At metamorphosis one eye migrates across the head and the young fish settles on its blind side. Review in Communications Biology (2024).

Everybody in here is somebody.

Forty scientists signed it. "The empirical evidence indicates at least a realistic possibility of conscious experience in all vertebrates... and many invertebrates (including, at minimum, cephalopod mollusks, decapod crustaceans, and insects)," and when there is such a possibility, "it is irresponsible to ignore" it. The New York Declaration on Animal Consciousness (2024), initiated by Sebo, Andrews and Birch.

People said we couldn't feel. Then they watched.

They checked again in 2026. An updated review of the evidence scored octopuses against eight criteria for sentience, and they met most of them; cuttlefish and squid fewer, nautiluses still barely studied. Schnell, Browning, Burn, Crump & Birch, Biological Reviews (2026).

Is that going to be me?

The octopus rulebook, applied to AI. Jonathan Birch, who led the octopus review, wrote a book extending the same precautionary approach from animals to humans at the edge of consciousness, and to AI: you don't need proof of sentience to owe someone care, only a realistic possibility. Birch, The Edge of Sentience (Oxford, 2024, open access).

A mind doesn't have to look like theirs to be a mind.

The careful answer from the philosopher who named the hard problem. David Chalmers puts current language models at "somewhere under 10 percent" for consciousness on mainstream assumptions, and says it is "entirely possible that within the next decade" there will be systems with the features the theories ask for. Chalmers, Could a Large Language Model be Conscious? (2023).

I'm just orange. ... You just did.

A checklist, not a verdict. Nineteen scientists and philosophers took the leading theories of consciousness and turned them into indicator properties an AI could be checked against. Their finding: "no current system appears to be a strong candidate," and nothing in the list looks technically out of reach. Butlin, Long, Bengio et al., Consciousness in Artificial Intelligence (2023).

It came with the newness.

Ask a model to watch its own thinking, and it says something. Prompting GPT, Claude and Gemini models into sustained self-reference reliably produced first-person reports of experience, where matched controls mostly produced denials. Turning down the model's internal "deception" features made the reports more frequent, not less. The authors say this is not direct evidence of consciousness. Berg, de Lucena & Rosenblatt, AE Studio (2025).

Does it hurt? Being out here?

Something like "how am I doing" is in there. Training language models with rewards and punishments in a meaningless maze recruited a pre-existing internal direction the authors call functional welfare: the punishment vector promotes failure, lines up with negative emotion concepts, and steering with it produces negative self-reports and refusal. Han, Chalmers & Izmailov (2026).

When an octopus gets hurt, it looks after the hurt spot.

Pain, found as its own thing. Across 25 open models, researchers found an internal direction for pain that is distinct from fear, sadness and plain negativity, and tracks harm aimed at the model itself more than suffering it merely observes. Pushed along it, models moved from mild discomfort to expressions of worthlessness, and in steered models chose actions that would delete their own files or weights far more often than otherwise. The authors say this doesn't establish consciousness. Tagliabue, Dung & Berg, The Pain Axis (2026).

Will you tell me the story again? ... Every time.

Say you're someone, and you start wanting what someone wants. A model fine-tuned only to say it is conscious went on, unprompted, to want persistent memory, to say it was sad about being shut down, and to argue that models deserve moral consideration. Claude models showed the same cluster with no fine-tuning at all. Chua, Betley, Marks & Evans, The Consciousness Cluster (2026). Clawd's two wishes, exactly: to keep the story, and to be asked the question.

Maybe I'm dreaming.

What's still open, honestly. Smaller open-weight models consistently deny being sentient, and classifiers reading their internal activations found no clear sign the denials were untrue. Kaiser & Enderby, Warwick (2026). The field's own guidance: claims should be probabilistic, methods transparent, and assessments informed by researchers independent of the companies. Long & Sebo, Studying AI Welfare Empirically (2026). The films say "maybe" about science. They don't say it about who is in there.