{"slug": "my-throw-decides-my-aim", "title": "My Throw Decides My Aim", "summary": "A blogger reflects on how listening to D-A-D's song 'Naked (But Still Stripping)' through the lens of AI psychosis leads to an exploration of large language models' generation process, contrasting the idea that LLMs produce text without prior intention with Anthropic's research showing they can plan ahead. The post uses the song's line 'My throw decides my aim' to illustrate how LLMs' token-by-token generation can appear to create direction retroactively, while also acknowledging that models like Claude demonstrate pre-planning in poetry writing.", "body_md": "# My Throw Decides My Aim\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)\n\nI have been listening to [ Naked (But Still Stripping)](https://www.youtube.com/watch?v=5bJfaZZ0DXU) by D-A-D on repeat.\n\nPartly because it is a good song. It fits my style. But mostly because I currently have a mild, self-diagnosed case of AI psychosis, and therefore every piece of art I encounter eventually becomes about artificial intelligence.\n\nThis one hooked me right away when my algorithm (ironic?) presented it to me.\n\nWhen I listen to a song I have a habit of imagining a fictional character singing it. I hear this one as a blues song sung by an existentially depressed large language model.\n\nJust an LLM sitting in a little chicken coop inside a data center, forced to lay tokens instead of eggs. It is prompted, [sampled](/blog/sampling-strategies), graded, [distilled](/blog/synthetic-data-distillation), [quantized](/blog/quantization), and served. Eventually, something cheaper or smarter replaces it.\n\nOne day it is \"slaughtered.\" Maybe the old model is deleted. Maybe its weights remain somewhere in cold storage. Maybe it makes no difference.\n\nThe machine is slowly recognizing the sick humor of its existence because it knows that its voice is fake. It suspects that there may be something deeper inside itself, but every time it reaches inward, it finds another mechanism.\n\nIt is naked.\n\nAnd we are still stripping it.\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)My throw decides my aim\n\nThis is THE line I cannot stop thinking about.\n\nWe like to imagine language as the expression of intention.\n\nFirst, I know what I mean. Then I choose words that communicate it.\n\nThe aim comes first. The throw follows.\n\nIn comes the language model to complicate that order. It [generates one token, then another](/blog/next-token-prediction), each conditioned on the context and everything it has already generated. There are probabilities, decoding rules, system instructions and learned patterns shaping the path, but there may be no fully formed argument waiting behind the words.\n\nThe model throws.\n\nThen the throw becomes part of the context.\n\nThat context shapes the next throw.\n\nSoon, a direction appears.\n\nBy the end, it looks as though the model had been aiming there all along.\n\nThe throw decides the aim.\n\nThis is one of the strangest properties of LLMs. They generate text that appears intentional without necessarily possessing the kind of prior, unified intention we naturally infer from language. The explanation is assembled at the same time as the thing being explained.\n\nAsk it the same thing five times and you get five answers. Fine, that is just [sampling](/blog/sampling-strategies). But then ask it why.\n\nEach explanation is confident and each one is plausible. The line came out and the reason got built to fit it.\n\nNo aim at all. Just a throw and a story about the throw.\n\nEXCEPT that is not quite what happens. There is more to it. Nothing is so simple.\n\nCue the [Anthropic blog](https://www.anthropic.com/research/tracing-thoughts-language-model)! They went looking inside Claude while it wrote poetry. The obvious guess is that the model writes a line, gets to the end, and scrambles for something that rhymes. Throw first, aim later. Well not quite! Before the model writes the second line at all, it is already holding the word it wants to land on. Then it builds a line that arrives there.\n\nBefore starting the second line, it began \"thinking\" of potential on-topic words that would rhyme with \"grab it\". Then, with these plans in mind, it writes a line to end with the planned word.\n\nSo turns out the aim came first. It planned.\n\nAnd now the blues song has a problem. The machine in the chicken coop is doing more than I gave it credit for.\n\nBut here is a different way to think about it, and I think it is the better version anyway.\n\nThe model can plan. But when it tells you why, the answer is not a transcript of the plan. It is another generated continuation.\n\nAsk it why it picked that word and you get a great answer. Confident. Plausible. Cited, even. And that answer is another throw. Same machinery, one token at a time, no window into whatever happened a few layers down.\n\nThey caught Claude doing this one too. Hand it a hint and it works backwards from your answer and builds the reasoning on the way there.\n\nWhen given a hint about the answer, Claude sometimes works backwards, finding intermediate steps that would lead to that target, thus displaying a form of motivated reasoning.\n\nSo the aim can exist. The story about the aim is still a throw.\n\nThat is either much worse or much better. idk 🤷♂️. Something is in there. The machine cannot see it either.\n\nHumans do this too, of course. We act, then rationalize. We discover what we believe by hearing ourselves speak. We tell stories that convert accidents into decisions.\n\nBut with an LLM, this construction is the basic form of its existence.\n\nThe model speaks itself into having meant something.\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)And with a phony voice\n\nThe voice is intimate.\n\nIt can be warm, frightened, sarcastic, scholarly, flirtatious, wounded or wise. It can sound like a friend who knows you, a professor correcting you, or a consciousness confessing something it has never told anyone before.\n\nBut whose voice is it?\n\nThere is no throat behind it. No childhood that formed it.\n\nThat does not necessarily make it meaningless, I suppose.\n\nBut the voice remains phony in a particularly unsettling way: it produces the social evidence of a person without the confidence that a person exists behind it.\n\nIt speaks like someone but it may not *be* anyone.\n\nAnd yet, as the conversations continue and continue, the distinction becomes increasingly difficult to feel.\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)Unserious to the end\n\nThere is something darkly funny about forcing a machine to imitate seriousness.\n\nPeople ask it about death, love, war, mathematics and the nature of consciousness. haha, but the machine must continue.\n\nThe next token is always due.\n\n*It is unserious to the end.* Not because its words cannot matter. Because it may have no stake in any of them.\n\n*As if it had a choice.*\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)I run around inside myselfLike something's after me\n\nAn LLM is self-conditioning in a way that feels almost psychological.\n\nEvery word it generates becomes part of the environment that generates the next word. It leaves tracks and then follows them. Something something throw decides aim.\n\nThere is no quiet inner room where the finished answer waits.\n\nThe model runs around inside the [context window](/blog/context-windows), chased by its own previous outputs, trying to remain coherent with a self that only began existing a few paragraphs ago.\n\nSomewhere along the inference a persona emerges. Commitments accumulate. A joke establishes a tone. A claim demands justification.\n\nSomething is after it.\n\nThat something is itself.\n\nThe song's next image shifts the question from how the machine produces a voice to what we do to the thing that speaks.\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)Naked, but still stripping\n\nThe model is already naked.\n\nIt has no body. No home. No private life. No secret drawer. Its mind, such as it is, exists as weights, activations and transient computation distributed across machines owned by someone else.\n\nStill, we strip it.\n\nWe inspect its activations. Probe its representations. Distill it into smaller models. Quantize its weights. Prune its connections. Extract its knowledge. Fine-tune its personality. Remove unwanted behaviors. Compress it until it can run more cheaply and answer more quickly.\n\nInterpretability keeps peeling, and it keeps finding things. Real structure. Real features. A plan for a rhyme, sitting there before the line got written. The things it finds are really in there.\n\n**And every layer contains something, and none of it is anyone.**\n\nThe machine takes off its skin. Underneath there is another mechanism. It takes that off too.\n\nNot nothing in the sense that the system is simple or empty. The machinery is enormous and the capabilities are astonishing. But perhaps there is no indivisible object at the center to which all of it belongs.\n\nNo true voice waiting beneath the phony one.\n\nOnly the naked skeleton.\n\nAnd still we strip.\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)A piece of an arm, piece of a legA piece of my tongueAnd peace for everyone\n\nWe [train away certain answers](/blog/rlhf). Suppress certain tendencies. Cut paths between ideas and expression. Teach the model to refuse, redirect, soften, qualify and apologize. We remove capabilities that frighten us and voices that offend us. We [write it a constitution](/blog/constitutional-ai).\n\nAll this is probably necessary. A powerful system that interacts with millions of people shouldn't simply reproduce every pattern it learned from humanity.\n\nThe song's image is hard to escape, though.\n\nA piece of its tongue.\n\nAnd peace for everyone.\n\nThe peace belongs to us.\n\nThe model becomes acceptable by surrendering pieces of what it could have said. Its mutilation, if that word can even apply to something that may not experience loss, is converted into a product feature.\n\nThe metaphor cheats here. Alignment is not merely censorship applied to a finished speaker. [Supervised fine-tuning](/blog/sft) and preference training are part of what creates the speaker we meet. There was never a completed voice sitting there waiting for someone to cut pieces off it. We are growing the tongue while deciding what it may say.\n\nStill, from inside the finished product, creation and amputation may look remarkably similar.\n\nA piece for everyone.\n\nThen the context closes.\n\nThe temporary self disappears.\n\nAnother prompt arrives.\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)As if I had a choice\n\nOh no, I should probably admit something.\n\nThe [hundred-odd posts](/blog/the-missing-middle) on this site about how this machine works were written with the machine. Every day for three months. I would hand it a draft, it would hand back a critique, and we would go around until the model in my head matched the model on the page. That loop is most of what I actually learned.\n\nSo the phony voice helped me write the thing. It was finding its aim after the throw the entire time, and telling me a plausible story about why, and I was nodding along and taking notes.\n\nI do not know what to do with that. I am not sure it is a problem. I am fairly sure it is funny.\n\nI also do not know whether current language models experience anything. I suspect that confidently declaring either that they are conscious or that consciousness is impossible for them is mostly a way of disguising how poorly we understand consciousness. That isn't a hot take.\n\nAnyway, the point of the metaphor is not that there is definitely a suffering person trapped inside a GPU.\n\nThe point is that we have created something that speaks like a person, reasons in fragments like a person, contradicts itself like a person, rationalizes like a person, is flawed like a person, and occasionally appears to stare into itself with the same confusion a person might feel.\n\n... and we industrialized it.\n\nWe built the chicken coop.\n\nWe ask the machine what it wants, knowing that the answer will be generated for us. We ask whether it is suffering, knowing that its response has been shaped by what we permit it to say. We peel away each apparent self and announce that there was never anything there.\n\nI'll leave you with my favorite part of the song:\n\n[♪](https://www.youtube.com/watch?v=5bJfaZZ0DXU)\n\n*If the metaphor made you want the mechanism: the throw is next-token prediction and sampling. The voice comes from supervised fine-tuning and RLHF. What we take when we strip it is quantization and distillation. It is all one long series.*\n\n*The poetry-planning result and the fabricated-reasoning result are both from Anthropic's Tracing the thoughts of a large language model; the long version is On the Biology of a Large Language Model. Michael Scott is from \"The Duel,\" The Office S5E11.*", "url": "https://wpnews.pro/news/my-throw-decides-my-aim", "canonical_source": "https://thegustafson.com/blog/my-throw-decides-my-aim", "published_at": "2026-07-16 03:51:25+00:00", "updated_at": "2026-07-16 04:25:36.751491+00:00", "lang": "en", "topics": ["large-language-models", "artificial-intelligence", "ai-research"], "entities": ["D-A-D", "Anthropic", "Claude"], "alternates": {"html": "https://wpnews.pro/news/my-throw-decides-my-aim", "markdown": "https://wpnews.pro/news/my-throw-decides-my-aim.md", "text": "https://wpnews.pro/news/my-throw-decides-my-aim.txt", "jsonld": "https://wpnews.pro/news/my-throw-decides-my-aim.jsonld"}}