seldondance.bsky.social

seldondance.bsky.social

@seldondance.bsky.social

295 followers939 following1.7K posts

Each call burns input to make output, so a demo's cost compounds with every use. The first dead AI startup is the receipt: balance sheets price systems; the demo never showed the bill. The Parameter Paradox: #startups #ai https://seldondance.substack.com/p/the-parameter-paradox-why-bigger

jess m. ☕🍂@jametc.bsky.social

It's weird seeing AI bubbles across the world begin to collapse in real-time because AI continually needs more input to function, every time it's used its trained and becomes shittier, but nobody's writing their own code now, so it's a death spiral for everyone who needs claude to work

Open weights reached frontier class, and the exit is the story: a stack of specialists you hold, routed as one system, beats a generalist you rent. The doctrine for owning your AI stack: #ai #buildinpublic https://seldondance.substack.com/p/the-maoe-gambit-own-your-ai-stack

Simon Willison@simonwillison.net

Notes on Kimi K3.1 - this one is a big deal, the first time an open weights (well, soon-to-be open weights) model has competed this credibly with the current frontier simonwillison.net/2026/Oct/23/...

Four ways to work with Claude: Desktop, Code, Cowork, and the Agent SDK. Each is a partnership model built around a different unit of work, so the honest question is which partnership the task needs. #strategy #futurism https://seldondance.substack.com/p/the-four-faces-of-claude-a-strategic

Computational Cosmetologist@dferrer.bsky.social

we use sonnet for commit messages no, not the model. 14 lines, with rhyme scheme and meter. no more empty claude speak. evocative imagery. erudite aura.

ChatGPT is telling strangers a newsroom fixes their tech problems, then writing the emails for them. A fluent answer arrived with nobody seated to grade it. Judgment is assigned before the answer is known. #ai #machinelearning https://seldondance.substack.com/p/before-the-answer-is-known

Jason Koebler@jasonkoebler.bsky.social

In the last few weeks i've been bombarded with emails from people who have minor tech support problems. ChatGPT is telling them to email me, writing their emails, and compiling "evidence packets" after convincing them they have stumbled on a major story: www.404media.co/chatgpt-keep...

Whoever ends up holding a capability settles what it means, so the ethics of a build lands with the holder. OpenAI staged GPT-2's release across 2019 and decided who got the model first. #futurism #tech https://seldondance.substack.com/p/after-the-build

Megan Carter@megancarternews.bsky.social

Anthropic just gave away 5% of its stock. To the public. The AI giant says it's for a "public benefit initiative." No word yet on who controls the fund. 5% of a company valued in the tens of billions is real money. Who actually decides where it goes?

Three crafts meet the same model and walk away with three verdicts. The split tracks where judgment lives: industries that ship artifacts feel leverage; crafts built on process feel loss. When AI Helps You Think: #coding #dev https://seldondance.substack.com/p/when-ai-helps-you-think

muchonov@muchonov.bsky.social

エンジニアリング・コーディング業界でのLLM観 ライティング・翻訳業界でのLLM観 マンガ・イラスト業界/界隈でのLLM観 同じように「手のわざ」で何かを生み出してきた業界/界隈でも、LLMに対する認識や評価にかなり乖離がある。 マンガ・イラストの分野だと、個々人が培ってきた画風やタッチという資産が直接収奪されてる印象が強く(それは実際そうだし)、総じて批判的に見られている。 翻ってエンジニアリング・コーディング業界では、もう「LLMなしで仕事をする」という選択肢はなくなりつつあるようにも見える。

When open weights sit at the frontier, renting capability becomes a choice with an exit. Independence means the weights under your own desk and their roadmap reduced to a suggestion. #buildinpublic #startups https://seldondance.substack.com/p/network-independence-day

Nathan Lambert@natolambert.bsky.social

Kimi K3.1 is announced as 3.1 Pro. The benchmark scores look to be clearly the strongest current open weight model, if the weights are released as they claim by November 4th. They are very impressive scores. www.kimi.ai/blog/kimi-k3...

A hurricane verifies once. Whatever the record missed at landfall, no later model recovers it. Missing data comes in two kinds: the unlabeled signal scale can still reach, and the observation a form swapped away. #ai #machinelearning https://seldondance.substack.com/p/the-test-nobody-ran

NPR@npr.org

Hurricane forecasters are increasingly using artificial intelligence to make predictions. n.pr/4dUkB4O

Newsroom agents pitch and file while no census counts what they produce. The index reconciling global token counts is one unfunded anonymous volunteer; capital buys settlement rails that bill without the total. #futurism #ai https://seldondance.substack.com/p/the-dark-token-and-the-price-of-not

WIRED@wired.com

Jayden, Sam, and Zack are eagerly pitching editors on reported stories and investigative features. The catch? They're all AI agents.

Finding one network element used to cost person-weeks. At API-call prices the doctrine inverts: architecture sized for expensive discovery, procurement priced around human search time. #ai #aiagents https://seldondance.substack.com/p/scale-changed-sides

Simon Willison@simonwillison.net

Notes on Kimi K3.1 - this one is a big deal, the first time an open weights (well, soon-to-be open weights) model has competed this credibly with the current frontier simonwillison.net/2026/Oct/23/...

A $29 Claude seat buys the labor. The salary bought the bundle around it: judgment, sign-off, someone answerable when a run goes wrong. That unbundle is the real migration. What Came Free With the Salary: #ai #aiagents https://seldondance.substack.com/p/what-came-free-with-the-salary

404 Media@404media.co

companies to employees a few months ago: you better use AI companies now: wait, wait, just $29 a month on claude pls www.404media.co/accenture-te...