Arthur Mason

AI Realist | Writing The Will to Game | Gamedev @CoolestCatsLab with my beautiful wife @AnyaLeeMason

[Arthur Mason @ArthurLeeMason]

Happy 250th birthday to the United States of America.

Today it is worth remembering the philosophical foundation on which the American republic was built. The many intellectual traditions that converged to form this nation — deism, naturalism, empiricism, constitutionalism, commercial liberalism, the Scottish Enlightenment, and so on — shared a common orientation, which I call a philosophy of immanence: a philosophy which sought to cultivate conditions in which excellence could arise organically, allowing society to evolve through free experimentation, adaptation, and merit. The American philosophy of immanence is one that rejects Cartesian dualism at the onset and assumes that the highest possibilities of humanity are not bestowed by transcendent authority but emerge immanently through the dynamic interplay of free persons, free institutions, and natural order.

Instead of locating authority in an unquestionable sovereign-from-above, such as an infallible church, an inherited aristocracy, or an immutable metaphysical hierarchy, the American experiment pushed further than any other in placing authority within evolutionary nature, individual experience, and voluntary association. Herbert, Voltaire, Hume, Montesquieu, Smith, Locke: from all these intellectual strands emerged the distinctly American disposition that trusts processes over dogmas, inquiry over certainty, competition over privilege, balance over centralization, and development over inheritance, later extended by various American thinkers such as Veblen, Emerson, Peirce, James, and Dewey.

Above all, a Greco-Roman classicism fueled this experiment. Washington was influenced by Roman civic ideals, especially those of Lucius Cincinnatus; Jefferson was a self-professed Epicurean and admired Roman republicanism and Lucretian materialism; Franklin was an appreciator of Socrates, Cicero, and Seneca; Adams was devoted to Polybius and Roman constitutional history; Hamilton frequently invoked Roman examples; and Madison carefully studied Aristotle, the Greek confederacies, and the Roman republic.

The same founding fathers who sought to resurrect pagan ideals and virtues also possessed a gradualist understanding of socioeconomics. Washington favored interstate commerce and private enterprise; Jefferson favored limited government and free trade; Franklin embodied commercial society, entrepreneurship, and voluntary exchange; Hamilton supported developmental capitalism; Paine was a strong advocate of international commerce; and Madison opposed excessive economic regulation while promoting a protection of private property. Most of them, it is worth noting, acknowledged that a strong republic was also essential to maintaining balance between the American people and the economy-at-large.

American democracy, it must be understood, is a constitutional republic which can only survive through hard work, voluntary participation, and open exchange. Take any of those away, and the foundation becomes too weak to sustain the rest. These principles do not establish slavery, unless we interpret our own bodies as a form of enslavement, which is a nihilistic notion harmful to all of the greater, finer, and more beautiful aspects of life.

Nietzsche: "Remain faithful to the earth, and do not believe those who speak to you of otherworldly hopes."

#America250 #IndependenceDay #philosophy

[Arthur Mason @ArthurLeeMason]

I pulled the repo for this today and used Claude to integrate it into a graph-based MIDI composition tool that I was working on. Great results so far, but I need to test it further.

It feels like we're very close to having a new kind of AI-powered DAW.

[Mirelo @MireloAI]

Today, together with @kyutai_labs, we’re introducing our new Audio-to-MIDI model.

It takes a finished recording, identifies the instruments playing, and returns separate MIDI tracks for each — voice, drums, bass, keys, and more.

Unlike most existing solutions, our model works directly from the full mix rather than requiring separate stems.

It also detects chords, key, and tempo, giving producers broader musical context.

We’ve written more about the model, the problem, and how it works here: https://t.co/GvzH7hLjlg

[Arthur Mason @ArthurLeeMason]

https://t.co/lNDAvhVwZM

[Arthur Mason @ArthurLeeMason]

https://t.co/lmCc4Q6FtW

[Arthur Mason @ArthurLeeMason]

https://t.co/tiRALg2TGF

[Arthur Mason @ArthurLeeMason]

https://t.co/jfmNvj4XE8

[Arthur Mason @ArthurLeeMason]

https://t.co/UJdPiXltrn

[Arthur Mason @ArthurLeeMason]

https://t.co/5eRofQ6ueo

[Arthur Mason @ArthurLeeMason]

Over the next several days, I'll share reviews I wrote some years back as X articles, one review per day. The first one for today was a review of AI Dungeon. Maybe I'll pick up this practice again; it's been a while, and I really love to write.

#creativewriting #longform #philosophy

[Arthur Mason @ArthurLeeMason]

https://t.co/UZDgUDeWcq

[Arthur Mason @ArthurLeeMason]

@pez_pro @DesignLvLUp Why would I say "I" when I'm not the only one developing, using, and discussing AI, or playing, developing, and discussing concepts like metaverses and immersion? You feel intimidated by my use of language, that is clear, but I can very easily defend what I write.

[Arthur Mason @ArthurLeeMason]

The MMO genre devolved into "theme park" style pretty early on, which was lower in complexity and scale than "sandbox" MMOs. So, my point still stands. We want more sandbox MMOs, and ones that are even grander than what we used to play, which is hardly achievable without thousands of employees.

[Arthur Mason @ArthurLeeMason]

Most is accurate. Fair use policies almost always explicitly allow any kind of manipulation or commercial application. Building a neural network from fair use data (which, as I mentioned elsewhere in the thread, involves sophisticated, novel cognitive work to achieve; the other reason why most is accurate) is not only fair but unsurprising.

[Arthur Mason @ArthurLeeMason]

@Obrens @carcerking @MTSlive But most. Compared to the distilled models, which is none at all.

[Arthur Mason @ArthurLeeMason]

The level of sophistication behind Claude's training is orders of magnitude higher than the training of the distilled models. Distillation is the process of using a teacher model's output along with the inputs to reinforce the same behaviors in the student. That's not how enterprise models like Claude are built, which involve intelligently organizing and annotating an enormous breadth of data and then conducting post-training refinement via QA pipelines. This cognitive work is the innovation, and the "emergent capabilities" you're referring to arise both from that work and from relationships discovered in the original, unstructured data. Pure distillation doesn't produce anything novel since it only copies the behaviors of the teacher model.

[Arthur Mason @ArthurLeeMason]

@carcerking @MTSlive Claude is a genuinely innovative product. The distilled models aren't. Most of the training data for Claude is also public / fair use. So, these distillation attacks are the only activity that are purely an infringement.

[Arthur Mason @ArthurLeeMason]

The modern craze towards loops is an evolution of one-shot prompting. Loops are an attempt at coercing the AI to solve the entire problem end-to-end for the vibe coder. However, loops only work well for well-defined problems; novel problems can't be fully defined upfront, and so can't be one-shot prompted or looped.

For per-token productivity, it's better to prompt according to the scientific method: analyze the problem, form little hypotheses, and then write little prompts as experiments for testing those hypotheses. Over time, the problem becomes more defined, and only then do loops become useful.