🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Exactly. • Error correction is exclusively an agentic process. • That is, by itself, a dynamical system is not capable of its own error correction. For example, the free-energy-principle, by construction, is not error correctable. • Therefore Barbara McClintock's experiments

The key to true autonomous AGI will be embodied intelligence. And the key to that development will be our ability to engineer, build, and program systems of nested agency. • A deeper implication from this observation is that however we choose to define agency, that definition

Is agency a local or nonlocal phenomenon?... • At the foundation of any theory of agency, there must be a physical layer of hardware which is instantiating it. • A fair question then “Is agency (in terms of the physics involved) a local or nonlocal phenomenon?” • Physical

Imagine a language model truncated to a dictionary of only a few dozen words. And rather than a “next word” selection based on a probabilistic search, use instead a hard-coded selection function. Such things have been done and are called string replacement schemes. • While they

The term "reductionist" seems to have become a synonym for any theory that rejects/excludes telos. • But what if a comprehensive physical theory of agency actually ended up explaining how the process of telos works? • Would such a theory still be considered a reductionist

Reposting a recent reply thread (1/2). Balázs Kégl Karen Wong • Question, what would be the difference between small language models and string replacement schemes like Post-Tag and Semi-Thue? • With that said, the future of true autonomous AGI looks to be small language

Reposting a recent reply thread (2/2). Balázs Kégl Karen Wong My question is driven by curiosity. • Yes, LLMs are based on probability selections but what happens if your probabilities take the default values of one and the rest zeros? Your language model becomes hardcoded and

When exploring questions of consciousness, intelligence, free-will, agency and etc., one can start at the level of physics with Maxwell’s demon and work up in complexity. Or start at the level of human experience and work downward in complexity. • Any reality check for one’s

The ability to self-prompt is exactly what happens when you say that a conversation can “turn on a dime.” It is an ability quite unique to language models. • Control algorithms, if recast as conversations between appropriate small language models, will inherit this ability to

Any statement about the limits of computation based on abstract mathematical principles, is also a statement about what the laws of physics allow. • Physical systems, capable of computation, being the bridge connecting the realm of abstract mathematics with the world of the

When you dive into the Theory of Computation you find that there is a correspondence, or a dual relationship, between a computational structure and the language that it will accept/recognize. • That is, a language model with appropriate grammar will generate exactly the same

The most appropriate programming paradigm for systems of nested agency will be nested language models. • But what would source code for such a architecture and corresponding language model look like? • And what about compiler tools, ...with different words within a source code

Small language models (*), as templates for computation, can be run interpretively, sans-operating system. That is, they lend themselves to being run bare metal. Resulting in very minimal memory and computational footprint. • As opposed to programs being compiled externally,

A Post-tag system is a string replacement scheme which is also Turing complete. • Tag systems can also be recast within the framework of a language model; that being, in a reduced-instruction-set "RISC" sort of way. • This suggests that within LLMs there will exist at least

Reposting this reply from another thread in hopes of a better exposure than just a dozen views. ...It has everything to do with autonomous general intelligence, the natural structures for which are systems of cooperative and nested agency. • And with the natural programming

When true autonomous AGI arrives it will be embodied and based on nested agency. • The current conceptual barrier to progress is that software development commands the field of AI right now. • But the concept of nested agency doesn’t make any sense on a strictly software basis.

By almost any measure, bacteria and archaea are the overwhelmingly dominant life forms on this planet. But when conversation tends to the origin and tree of life, these little guys hardly ever get any recognition. • In a similar fashion, the overwhelmingly dominant computational