sark

cultivated abandon |

[sark @afoolswisdom]

Our language is wiser than ourselves: when we say that something is "a thing" we implicitly recognise that questions of ontology are questions of social coordination.

[sark @afoolswisdom]

LLM coding agents are to software what informal proof is to mathematics: a social technology for coordinating above the level of formal objects.

[sark @afoolswisdom]

The LLM agents will foreground (i think, and i hope) that following rules is trivial, and the hard part is in deciding which rules to play by in the first place, a fundamentally arational process.

[sark @afoolswisdom]

Will LLMs finally wake us up from the thousands-year-long fugue of rationalism, and recover what it means to be human?! https://t.co/TYJiz95qMf

[sark @afoolswisdom]

The rooms of different individuals are more different than how their computers are set up. And do we say "Oh, they are not skilled in carpentry, or interior design, so the professionals should decorate their rooms"?! No. Software is bullshit, and it WILL be disintermediated away.

[sark @afoolswisdom]

Ontology is coordination equilibria. Ontology takes up space on the heap.

[sark @afoolswisdom]

The point of agentic AI isn't to help software engineers write more code. The point is for LLMs to reinstate natural language, and to disintermediate software away entirely.

[sark @afoolswisdom]

I really really hope that LLMs serve as a reductio that that we can operate everything as a matter of optimisation and forget about the role of choice. Because if does not, and we are still as enamoured by the spectre of rationality as ever... then we are well and truly fucked.

[sark @afoolswisdom]

The very weird but important disintermediation about to happen is simply traditionally written software getting out of the way of humans communicating their intention in natural language for LLM agents to implement.

[sark @afoolswisdom]

The greatest question of our time is whether the adoption of LLM agents will cure us of the hangover of thinking judgement is possible without choice, or if will we simply double down on the idea of "automation" and spiral ever faster into the abyss.

[sark @afoolswisdom]

Chronic anxiety is a runtime/compile-time confusion, about when to think and when to act. And this is why software engineering feels like chronic anxiety.

[sark @afoolswisdom]

The superpower of homo sapiens is not cooperation, it is coordination.

[sark @afoolswisdom]

The boundary of the self is not a fortress, nor should the body merely a channel through which currents without course, but who you are a factorisation for the structure of reality itself.

[sark @afoolswisdom]

All ill from how the wider culture mythologises mathematics comes from prioritising truth-preservation over structure-preservation.

[Mario Krenn @MarioKrenn6240]

After the apparently amazing announcement by @mathematics_inc on the formalization of a major recent Fields-medal winning theorem, i had no idea how pissed the math-formalization community is.

Very worrying discussions by some of the leaders/founders of Lean's mathlib.

cc @ChrSzegedy

[sark @afoolswisdom]

One of the greatest advantages mathematics has over software is that doing mathematics informally saves you from having to write a compiler every time you want to perform abstraction. And in software, in order to avoid having to write a compiler, you do it in runtime instead.

[Jonathan Gorard @getjonwithit]

Now that I'm back in Princeton following the @DARPA expMath kickoff event, I'm beginning to collect my thoughts on the future of autoformalization, AI for Math, and AI for Science more broadly. Here's where I've got to.

"On productive and unproductive frictions"

[sark @afoolswisdom]

God of the gaps, but the gaps are actually the parasitic niches of elite underproduction.

[sark @afoolswisdom]

It seems the LLMs are better at generating code than producing math. This is not a skill issue, but comes from the fact that producing math involves choice, and producing code doesn't.

[sark @afoolswisdom]

More symmetrically, this is a double instrumentalisation of thought:

Meditation: We are not our thoughts, so they can instead serve us.
Mathematics: Our thoughts are not reality, so they can more freely model it.

[sark @afoolswisdom]

More pithily:

Meditation: We are not our thoughts.
Mathematics: Our thoughts are not about anything.