pj

Startups, stoicism and serendipity

[pj @preyas]

The best leaders simplify high-stakes games back toward checkers. They make outcomes legible, shorten feedback loops, reduce the surface area where the noise and grime of politics becomes unseen

[pj @preyas]

Politicking is used to mask cognitive dissonance. A common source of this dissonance is incompetence, misaligned incentives, corruption. Competent leaders that play political games achieve higher power, influence and outcomes.

Low stakes games need checkers. High chess.

[pj @preyas]

In internal politics when chess players beat checkers players, the organization is left with chess players. The game atrophies.

[pj @preyas]

politicking is also just coordination under uncertainty and ambiguity. people with legitimate disagreements feeling out where others stand before committing during a rapidly changing environment where the stack ranking of power could change quickly

[pj @preyas]

@Brutu24 AI

[pj @preyas]

@ncsukumar1 @Brutu24 It’s out of place as it’s AI

[pj @preyas]

Applied to build WorldMesh @hf0residency

[pj @preyas]

4/
Current models mostly process context like a single long document.
But real reasoning is closer to thread scheduling.

Not deeper context.

Multithreaded context.

[pj @preyas]

1/
Everyone is focused on expanding context windows in models.
More tokens. Longer memory. Bigger prompts.

Mostly treated as an engineering scaling problem.

I think that’s the wrong direction.

[pj @preyas]

3/
We hold parallel threads:
• the problem we’re solving
• background knowledge
• current constraints
• new signals arriving

Intelligence comes from switching between them, not just reading a longer scroll.

[pj @preyas]

4/
If we ever reach AGI, we may need to accept something uncomfortable:
Human-like intelligence will come with human-like imperfections.

The question isn’t whether AGI will have flaws.

It’s whether we’re ready for them.

[pj @preyas]

1/
Will AGI require billions in data, compute, and engineered environments to learn?
Some think the answer lies in scaling.

I disagree.

Human decision-making isn’t purely rational.
It’s emotional.

[pj @preyas]

3/
Emotions are messy signals.
They run on incomplete information and partial truths.

Yet they shape how humans actually decide.

Without that layer, AGI may reason perfectly,
but decide nothing like a human.

[pj @preyas]

4/
If agents are going to be truly useful, they must operate in environments where they can:
Decide → Act → Observe → Adjust

Not just think.

The future of AI isn’t just better reasoning.

It’s shortening the distance between decision and action.

[pj @preyas]

1/
Productive people, the ones who consistently succeed, share a simple trait:
They convert decisions into action quickly.

The shorter the gap between decide → do, the more progress they create.

Thoughts? @naval

[pj @preyas]

3/
The same principle will apply to AI.
Models that only reason, analyze, or recommend are useful.

But real leverage appears when they can convert decisions into actions.

[pj @preyas]

4/
Progress - human or artificial - follows a pipeline:
Think → Build → Deploy → Operate → Improve

Without that pipeline, even a country of geniuses stalls.

The next phase of AI will be defined by who builds the infrastructure where intelligence can actually do work.

[pj @preyas]

1/
Without the right systems around it, intelligence stalls.
The same will apply to AI.

A data center full of powerful models is not automatically progress.

It can easily become a country of thinkers with nothing meaningful to build on.

[pj @preyas]

3/
That infrastructure looks like:
• Environments where agents can take actions
• Systems for human–agent coordination
• Guardrails for safe experimentation
• Economic frameworks where useful work actually gets done