Cryptic Swarm

Building Decision Plane for real-time optimization. Models compete - no single model rules them all.
Hire Human AI courses contributor

[Cryptic Swarm @CrypticSwarm]

Software engineering is phoenixing.

Companies are downsizing.

Agent swarms are promising to replace entire engineering teams.

And yet, the best engineers are busier than ever.

[Richard Sutton @RichardSSutton]

The bitter lesson in 26 words:

Don’t be distracted by human knowledge, as AI has been historically.
Instead focus on methods for creating knowledge that scale with computation, like search and learning.

[Cryptic Swarm @CrypticSwarm]

@ThePrimeagen A flat rate on tokens would also disproportionately penalize smaller cheaper models compared to their energy usage.

[Cryptic Swarm @CrypticSwarm]

@mcuban Not all tokens are created equally. Flat rate penalizes smaller cheaper models disproportionate to their energy usage.

[Cryptic Swarm @CrypticSwarm]

The time is now.

Learn the skills to use the orchestration rigs of tomorrow.

Learn to build your own rig, or get left behind.

Start here: https://t.co/xaO4zET9zv

What's the best tool in your rig right now?

[Cryptic Swarm @CrypticSwarm]

Software engineering is phoenixing.

Companies are downsizing.

Agent swarms are promising to replace entire engineering teams.

And yet, the best engineers are busier than ever.

[Cryptic Swarm @CrypticSwarm]

The old org chart is dead.

Now it's about coordinating humans and rigs the same way we used to coordinate engineering teams.

The companies that figure this out first will eat everyone else's lunch.

They will build massive systems.

[Cryptic Swarm @CrypticSwarm]

@theo Yes. I've lived in the terminal for years. It fits well with my flows in Tmux + git worktrees + containerized harness + containerized MCP, etc.

[Cryptic Swarm @CrypticSwarm]

@slicknet But do these additional improvements only further incite Jevons paradox and drive increased usage and costs.

[Cryptic Swarm @CrypticSwarm]

@slicknet Tooling can also offset usage costs:
- Dynamically route to cheaper models by task complexity
- Subagent and tool delegation to keep context tightly scoped
- Smart memory: only pull in what's needed
- Skip agents for repeatable flows (use scripts)

[Cryptic Swarm @CrypticSwarm]

In AI alignment, what matters more: shaping the AI's utility function (its core objectives and values), or curating its training data (what it learns from the world)? This reframes the entire alignment problem as nature vs. nurture for machines.

[Cryptic Swarm @CrypticSwarm]

Prompt/System prompt learning has the potential to break Jevons' paradox for any repeatable tasks delegated to an LLM.

[Nick Apperson @nick373]

@CrypticSwarm It actually speeds up code on Haswell by 25%. #Fact

[Cryptic Swarm @CrypticSwarm]

People often end emails with "Best, [name]". Wonder how often people mistype and instead write: "Beast, [name]"

[Cryptic Swarm @CrypticSwarm]

@0x00A slightly more flexible is setting tabwidth to 80. Then running gq with appropriate motion such as:

gggQG

[Cryptic Swarm @CrypticSwarm]

What kind of fertilizer should I use on my server farm?

[Cryptic Swarm @CrypticSwarm]

There are 2 types of people in the world: Those who can extrapolate from incomplete data.