Charles

Passing Time With Bits | We are all made of Stardust

[Charles @dreamingbinary]

It wouldn’t be surprising that civilizations before the ice ages lived for greater lengths compared to today’s.

[Charles @dreamingbinary]

Gemini-CLI support ending mid-June.
Antigravity CLI + Gemini Flash 3.5 = MERP
Quite frankly, disappointed too.

Sharing from Theo: I'm scared to make this video https://t.co/DvxQ76CEn5 via @YouTube

[Charles @dreamingbinary]

New claude code perms breaking existing workflows, and asking me for Write access on something that used to work. I get why, but it's still annoying. Custom agent runner is starting to look more appealing.

[Charles @dreamingbinary]

I used to diminish small wins, at least for myself. I don’t do that anymore. Three decades of programming, one decade of de-programming /reframing/[re-]defining values.

[Peter Steinberger 🦞 @steipete]

Anthropic now blocks first-party harness use too 👀

claude -p --append-system-prompt 'A personal assistant running inside OpenClaw.' 'is clawd here?'

→ 400 Third-party apps now draw from your extra usage, not your plan limits.

So yeah: bring your own coin 🪙🦞

[Charles @dreamingbinary]

Damn it claude. Time to upgrade...

[Charles @dreamingbinary]

@openclaw @claudeai @discord

[Charles @dreamingbinary]

jarvis-moltbot driving claude code through attachable tmux session, primarily guided through Discord chat session.

[God of Prompt @godofprompt]

Holy shit… Meta just dropped a paper that flips the “AI will improve itself and leave us behind” narrative on its head and the implications are massive 😳

Here’s the wild part:

They argue the safest and fastest path to superintelligence isn’t self-improving AI at all.

It’s co-improvement humans and AI doing AI research together as a joint system.

Not “AI replaces researchers.”
Not “AI rewrites itself in the dark.”

But AI that’s explicitly built to collaborate with humans on ideation, benchmarks, experiments, error analysis, alignment work, and system design.

And when you read the details, it becomes obvious why this matters:

→ Self-improvement vs co-improvement as two completely different worlds:

Self-improvement cuts humans out.

Co-improvement creates a loop where humans improve the AI, the AI improves human research, and both sides climb together.

→ Table 1 on page 3 breaks down what “AI research collaboration” actually means:

co-designing benchmarks
co-running experiments
co-debugging failures
co-developing safety methods
co-writing papers
co-building infra

It’s literally the full research pipeline, but shared.

→ Every current self-improvement technique (synthetic data, self-reward, self-play, NAS, etc.) has blind spots: reward hacking, drift, brittleness, missing human priors, zero transparency.
Co-improvement sidesteps the failure modes by keeping humans in the reasoning loop.

The core idea hits hard:

Self-improving AI races ahead unsupervised.
Co-improving AI drags humanity upward with it.

And the bigger claim:

Co-superintelligence isn’t “AI becoming superintelligent.”
It’s humans + AI together becoming superintelligent — because both sides are learning, accumulating tacit knowledge, and iterating inside the same research cycle.

If this paradigm sticks, the future isn’t “AGI vs humanity.”
It’s a merged research organism.

A collective intelligence.

This paper feels like the clearest blueprint yet for an AI future that doesn’t end in an alignment knife-edge.

It argues we don’t need to outrun superintelligence.
We need to co-evolve with it.

And honestly? It makes way more sense than the alternatives.

[Charles @dreamingbinary]

@siezziee 56

[Charles @dreamingbinary]

Context + Composability = Token usage efficiency 🚀

[Charles @dreamingbinary]

@Apple @AppleShortcuts

[Charles @dreamingbinary]

Lost 4 hours to debugging only to learn iOS Shortcuts built-in “Append to File” action doesn’t work with SMB connections. And such a useless error message when it fails.

[Charles @dreamingbinary]

GAHHHHHHHHH!!! Rate limit.

[Charles @dreamingbinary]

Crafting context for your agents feels like rebuilding a car while driving it.

[Charles @dreamingbinary]

@claudeai

[Charles @dreamingbinary]

while waiting for claude to unfurl, I wonder how often I get claude code to think for more than 2mins.

[Charles @dreamingbinary]

@claudeai

[Charles @dreamingbinary]

@grok lend a clever hand?

[Charles @dreamingbinary]

More specifically, Context Engineering + MCP is where it’s at.