This is a good thread about about why AIs are unlikely to become single minded doom creating paperclip maximizers. Unfortunately no such reassurance applies to corporations

Colin@colin-fraser.net · last wk.

Alright here’s what I think about this. For me, this argument is a little bit slippery about what it means for a system to have “a final goal”, and takes for granted that any intelligent system must behave as though it has such a thing. I don’t think this is necessary, and I don’t think LLMs do.

Interesting new pattern I'm seeing in multiple places: - Person has brilliant idea. - They use AI to write it up. - The write-up now contains so much indirection and so many AI tics that the only way to understand it is to talk directly to the person. - ???

It does seem like making companies liable for their own containment breaches might introduce a useful level of caution into the industry. Like sure, we get that at the current speed you can’t avoid disaster. That’s why we’re asking you to slow down to the speed you can be responsible.

Vicki@vickiboykis.com · 2w ago

It’s pretty wild how cleanly the labs have been able to spin basic security incompetence into positive marketing

If your method of building smarter models quickly involves accidental hacks with growing and hard-to-detect external damage, then that may not be the ideal strategy to prevent accidental hacks with growing and hard-to-detect external damage.

Thinking about Xanadu and Vingean Failed Dreams: gwern.net/xanadu

Project Xanadu: Even More Hindsight

Retrospective on Project Xanadu’s success and failure: a lack of design iteration, meaningful use-cases, or practicality stopped a valuable vision from maturing into something useful. (And contrasted ...

gwern.net

If you don't read the code then you'll have to find or create some other abstraction/notation that gives you legibility into the system and lets you change/guide the system. What will those things be?

Great article relevant to bill C-22 debates in Canada. The argument that backdoors only weaken your *own* systems because adversaries are likely to use someone else’s systems is even more relevant in Canada than in the US, simply because we’re smaller and most systems come from elsewhere.

Matthew Green@matthewdgreen.bsky.social · last mo.

I wrote up a new post about what AI software finding might mean for the backdoor debate. blog.cryptographyengineering.com/2026/08/14/e...

chapter 1 of Greg Egan’s DIASPORA is the realest depiction of the birth of a continuous learning AGI that we’ll have until we actually have AGI and it can tell us he shared it for free on his website. Read it (then buy the book) Orphanogenesis www.gregegan.net/DIASPORA/01/...

Currently: wondering what hardware an AI-native equivalent of teenage engineering would be making?

Currently: thinking about the Pi ecosystem and horizontal gene transfer: en.wikipedia.org/wiki/Horizon... in a world where anybody's agent can exapt/copy/be inspired by any aspect of anybody else's work.

Horizontal gene transfer - Wikipedia

en.wikipedia.org

On 2024-02-22 I reached out to @antithesis.com fanboying and wanting to use it for @tailscale.com. They were expensive & I couldn’t get people on board w/ buying their service. Instead, more time + money went to: tailscale.com/blog/sqlite-... Nerd sniped them and: antithesis.com/blog/2026/wa... 😅

Breaking the WAL | Antithesis

I did this from a lawn chair on the Sunshine Coast.

antithesis.com

Theory: software used to be an evolutionary system. Software is now a cultural system.

What are the disciplinary values in the craft of software?

Paul Ford@ftrain.com · 2mo ago

contrast that with this lecture by Terence Tao on how to address LLM-driven changes in your field but hold onto disciplinary values teorth.github.io/tao-web/slid...

Great lens. I often think of Gregory Bateson’s notion of “making circuit”. You want the important feedback loops to make circuit through you, or else it’s no longer your OODA loop.

skynet does not have to be conscious for terminators to happen

Currently: thinking about agentic education. Example: pi-school.ziki.boo

A new TIL on adding custom MCP servers to both the ChatGPT and Claude regular chat interfaces - it's a little less obvious than I had hoped, but I got there in the end til.simonwillison.net/llms/mcp-in-...

"A change of perspective is worth 80 IQ points."-- Alan Kay

Gordon@gordon.bsky.social · 2mo ago

How to survive in the AI information economy: you gotta be Rick Rubin maxxing. The thing he is bringing is not a set of mechanical skills (commoditized). What he is bringing is a unique set of tastes, perspectives, and values.

In 2026: how can we tell that the quality of our thinking is deteriorating?

“The primary feature for easy maintenance is locality: Locality is that characteristic of source code that enables a programmer to understand that source by looking at only a small portion of it.” #RichardGabriel

“I think what's really important right now is for programmers to know what's worth making.” #HalAbelson https://gigamonkeys.com/code-quarterly/2011/hal-abelson/

some day, when I retire, I might catch up on the public ranting I wanted to do that would have been politically indelicate at the time.

alec@alecsparks.bsky.social · 4mo ago

if I was the largest grantmaking organization at the intersection of the arts and technology I would simply attempt to produce actual deliverables that benefit the field

Keyboardia, my music synthesiser, now has MCP support: keyboardia.dev/mcp How should agents leverage it?

Picard leadership tip: The larger your audience and the higher your position, the greater the need to communicate with care and precision.

I think this is a big gap in software design culture. There's lots of commentary on visual design and micro-interaction, but remarkably little explicating conceptual and structural design. This section from Magic Ink is lodged in my mind. buff.ly/Bl9JwTv

Do evals make more intuitive sense to you if I describe them as "benchmarks for quality?"

Theory: it's not a "dark forest." We're just blind. If that's true then as we develop the tools and intellectual capabilities to 'see', the universe will come to seem like a rainforest teeming with life.

I see this more in the wealthiest and most connected people not necessarily the smartest.

Russ Wilcox@russwilcox.bsky.social · 3mo ago

Is anyone else having the sensation that the essays you are reading from the smartest people you follow have all started to reason and even sound just like Claude?

Currently: Trying to stay out of the gravity well of Computational Aesthetics research long enough to ship this PR: github.com/adewale/agen...

Complete Section B brand primitives with executable evidence by adewale · Pull Request #172 · adewale/agentic-mermaid

What Complete the mandatory Section B brand-primitives program (B0–B3 and B5) through the existing StyleSpec stack, while preserving authored Mermaid paint and deliberately leaving evidence-gated B...

github.com

I wish people knew more about Game Theory than the Prisoner's Dilemma. So I (and Fable) made this to help people think through the other payoff matrices that are available: claude.ai/public/artif...

Claude Artifact

Try out Artifacts created by Claude users

claude.ai