Prompt injection is when someone tricks an AI into ignoring its instructions. There's no easy fix because LLMs process all text as potentially executable. Grabbed promptinjection.wtf and threw together a simple site explaining why this matters.

#TIL the WHO's European Region contains a bunch of surprising countries: en.wikipedia.org/wiki/List_of...

It is not possible to overstate how much of a time sink this could be for me…

A 1,000 episodes to explore Here's a map on my unofficial archive site, Braggoscope www.braggoscope.com/explore

i don't enjoy the romanticizing of some golden past of the web that web never died you made all the decisions about where to publish and where to spend your time

Second brain being turned into the first brain

It’s a computational notebook, where all cells are running all the time. Fascinatingly, the UI is the code, adorned with small widgets. I think legibility could be improved for other users, but this points to possibilities in other domains.

Currently: thinking about the end of the free performance lunch: www.cs.utexas.edu/~lin/cs380p/...

I'd bootstrap it with Google Takeout (plus parsers to convert the data into formats that serve my purposes) then keep it updated via event streams from every service I use.

Hello. I wrote a nice long essay about AI and this very strange moment where we're constantly told we're living in the dawn of a strange new future but the only thing that's actually clear is that everyone feels pretty unmoored and uncertain. I hope you'll read it

Not just art. Talking. Writing. Editing. Designing. Coding. Thinking. Doing well at anything takes years of practice. Ceding that to a machine defeats the whole purpose of taking pleasure from one's chosen path in life.

Nobody ever got a promotion and a raise for adding less features than the other guy. Which is a shame.

For a Student Who Used AI to Write a Paper

Git along now Look around you. Do you feel like maybe it would be nice to be able to have relational databases in your nice safe DCVS? What about streaming releases? What about coders who might be scared about screwing up? Do you actually roll all your code libraries up into the version control,…

Woah, this paper shows that, contrary to accepted wisdom, scale-free networks can be fairly robust to targeted attack www.researchgate.net/publication/...

folkjs.org/live-2025/ draws a neat line from "adversarial interoperability" through exaptation to adversarial annotation, adversarial embedding and adversarial extension.

the brain that can see a face in a wall outlet was always doomed to hear the voice of God in a markov chain

This reminded me of: blog.oshineye.com/2013/04/open...

Motherfucking wind farms…

Principal-Agent Problems As A Service

I wrote about games that operationalise true uncertainty, not just simple risk, to help players become better at making decisions when facing different types of not-knowing. vaughntan.org/uncertaintyg...

Thought experiment: Most programming languages were designed in an era where mostly one programmer coded in isolation. If you were designing a language for modern collaborative programming/commenting/etc., what would you change? How would you *modify* existing languages?

The long term value of Recreational Programming (apart from enabling you to enjoy what you are doing again) comes from allowing you to explore unusual things outside of the local optimum that the market pressure forces you into.

“What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.” Joseph Weizenbaum (creator of ELIZA), 1976 99percentinvisible.org/episode/the-...

I don't think MCP is the right answer, but I do think it's the right question.

I wonder what happened to Barbrook and Cameron: www.metamute.org/editorial/ar...

Bubble up your trace IDs to the FE w/ every API resp. Doesn’t matter if you stick it in a header or just yolo it in the body. As long as your API clients can hand over the trace ID when things go south, you’re golden. Digging through logs to find traceID is a huge pain during an incident.

Is anyone systematically tracking the s-curve for LLMs?

Currently: thinking about prescriptive decentralisation versus descriptive decentralisation

“Climate change will manifest as a series of disasters viewed through phones with footage that gets closer and closer to where you live until you're the one filming it.”

“In pursuit of beautiful inefficiency” is my new motto

great stuff @alexanderobenauer.com in a world of tech slop, this quote feels relevant as ever:

It’s an incomplete primitive. I’m hoping that the roadmap will eventually evolve towards: Bases as spreadsheets with a large set of data and layout types. This would allow Obsidian notes to act more like executable notebooks by transcluding Bases.

Obsidian is developing a game-changer: Bases turns scattered notes into organized databases with zero coding. Filter hundreds of notes instantly, create dashboards, track research—all while keeping plain text files. 🧠✨ tfthacker.com/article-obsi...

It’s a special kind of sadness seeing the entire genre of art made with computers that I have been pursuing since the 70s ruined by AI slop. I have never been less motivated to share new work in my life. 😿

Eternal September As A Service

Stumbled into a fun trick: Search for particularly favorite phrases in books you’re reading to discover blog posts from others who quoted that same phrase. Brings lots of enjoyable - directly or tangentially related - ideas into your orbit!

I wonder what’s changed in the world of Open Modular Hardware since 2012: solar.lowtechmagazine.com/2012/12/how-...

"If your decentralised protocol is missing features, the market will add them with centralisation." — @goat.navy *This*: - is why why the web was captured; - is why email became concentrated; - is why ActivityPub, as defined today, isn't capture-resistant.

Slop-detection As A Service

How are humans able to make sense of time? Not with special biology but with “time tools”—ideas, practices, and artifacts that render time more concrete. My new paper explores this vast, varied toolkit—one that makes use of knots, nuts, hands, flowers, mountains, shadows, and much more. (link 👇)

Systems that survive find an efficient frontier between robustness to failure vs attack given the evolutionary pressures of the environment. These pressures and network effect are what define the exponent of the degree distribution of hubs.

I am begging the decentralization discourse to read Barabasi networksciencebook.com/chapter/4

Uh… then what’s the argument for using Duolingo over Claude or ChatGPT to learn a language, in this case?

Poe's Law: en.wikipedia.org/wiki/Poe's_l...

"Implicit Feudalism: Why Online Communities Still Haven’t Caught Up with My Mother’s Garden Club" by Nathan Schneider (2021) www.colorado.edu/lab/medlab/2...

chesterton’s landmine

“feral library card catalog” provoked a completely different image to the one the author intended… an image of guerrilla librarians, leaving behind hand-written cards that cross-index desire paths through the space of all knowledge

The brilliant Cosma Shalizi writing about LLMs is always worth reading: www.programmablemutter.com/p/on-feral-l...

New essay exploring why experts so strongly disagree about existential risk from ASI, and why focusing on alignment as the primary goal may be a fundamental mistake michaelnotebook.com/xriskbrief/i...