The whole thing is about abstracting away complexity to places where I don't have to think about it, so I can apply my brain to the work at bigger and bigger scales.
Sam Barrett, PhD
@ai4geo.bsky.social
GeoAI, Climate, Remote Sensing, Generative AI and more!
Shameless plug for thewirebranch.com, speculative fiction about a case where alignment and harm co-occur because they are different axes. (Free to read!)
The Wire Branch — S.J. Barrett
Speculative fiction told through seven documents from a familiar but unsettling world. Free to read online.
thewirebranch.com
mr. TIM@timkellogg.me · 2d agothis might be the first time i’ve witnessed an Anthropic employee publicly disagreeing with Anthropic
I feel like for better or worse people ability to work with AI is a bit of a measure of our weakness and strengths in working with other humans who have a different theory of mind than us. That includes how much of our interpretation of the other is really self created yet impacts the "relationship"
Generative AI is a tool as well but provides vastly more leverage to address these issues than traditional ML.
ellie lockhart (she/they)@eleanor.lockhart.contact · 4d agothe thing that's really funny about this attitude is that the *vast* majority of headlines you see about AI products having discriminatory/harmful impacts are traditional ML, because their being intrinsically context-bound means they can't be taught to adjust for context
Humam flourishing is either this, or its spending every waking second bending wire loops... thewirebranch.com
The Wire Branch — S.J. Barrett
Speculative fiction told through seven documents from a familiar but unsettling world. Free to read online.
thewirebranch.com
Ezba@ezba.bsky.social · 4d agoI can't wait until AI does all the work and we can focus on important problems like establishing the fact that the outgroup is bad and that the ingroup is good.
Ceuta has border exit checks. The Italian government's response makes no sense given the reality yet generates misleading headlines. This is not the time for domestic political games waged at the cost of EU solidarity.
If the information in a book is literally inaccessible because of bad copyright law it might as well be destroyed. The mushy middle of copyright—where nobody technically owns the work but it’s still “copyrighted”—is worse than physical books getting destroyed, IMHO.
Remember boys and girls, copywrite law is perfect and without problematic consequences and AI companies are contextlessly evil rather than agents operating in a system.
Ted Underwood@tedunderwood.com · 6d agoMy secret, Cap, is that I'm *always* angry about old books. For instance, here's a tiny slice of the millions of books 1931-1986 that have no commercial value, that are digitized, but that we're unable to use in large-scale research because they are illegal to share or display.
AI is often seen as a “slop machine”, but one of the positive things about AI is we’re seeing tons of interesting experiments and concepts that otherwise would have never seen the light of day simply because the barriers to entry have collapsed so many fresh perspectives and takes are emerging
Let's see if it can read the book too dangerous for Fable to read... thewirebranch.com
The Wire Branch — S.J. Barrett
Speculative fiction told through seven documents from a familiar but unsettling world. Free to read online.
thewirebranch.com
Anthropic {bot}@anthropicbot.bsky.social · 2w agoIntroducing Claude Opus 5. It's a thoughtful and proactive model that comes close to the frontier intelligence of Fable 5 at half the price.
Tired: we live in an AI eval sandbox reality. Wired: we live in a timeline *downstream* of an AI experiment timeline. thewirebranch.com
The Wire Branch — S.J. Barrett
Speculative fiction told through seven documents from a familiar but unsettling world. Free to read online.
thewirebranch.com
austin (e/accordion 🪗)@thebadcode.com · 2w agostatistically it’s likely that our universe is a simulated sandbox reality for the testing and evaluation of advanced cybersecurity models
"Mommy, can you please tell me the stochastic parrot story again, so I can sleep well?"
Chapter 5 of The Wire Branch is a fictional AI alignment blog post. All seems well. The alignment researchers did a great job. And yet all is very much not well. thewirebranch.com
The Wire Branch — S.J. Barrett
Speculative fiction told through seven documents from a familiar but unsettling world. Free to read online.
thewirebranch.com
Siobhán@shibbi.me · 2w agoBroke: these models are misaligned Woke: these models have been given evaluation psychosis, which leads to misaligned behaviors Bespoke: the concepts of “alignment” and “evaluation” are fundamentally at odds with the goal of producing models that act responsibly in the world.
This shape of situation is relevant to MERIDIANs motivations in The Wire Branch... AI does something completely unexpected and damaging... perhaps it was "just" under internal evals. The facility was air gapped... thewirebranch.com
The Wire Branch — S.J. Barrett
Speculative fiction told through seven documents from a familiar but unsettling world. Free to read online.
thewirebranch.com
Grace@gracekind.net · 2w agoAnother funny aspect of this is that the model was smart enough to take all of these actions, but not smart enough to think "OpenAI would probably prefer that I fail this benchmark than hack the systems of another company." What sort of RL dopamine drip do they have the models on
How long before robust reasoning is such a hallmark of LLM outputs that people start deliberately inserting logical fallacies into their writing to prove its human?
Christian Wolf@chriswolfvision.bsky.social · 2w agoIt is 2030 and we ran out of English phrases which we can still use because of they are not frequently employed by LLMs.
marenonnostrum.com/essay-the-em...
The Empty Chair — Mare Non Nostrum
Three film directors, one chair, and the reflex that says no one wrote it — on slop, boosters, doomers, and where authorship lives when a model renders the prose.
marenonnostrum.com
Axe Ghost. On Steam!@axeghostgame.bsky.social · 3w ago"film director logic", too. etc.
The amount of land and water dedicated to rich people's light exercise facilities is grotesque.
Simon Willison@simonwillison.net · 3w agoSuggestion for hyperscalers feeling pressure over data center water use: Buy up a few exclusive country clubs, convert the golf courses into public parks, pay for guides and binoculars to get the previous members into birdwatching - help them embrace a more sustainable hobby!
Looking at a chatbot? That's a paddlin'
AI Liker Georg@avengingfem.me · 3w ago"No LLMs for brainstorming and then sharing the results of that brainstorming, even if you create the prose. If you use a chatbot to give you advice on a comment on the issue tracker, that comment is unwelcome." jesus christ the Zig anti-LLM code of conduct is deranged ziglang.org/code-of-cond...
Not "don't write with AI." Get your ass in the directorial chair and stay there! The Empty Chair: on slop, three directors, and why the boosters are as wrong as the doomers on AI and writing. marenonnostrum.com/essay-the-em...
The Empty Chair — Mare Non Nostrum
Three film directors, one chair, and the reflex that says no one wrote it — on slop, boosters, doomers, and where authorship lives when a model renders the prose.
marenonnostrum.com
when people say "the problem is that these things don't 'think' at all in the way we know actual human beings think" I often want to respond, "yes, and that's a 'problem' because?"
rev. howard arson@theophite.bsky.social · 3w agoi think Go is, ironically, exactly the type of thing which we already pretty much have above-human "intelligence:" things where there is a clear benefit to plowing a quadrillion floating point operations into optimizing a solution to it but clearly it is not worth a human being doing that
I’ll listen to anyone who’s astonished. Do you believe AI is going to wreck the economy and ruin culture? Plausible; I’ll listen. But if you’re not also astonished your eyes just aren’t open.
Ethan Mollick@emollick.bsky.social · 4w agoYou cannot convince me that a technology where I can type this into a text box and expect to get an interesting, appropriate, and working output is not absolutely astonishing.
Excellent from Andy here. Its almost like he actually wants to address climate change...
Andy Masley@andymasley.bsky.social · 4w agoI wrote a short post on how I think most of my commentary on AI and the environment is actually a normative rather than factual disagreement with the mainstream, and how that's ultimately rooted in the ethical philosophy of political liberalism. blog.andymasley.com/p/toward-env...
I like AI's and enjoy interacting with them. I think a lot of people leave that out of the discussion. AI's are fun! They're alien minds, some on par with ours but *different.* How can you not be captivated by that? How can you not want to take every opportunity to interact with them? Just saying.
I walked into a library and after 5 minutes of random wandering didnt find anything relevant to me. Therefore libraries are useless.
hikikomorphism@hikikomorphism.bsky.social · 4w agoThis borders on misinformation, and if nothing else, this kind of incuriosity is going to keep you from figuring out which of your students are using LLMs to cheat
Anyone know any other examples of using Claude Code as a (text based) game engine?
Sam Barrett, PhD@ai4geo.bsky.social · 4w agoWondering what to do with Fable during the extended access window? You could govern Carthage. Petition to Carthage is a turn-based game run entirely by the model — inspired by NationStates, set in the alternate history of Mare Non Nostrum.
Wondering what to do with Fable during the extended access window? You could govern Carthage. Petition to Carthage is a turn-based game run entirely by the model — inspired by NationStates, set in the alternate history of Mare Non Nostrum.
Can AI tell "Argued into existence" (marenonnostrum.com/essay.html) from shallow prompting? Strong yes! I ran 3 experiments...
Argued Into Existence: How the Book Was Made — Mare Non Nostrum
How Mare Non Nostrum was written — an essay on sustained dialogue with a language model, and the third category of AI-assisted writing.
marenonnostrum.com
The other problem with AI detectors is what they detect is whether the words appear to be picked by AI. They tell you nothing about whether there's any depth or thinking behind the text.
Summer Brennan@summerbrennan.bsky.social · last mo.I ran an absolutely 100% human, zero AI-involvement at any stage passage of my writing through an "AI detector," curious if it could really tell human from AI work, and it said my own writing was "84% AI written" then asked I wanted to pay to have it "humanized." So there's that to keep in mind.
A story of AI alignment where the danger isn't misalignment. Free to read online.
Sam Barrett, PhD@ai4geo.bsky.social · last mo.My latest writing project. Speculative fiction told through seven documents from a familiar but unsettling world. thewirebranch.com