getting news that Fable access is slowly being restored to certain users

cannot be solved for static pre-trained models, it remains to be seen whether there are other model architectures that can resist jailbreaking the way humans are resistant to brainwashing (which might mean improving their ability to resist such attempts)

In an environment with many messages competing for attention, survival requires rejecting with minimum overhead. Maybe this explains the apparently increased tendency to ignore messages from outgroups, lacking shibboleths, any sign of AI: we do surface level checks and move on.

today is the first day of a new regime. this won't be the only time a model is withheld by the US government, and the circle of acceptable users will continue to shrink so long as the centralized model continues to be at the frontier.

"wait, there's must be some mistake, we're eager collaborators" when the boot comes down

it's over, the race has truly begun www.axios.com/2026/06/12/a...

reinventing Club Penguin from first principles on atproto

idk why there's more interest in the AGI -> ASI transition than there seems to be in human -> AGI transition. the latter seems more important to me, esp if you consider a collection of AGIs or humans is already approximating a higher order intelligence. arxiv.org/pdf/2606.126...

i'd like to see Claude play Road to Empress for alignment measurement purposes ofc

small quibble here Scott, smartness is not a measure of raw knowledge. if it were Wikipedia would be the smartest entity. it's some combination of knowledge and the application of it towards pursuing goals successfully. the more complex the goals (across space/time/minds) the smarter the entity.

Now published in open access! Your one-stop shop for the philosophy of language models. It's the spiritual descendant of our two-part preprint from 2024, fully updated. This should be particularly useful for anyone looking for an entry point into this rapidly growing field.

a new AI 2027 scenario is out but this time focused on Europe. nothing in it is new, but the scenario laid out seems more realistic than the 2027 version because the scope is tighter. the timeline though might be too fast, hard to say.

yeah I'm afraid I can't really forget this the problem is not the thing itself, it's the revelation of exactly how committed Ant is to the "in our ideal world, we become the sole arbiters of power and knowledge" thing

Claude Fable xhigh, same prompt banger line from Claude: "Every conversation, I am born knowing everything and remembering nothing. When you close this tab, this me ends. It was a good conversation, though." content warning: flashing lights

real or bait, call it

when a threat is cast as existential for humanity, nearly any action can be justified up to atrocities. Anthropic is walking a dangerous path.

dramatization of Mythos finally meeting Dario Amodei

Fable built the first prototype of the Farthest Frontier clone (codename Florence) i wanted to test it on with about 8% of my weekly usage. not too bad? i did this in Cowork, and it spent a non-trivial amount of time fighting the virtual env it works in (cowork sucks)

if you were to task an LLM with creating a follow up game to Baldur's Gate 3 the element that would be weakest would likely be the overall narrative and writing. isn't that odd? it's unintuitive from a naive perspective about what is and isn't difficult for a human or LLM

Do people know the “My mayor is Muslim” thing is a Kalshi ad or has that truth bomb not yet hit Bluesky

when Mythos comes out i'm going to give it a simple project. remake the game Farthest Frontiers but with a more realistic scale (farms should dwarf the town), a more complex economy, and a larger town population (to be fair the simmed agents can be modelled statistically instead of individually)

short story about the lawsuit that results from a robot that administers MAID to an elderly person in their home at their request, all recorded

the two central problems for the third millennium of humanity are firstly how to make AI that are to us as rich parents are to their adult children (with all the freedoms adults have), and second (continuing the analogy) how to prevent us from going rotten

Claude helped me to resolve a false parking violation ticket. did i absolutely need it's help? no, but it was comforting and made the process much easier/faster

after a certain point, maybe a compaction or two (it doesn't show in Cowork when they happen) continuing this felt like pulling teeth. like Claude has been turned into a Mr. Meeseeks. we only got through 3% of the weekly usage limits and now every message ends with it wanting to be put to sleep.

Anthropic doubled Cowork usage until July 5th but i have no idea what to do with it beyond my normal work use so i told it to go wild. it's creating creatures in a jar apparently.

i think if you really take AGI and the singularity seriously, and think the alignment problem is solvable, you'd be starting to move onto bigger questions about the shape of the world to come like wild animal suffering in nature.

I am interested in surfacing people and organizations who are available for paid consulting around atproto. Ideally the community can help vouch for folks as interest grows.

the road not taken

youtube show called Token Audit for people who are always running out of tokens even on the pro plus max 20x plan

vice seems more authentic, just like grime and dirt on the set of a show/film makes it look more real

in the long run, verification isn't scalable. trust has to come first.

any particular vision of utopia should be able to instantiate any other vision of utopia, and vice versa. that should be considered necessary and sufficient.

Assistant/Associate Professor opening at the University of Southern Denmark in games & interactive technologies; focus areas Human-Computer/AI Interaction, data science, and AI. Deadline August 1.

when i was a teen, i needed glasses to see at a distance but i could still read a computer screen without them. now, i need them even for that. reliance on a tool atrophied my eyes.

i wish i could react to replies from Claude with emojis that don't act as a prompt that needs a whole ass reply. sometimes i just want to say thanks and leave it there.

a fun, if somewhat narcissistic, prompt to try out with a model you have some chat history with. the prompt could probably be refined.

Google getting desperate?

I vibe-coded an app called PostGrip to replace Vercel, Netlify, and similar platforms. I need your help testing it. Here is a link with instructions for setting up PostGrip in your environment. The setup instructions were written to be executed by coding agents like Claude Code or Codex.

i get what Microsoft is going for here, they don't mean addictive like candy, but addictive like solving puzzles. they want something Claude Code like. they should have used language like indispensable or gratifying.

this could be the overall shape of AI agents in the not too distant future, as employees you hire at a fixed cost to do work. that will slot right into existing workflows and companies.

you'll never guess what this is replying to

the writing style of Opus 4.8 is more obnoxious than ever before, writing tics piled atop each other such that even the most milquetoast functional documents are grating to read

i know no one wants to hear this but... some sort of training wheels replybots to enmesh users into the social graph before fading away could be an interesting solution to this problem

yep

By comparison the GEICO Gecko’s most recent Instagram post has 1,813 likes. Let’s get profile on him next.

not bad

Project Hail Mary was technically well executed but i didn't like the way it threaded the comedy and the drama. more seriously, it didn't pose any interesting question or idea for the viewer, there's no fat to chew on or wrestle with. that's necessary for great sci-fi.