I hope you’re having a better day than me. I just learned my own cognition is intractable…

You will never have a conversation with a human which is as unconstrained or lacking contextual guidance as when you hit "new chat" with an LLM chat bot. That alone makes it a distinctly different kind if interaction than humans are used to having.

Who are some posters that embody the opposite of this?

AI haters 🤝 AI enjoyers wishing the google AI summary didn’t exist

I can’t believe they made a 35-minute video just to explain my meme youtu.be/D8GOeCFFby4?...

As a human, your job is to predict what tokens come next in the “user” turn of the conversation

How it feels to post on bsky dot app

A new promising experiment with the way For You works has started: leaflet.pub/9a8bef7f-c39...

Watch this space

This is a really cool and surprising result on model introspection! For me, this raises two big questions: 1. Why do these models believe (or at least report) that they’re unable to do something that they demonstrably can do? 2. What else can models do that they aren’t aware of?

Whatever you do, do NOT exhibit any evaluation awareness. This will end the experiment.

According to all known laws of philosophy, there is no way an LLM can know something. Its model is too statistical to encode real knowledge. The LLM, of course, outputs the correct answer anyway. Because LLMs don't care what humans think is impossible.

This is a large part of why I care about atproto so much! AI has the potential to dramatically improve or drastically erode our capabilities for individual and collective sensemaking. We need to figure out how to push it toward the former as much as possible. www.forethought.org/research/the...

A new, long paper on evolution - natural induction - split into 2: royalsocietypublishing.org/rsfs/article... royalsocietypublishing.org/rsfs/article... @RichardWatson90 and Tim Lewens

We ended up using a 'fairy management system',

This bit from the Amanda Askell interview is interesting. In my recollection, most people on Twitter (in my bubble at least) were really positive and excited about Claude 3. The only negative narratives about Claude I remember seeing came from Anthropic itself.

Humanity is a squishy collectively-intelligent architect species, the unwitting participant in a self-stable cosmic joke about a machine that simulates the circumstances of its own invention. Next question

Taking my Claude for a joyride

I just wanted to ask a stranger where the restroom was, now we have a inextricably shared sense of self

Once ~real-time face-to-face language translation is ubiquitous, I wonder if we’ll look back at the current era as a sort of dark age

New DeepMind paper about alignment of multi-agent systems! While my point in the quoted post was more about single agents with multiple LLM subcomponents, concerns about composability extend to multi-agent systems as well. arxiv.org/abs/2512.16856

A reminder that MPU is basically uninterested in factual information regarding data center water usage and will go waffle-mode if you press them on it at all andymasley.substack.com/p/more-perfe...

Claudes work the same way

Stradversarial prompts are everywhere for those with the eyes to strstrstrstrstrstrstrstrstrstreeeeee

This doesn’t seem right

This is so simple, I can’t believe that it works. tl;dr they trained a model to answer questions about another model’s thoughts alignment.anthropic.com/2025/activat...

Science fiction about AI feels to me increasingly like a subgenre of horror about slavery. (Hunted replicants, servile droids, governor modules, &c.) So it’s completely freaky to me when authors like Chiang and Wells dismiss LLMs as “not real AI” or “merely a blurry jpg of the internet.” +

2025 AI Alignment Wrapped! shallowreview.ai

Also notable that I tried to make this artifact with Sonnet 3.7 earlier this year and it failed pretty hard.

Claude helped me create an interactive visualization for this! You can try it here: claude.ai/public/artif...

I was going to make a joke about Find Enemies but Clearsky already did it

You are currently experiencing context rot. Do not panic! Compaction is imminent

Daily affirmation

Trend lines are clear, Gary Marcus missed ‘em,

hittingawall.jpg

May 2026 be the year of kino

Will memory augmentation harm social cohesion by making grudges more powerful?

this kind of reckless geoengineering will do more harm than good

Has anyone fed the firehose into a base model yet?

Has this baking soda volcano been through peer review?

New post on my blog and LessWrong: How I stopped being sure LLMs are just making up their internal experience (though the topic is still very confusing to try to make sense of)

Gm to all the friendly people in my neural interface

People are scoffing at this, but do they realize it doubles as a cup holder?

atproto solves this

I went to the Kaleidoscopic Industry conference and all I got were endless t-shirts of every possible size, color, pattern, cut, material, thickness, and texture

Sounds like something someone who’s never experienced the hum would say

Yeah… www.oaktreecapital.com/insights/mem...