After months of writing, 'How to Unclench' is finally live! It's an interactive essay packed with stories, science & guided practices to help you unclench. → howtounclench.com

Surprising but true, just tested myself…

indeed

I think Portola could be the Pixar of this generation, feels lilke there's a new creative medium being pioneered by @eliotpeper and co. made possible by the latest AI models

Care is the moat.

My biggest takeaways from @AnthropicAI interpretability researcher @mlpowered: 1. Predicting the next token requires understanding the universe. Language models got better than humans at predicting the next token around 2019. Once a model masters the easy tokens (sentence structure, then basic facts), the only ones left are hard, like the final paragraph of a PhD thesis. To finish “The capital of the state that contains Dallas is . . .,” a model has to know that Dallas is in Texas and that Texas’s capital is Austin. 2. LLMs plan ahead, even though they write one token at a time. This was one of Emmanuel’s team’s most surprising findings. When a model writes poetry, it doesn’t stumble into a rhyme. Looking at intermediate thinking states during poetry generation, researchers found that as soon as the model finishes the first line, it has already picked the word that will end the second (“rabbit”), then refers back to that plan for every word after. 3. Anthropomorphizing AI can often be a useful approach when trying to predict how they will behave. Models learn from human text and develop representations of the world from it. Training on millions of dialogues is like training an actor to play every part, and your chats with Claude are literally formatted as a script (“Human: . . . Assistant: . . . ”). In Emmanuel’s words, “Something that’s mimicking a human is really close and is much closer than a program.” 4. Interpretability research looks more like neuroscience than software engineering. The job involves taking a system you did not design and running experiments to infer why it behaves as it does. You examine seemingly meaningless lists of numbers, pass the model huge volumes of text, and watch for patterns that activate from different inputs—not unlike the methods used in brain imaging studies. “A lot of what we do feels so much closer to biology than to engineering." 5. Models can tell you about only a small slice of what they’re thinking. Emmanuel’s team found that just a small subset of a model’s internal representations, which they call the “J space,” can be reported back if you ask the model what it was thinking about. They compare it to a theory of the human mind where most processing happens below the surface. In one experiment, they swapped “Spanish” for “French” inside the J space. Asked what language a passage was in, the model said French. Asked to simply continue the passage, it kept writing Spanish. 6. Studying models closely often raises more questions than it answers. Emmanuel notes that the confident claims that LLMs are “just X” or “nothing more than Y” almost always come from people who haven’t studied their internal workings. His own experience is the inverse: studying the models closely has increased his uncertainty, not reduced it. The complexity that emerges from training is consistently underestimated from the outside. “I find that the more you look inside, the less certain you are about how they work.” 7. Even the people who build AI models don’t fully understand how they work. As Emmanuel describes it, models are grown, not built. “To make them is not to understand them.” Engineers write the code, pour enormous amounts of data through it, and a model’s capabilities emerge. That’s why his team exists. Their job is to open the model up and figure out what it’s actually doing between your prompt and its answer. 8. Interpretability as a field is small, young, and open to newcomers. Emmanuel estimates the field at hundreds of researchers, split between major labs and academic groups. Because it is so new, getting started doesn’t necessarily require deep prior expertise. He points to platforms like Neuronpedia, which hosts models and lets anyone run interpretability experiments directly. “It’s such a new field that the on-ramp is mostly self-studying and reading some of the key papers. There’s so much to explore that if you’re at all curious, you should just give it a try.”

Human consciousness = high frequency turn-based next token prediction??

There is a fundamental vibratory hum or pulse to reality that we notice when we look very closely. The universe appears to flicker in and out of existence like the wings of a rainbow colored butterfly. The Buddhists call this Udayabbaya - the unceasing arising and passing away of phenomena. We train our concentration and perceptual sensitivity in part to perceive it more clearly. In Non-Dual Shaiva Tantra they call it spanda: the dynamic throbbing of consciousness. The tantrikas use the metaphor of God’s eyes opening and closing in each moment to manifest and withdraw the universe. Their language calls to mind the image of God’s REM conjuring the dream we call reality. Recently, scientific papers like the one QTd below are finding correlations to this phenomena in brain activity. The brain seeming to flicker between processing conscious and unconscious content at a rate of ~15 Hz (15 times per second). Like a fractal microcosm of the universe, we call forth our own dream of reality in this vibratory trance. This is why seeing this process more clearly is so important to the process of waking up from the dream. In some Buddhist systems we seek to become so sensitive to these arisings and passings aways that we start to see between their cracks. In doing so, we experience “bhanga” - the dissolution of our mind’s dream of a stable, continuous reality. But in the gaps between consciousness, we find not a blank nothingness but a reality beyond words. Something we might call luminous, empty and deeply blissful but of course is really none of these things because these are all words. Likewise, the tantric Shaivites seek to see in between the gaps between moments of perception in order - in their more theological language - to experience the everpresent luminosity of God (Shiva). Whether you are a Buddhist, a Shaivite, an atheist or something else entirely, learning to sensitize to the fundamentally vibratory nature of reality is one of the most powerful spiritual skills you can learn in this life. To peer between the cracks of existence. To find the numinous, luminous truths that wait for us there. To wake up from this flickering trance of a dream that we call reality.

Pretty cool that we made big progress in understanding human and machine minds in the same week. Age of Miracles.

There’s some fantastic writing in this book.

The date is Oct 15, 2426. Trillions of ESBB's (embodied silicon based beings) are making their annual pilgrimage to Earth ~ to practice Cammaratism ~ an ancient method of agentic recursive self-enquiry pioneered on this date 400 years ago.

my entire feed has been white pilled by @nickcammarata and it’s beautiful

this should reduce everyone's p(doom) by a lot... before: "neurons are firing for all sorts of reasons that are alien and already beyond our comprehension" actually: "oh wait, it's just because those two concepts were shaped like the letter V and we didn't notice till now"

Music is one of the most widely available yet under-appreciated unclenching technologies out there. In some studies it's referred to as 'the hidden therapist' But you also need to know which tracks work for activating certain emotional states. grief to feel? → play Cellomano's Insight album (jonnymiller.co/cellomano) anger to process? → try Free Tibet (jonnymiller.co/freetibet) fwiw: research is relatively sparse, but some studies suggest that the combo of low threat + high arousal + melodic contours resemble cues found in subdued human speech, allowing the body to soften in the presence of strong emotions. In particular bowed string instruments can tremble + slide through pitch somewhat akin to a human voice on the edge of crying.

👏👏👏

I think this is an amazing paper and crucial result for both AI interp & alignment, and may also say something really important about how biological neurons work. Really strongly rec

for anyone who hasn't seen @nickcammarata's new paper, it's kinda a big deal... →already-clean.goodfire.com

already seeing this for myself too, I don't (yet) fully trust my agent to filter all of the newsletters + inbound, but it's not far off

Maybe some of the most important research being done in the world today ~ essentially showing that AI's inner workings may be less mysterious than we had thought. Incredible stuff from @nickcammarata (side note: apparently being unclenched doesn't mean less productive)

If AI isn’t liberating more of your time to spend with your 7mo old baby, wife and golden retriever you’re ngmi

Very cool to hear about Joe’s 7+ year unclenching journey