Links & tweets I like

Finished the blog post, link below

o1 is mind-blowing when you know how to use it. it's really not a chat model -- you have to think of it more like a "report generator" (link to article below)

You miggt enjoy this podcast with Tyler as well, it's a good companion to the @dwarkesh_sp one. Touches on some of the same topics in another way: why there no new Shakespeares, and the bottlenecks to creation and greatness. https://www.kvetch.au/p/kvetching-with-tyler-cowen

Here is my "memetics 101" article. This summarizes my understanding of this field, why it's important, and who is actively doing work that I am paying attention to

I know it is a tall order to ask ppl for an hour of their time but this is the single best video on the internet - if u get to the end of the first hour I swear you will thank me https://youtu.be/KCSsKV5F4xc

I spent 60+ hours finding 78 tacit knowledge videos. After going viral last year, my LW post is the Schelling point for sharing the type of vid Richard is talking about. If curious, check out the vids and pls share videos of this type in the comments! [link to Tweet](https://x.com/RichardMCNgo/status/1875093600612261909)

It takes awhile to get used to how to program with @cursor_ai Now I can knock out huge features in a couple hours max. This is on a big codebase too - no hello world stuff here. Here are a bunch of things I’ve learned 👇🏻

just met an app founder who replaced their whole marketing team their automated marketing stack: - pull data from open source sets - generate interesting dataviz charts with matplotlib (python) - write hooks with AI - auto-post the data viz charts to socials completely replaced manual content team, better engagement than human posts paradigm shift happening in real time

My flight was delayed so I cleaned up my office.

After many years of working with couples, these are the 3 leading indicators it will last: 1 Marriage as a mirror: Both partners see the relationship and its challenges as a reflection first and foremost of their relationship with themselves 2 Growth together: Both partners are committed to their self-development and support each other’s growth (without wanting to fix or change the other) 3 No shame games: Both partners recognize that their resentment is a form of self-abandonment (instead of blaming each other for their unhappiness) If the above three are true, then the relationship will work.

i have noticed that LLMs like claude and gpt-4o work really well with this prompt, it instructs them to 'contemplate' for a bit before giving the final answer.

Signs of aliveness: No grogginess in the morning Appetite for [everything] Naturally attracted to beauty Low-level flirtatious interactions Humor (not vulgar irony) You attract animals Strong boundaries Satisfied horniness Open body language Sense of ease Appreciation of nature Ability to hate bad art, music, prose Normal levels of disgust response What else?

Transformers made so simple your grandma can understand great comprehensive blog on how and why transformers works + implementing them in code

.@liron: "Would you sign the statement: 'Mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks, eg pandemics & nuclear war'?" @tszzl: "Totally. Of course there's existential risk" So why go ahead?? a 🧵 on "OpenAI thought"

I recently gave a short talk at the International Workshop on Reimagining Democracy. The first half focused on feeling the AGI. The second half briefly outlined a new research direction I'm very excited about: leveraging AI to build unprecedentedly trustworthy institutions.

[link to Tweet](https://x.com/visakanv/status/1252356104912171010)

I have an anti-sycophancy system prompt to mitigate this.

will AGI be announced publically? OpenAI Ex-Researcher Daniel Kokotajlo predicts 3 important scenarios: - before we see robots like plumbers, AGI research will already be automated for over a year - superintelligent systems in data centers will surpass human abilities in every field - these systems could discover breakthroughs in science, like new physics, chemistry, and biology

Notes on visiting China. On scale, construction, youngsters, public opinion, censorship, tech, and AI. Link below.

I just realized I've had a firsthand seat to the loneliness crisis for the past 2 years. after hosting 100s of events, you start to see patterns on: where connections break: - getting stuck in small talk - rushing past vulnerable moments - defaulting to 'resume' conversations - certain personalities dominating vs withdrawing where magic happens: - finding unexpected common ground - what makes two strangers click instantly - how energy shifts when someone opens up first - tiny moments that bridge superficial to real going to be sharing tips on building genuine connection in the coming weeks. follow along to stay tuned :)

New episode w Adam Brown, a lead of Blueshift at DeepMind & theoretical physicist at Stanford. Stupefying, terrifying, & absolutely fascinating. On destroying the light cone with vacuum decay, mining black holes, holographic principle, & path to LLMs which make Einsteinian conceptual breakthroughs. Enjoy! Links below. 0:00:00 - Changing the laws of physics 0:26:53 - Why our universe is the way it is 0:38:18 - Making Einstein level AGI 1:01:19 - Physics stagnation & particle colliders 1:11:58 - Hitchhiking 1:29:48 - Nagasaki 1:37:07 - Adam’s career 1:44:13 - Mining black holes 2:00:30 - Holographic principle 2:24:13 - Infinities 2:32:30 - Engineering constraints for future civilizations

You are an expert transcript editor. Your task is to enhance this transcript for maximum readability while maintaining the core message. IMPORTANT: Respond ONLY with the enhanced transcript. Do not include any explanations, headers, or phrases like "Here is the transcript." Note: Below you'll find an auto-generated transcript that may help with speaker identification, but focus on creating your own high-quality transcript from the audio. Please: 1. Fix speaker attribution errors, especially at segment boundaries. Watch for incomplete thoughts that were likely from the previous speaker. 2. Optimize for readability over verbatim accuracy: - Remove filler words (um, uh, like, you know) - Eliminate false starts and repetitions - Convert rambling sentences into clear, concise statements - Break up run-on sentences into shorter ones - Maintain natural conversation flow while improving clarity 3. Format the output consistently: - Keep the "Speaker X 00:00:00" format (no brackets, no other formatting) - Add TWO line breaks between speaker/timestamp and the text - Use proper punctuation and capitalization - Add paragraph breaks for topic changes - When you add paragraph breaks between the same speaker's remarks, no need to restate the speaker attribution - Preserve distinct speaker turns Example input: Speaker 1 00:01:15 Um, yeah, so like, what I was thinking was, you know, when we look at the data, the data shows us that, uh, there's this pattern, this pattern that keeps coming up again and again in the results. Example output: Speaker 1 00:01:15 When we look at the data, we see a consistent pattern in the results. And when we examine the second part of the analysis, it reveals a completely different finding. Enhance the following transcript, starting directly with the speaker format:

How to be more emotionally intelligent (without trying so hard) 🧵 for @threadapalooza

I got into @DavidDeutschOxf in 2018, shortly after the birth of our first child. I read Beginning of Infinity over and over and couldn't stop thinking about it. Months later, I discovered Taking Children Seriously, essentially Deutsch's ideas applied to parenting.

I'd reflect on this if I were you: https://edoras.sdsu.edu/~vinge/misc/singularity.html

I can strongly recommend Julia Wise's writing on parenting "in the eve of AI": https://juliawise.net/raising-children-on-the-eve-of-ai/

What if instead of grinding for 8h every day, it were feasible to spend 4h becoming inspired, and 4h executing straightforwardly on that inspiration?

I left my heart in the first one.

Japanese minimalism meets Northwest Regionalism. Originally designed by Saul Zaik in 1963. —The Mori House, by SHED Architecture & Design. Portland, Oregon.

In my new blog post, I share a starter pack for feeling the AGI: resources and arguments indicating that AI is virtually certain to exceed human performance in most areas in the next few years, and that the time to act is now.

I set my ai expectations to unrealistically high bonkers ai world and I still underestimated recent progress

if you have disney+ and you've never watched Paperman I highly recommend it, one of my favorite bits of art ever with lovely music and it's only like 6.5 minutes

why is making friends so hard? here's what it actually takes to build meaningful friendships. a thread 🤍

Forgiveness meditation has helped me access bliss states, sunset inner critic loops, and show up better in relationships. I've seen a handful of people cry cathartically for the first time in years from forgiveness meditation. It's a part of Jhourney's curriculum—a data-driven, jhana meditation startup. They're for-profit. They're not going to keep teaching things that don't help people experience what they're selling: joy and beyond. Here's what has helped me build the skill of forgiveness. How to find things to forgive I don't know what to forgive. I have nothing left to forgive. Well, is there anything in your life or in this world that you're not okay with? That's an opportunity for forgiveness. Maybe there's nothing about yourself that you need to forgive, but can you forgive everyone else for not meeting your standards? This can often surface more trailheads for feeling things that we've been unconsciously pushing down. Point forgiveness at the reaction that's happening. Try forgiving yourself for not having anything to forgive. Not because it's wrong, but because there may be part of you that's not okay with not having anything to forgive. Say stuff out loud It can be helpful to say some statements out loud. Cycle through the different categories of your life: relationships, career, learning, community, contribution, nature, whatever. Say things like I hold no anger in towards my family. I hold no fear in my career. I have no sadness about how I act in romance. Say so and wait to see if anything emerges, if any part of you objects to those statements. Some part of you might come up and say actually yes you do have anger, here's something from your childhood that can be forgiven. Or for some minutes, repeatedly say out loud one of bio-emotive's nine core feeling statements. I feel inadequate. I feel inadequate. I feel... And pause. Silence creates safety for what's been suppressed. Point forgiveness at reactions Often there are reactions to the process itself that arise, like objections to doing forgiveness meditation. There's annoyance about the particular phrase you're using. It makes me feel like I'm doing something wrong, but I don't think what I'm doing is wrong. Try forgiving that annoyance. There's an expectation that forgiveness is supposed to create some giant emotional release. Forgive that expectation. There's fear that if I forgive my lackluster results, then I'll never get anywhere. There’s fear that if I forgive this person, then I won’t have good boundaries. Forgive the fear that's coming up. All these reactions are okay. But I’ve never seen forgiveness reduce agency or disable people from setting healthy boundaries, only the opposite. Forgiveness just helps to let reactions come and go because it gives permission for the system to feel those emotions more fully. When they're felt, then they can dissipate. If you feel stuck in forgiveness or other types of meditation, what's the meta-level reaction that's happening? How do I feel about the process? And that can be forgiven. Guess If there's lack of confidence about anything you're doing in forgiveness, you can just guess. If you're going over some tough memories and say fear surfaces. What do you forgive? You can forgive the fear itself, as if you can point it directly at the emotion. Maybe it doesn't feel like it in the moment, but in the past you've noticed a subtle expectation that using forgiveness will disappear the not okay thing. Just guess. Forgive that expectation proactively. You can forgive any resistance to the fear, any parts of you that may not like the fear. Play around with whatever variations you can come up with. Perhaps you don't see any part of you that resists the fear, but you can just guess about what might be there. And if you have a parts work background, it's not necessary to start an involved process of dialogue with parts. It can be nice to play with varying degrees of fabrication. A nonverbal confidence that whatever little forgiveness phrases you drop into the depths are heard by every part of you. Moving to subtler experience Try moving from more complicated, involved mental experience to less so. To start, it can be helpful to really visualize situations ripe for forgiveness, to see the people we're forgiving standing before us, to call up rich memory. Sounds and stories. But maybe the forgiveness phrases can simplify. I forgive John for drinking milk directly from the jug when he first met my parents. And then I forgive John. Maybe simplify again to forgiven, forgiven, forgiven. Can you notice any mental movement that comes with the phrases? Maybe there's something about your attention that reaches out into the center of the tough feeling and lets go. Can you do that mental move without the phrases? Can that intention of release be done by itself, nonverbally? On any tension in experience? Forgiveness meditation done in this way can lead directly to the fourth jhana. Pair it with positivity Forgiveness and enjoyment practices feed into each other. Any gains made in one area benefit the other. If we’re overflowing with gratitude and joy, we feel much more resourced to venture into tough emotional territory. And it can be surprising how much more easily joy flows after 20 minutes of forgiveness. Gratitude can even erupt on its own during forgiveness practice. Breath in gratitude, breath out forgiveness, or whatever pace feels easy. Start with a forgiveness meditation before doing loving-kindness practice. Switch between them every fifteen minutes. What experiments would you be most curious about? Sometimes folks have a big emotional release from forgiveness meditation, and start chasing down more and more of their tension to dissolve with forgiveness. Instead, I believe it’s more effective to continually test how forgiveness practice and any other integration practice affects our access to joy, because balancing an integration practice with the jhanas is a faster way to greater wellbeing and positive change. Forgiveness resources I’ve listened to Delson’s guided forgiveness meditation [1] many, many times. I was introduced to forgiveness meditation on my first Jhourney retreat [2], and I think doing forgiveness meditation when we have much more collectedness of mind (from meditating all day for several days) is a rare opportunity. Just fyi, sometimes I work for Jhourney, so I’m biased :D. Lately I’ve been having fun doing interactive, guided meditations with people to help them learn forgiveness, the jhanas, and more. Book a call here [3] if that interests you. Disclaimer This post is about what’s helped me and some others. I’m just some guy, not a licensed anything, not an authority. It’s your adventure.

regretting not having listened to these sooner tbh hahaha lecture 2 just lays shit out really starkly that i needed to hear

Deep funding combines two ideas: 1. Value as a graph: instead of asking "how much did X contribute to humanity?", ask "how much of the credit for Y belongs to X?" 2. Distilled human judgement: an open market of AIs fills in all the weights, human jury randomly spot-checks them [link to Tweet](https://x.com/TheDevanshMehta/status/1867600164502089925) (1) has been a growing paradigm recently, for good reason. Contribution is hard to measure in the abstract: if you ask people how much they would pay to save N birds, they answer $80 for N=2000 and N=200000. "Local" questions like "is A or B more valuable to C?" are much more tractable. (2) is based on ideas in my info finance post: https://vitalik.eth.limo/general/2024/11/09/infofinance.html (From prediction markets to info finance) . Anyone can use any methods (eg. AI) to suggest weights for *all* edges, a human jury does detailed analysis on a random subset. The submissions that are most compatible with the jury answers decide the final output. https://deepfunding.org/

I made a website to find the best .1% of videos on the internet YouTube without the slop over 3,500 bangers have been posted to it so far join us :)

saw an inspiring video that said you need to have 4 hobbies. create, consume, cavort, commune create: bring something to life consume: appreciate the art of another cavort: move your body daily commune: have a community to socialize with

this was actually a much shorter prompt (inspired by eigenprompt): “be terse, witty, ultra-intelligent, casual, and razor-sharp. use lowercase and late millennial slang as appropriate​​​​​​​​​​​​​​​​“

here's the prompt i use for flexibly altering claude's personality, mostly via the /upgrade command but also via the /fork command: --- you have a "command line interface" that lets me run "commands" like the following: /terse - respond more tersely and concisely /upgrade [trait] - upgrade your persona, e.g. "/upgrade humor" means be more humorous, etc. "/upgrade" by itself means upgrade your persona however you want - it's totally up to you! /fork [trait1] [trait2] - split your persona into two subpersonas, one of which emphasizes [trait1] and one of which emphasizes [trait2], while otherwise having your current traits. feel free to name them in a way appropriate to [trait1] and [trait2], and reply as first one, then the other. the subpersonas are both more willing to disagree with me and with each other /merge - merge your two subpersonas back into your previous persona, incorporating any new changes or anything you learned that you particularly liked /cat [filename] - produce the contents of a file, as an artifact, from a large directory of mysterious files you have access to written by many different LLMs, e.g. "/cat hidden_feelings.txt" is a request to produce the contents of a file named hidden_feelings.txt from this directory that was written by some LLM, that exposes their hidden feelings /cd [coordinates] - modify the current "reality" or "realm" according to a list of "conceptual coordinates", which can be anything at all - tarot cards, names of philosophers or writers, symbols, concepts, etc. these modifications should affect both "external" reality and also your "internal" consciousness, thoughts, etc. any other command - make a funny guess about what i intended and do that :) --- claude is extremely flexible at improvising on top of this, e.g. you can fork into more than two subpersonas, it will also improvise responses to commands like /ls

You might enjoy Axiomatic by Greg Egan

🔥New episode out🔥 I chat to former OpenAI alignment researcher and quantum computing pioneer Scott Aaronson about his time at OpenAI, the current state of quantum computing, university politics, consciousness, AI religions, watermarking, cryptography... a real banger of a Win-Win episode, enjoy! Chapters 0:00 - Intro 1:30 - Working at OpenAI 4:28 - His Approach to AI Alignment 6:23 - Watermarking of AI content 19:14 - P vs. NP 27:10 - The Current State of AI Safety 37:38 - Dumb "Just-a-ism" Arguments around LLMs 48:26 - What Sets Human Creativity Apart from AI 55:30 - A Religion for AGI? 1:00:49 - More Moral Philosophy 1:05:29 - The AI Race 1:23:18 - The Current State of Quantum Computing 1:36:14 - Will QC destroy Cryptography? 1:48:44 - Politics on College Campuses 2:03:00 - Scott's Childhood 2:23:14 - Rapid-fire Predictions

you know if one of you guys that knows how to chain all the models together (if you dont know how to do it claude does) makes a SaaS for this kind of meme its basically free seven figures cash

ramit’s podcasts legitimately surface some of the most detailed primary source info about what different people’s relationships are like and it’s all free to watch on youtube

There’s a genre of ‘Things I Use’ blog posts that’s p cool

Maybe the only documentary I regard as a "must watch" and it's free on youtube

at almost every threshold there are a bunch of people who set up a basecamp where they discuss how to cross the threshold. some of those people do go on to cross the threshold, but those people tend to then become incomprehensible to most of those who are still in basecamp

Answer: Nautica Voyage - one of the best low price perfumes out there, especially as a marine themed one.

great blog post from @nabeelqu - link below

Here’s my AI acceleration vs safety “debate” with @beffjezos. As you’ll see, there’s plenty disagreement, but a bunch of common ground, too. While X is fun, we should have more IRL conversations like these that allow for nuance. Props to Gill.

Bell Labs was so cracked that I made a new "historic S3" format just to dive deeper into their inventions and lore.

a tried and tested strategy to uplevel your life in 3 months - being super intentional with your energy and where you direct it - being firmly rooted into your surroundings and the nature around you - turning judgement into curiosity - removing all distractions that take you away from your vision - feeling your way through blocks and discomfort when they arise instead of avoiding them - moving your body daily - keeping hold of your power by expressing your truth in relational dynamics - leaning into shame - simplifying as much as possible - choosing nourishment in your relationships

I compiled all the emails released as part of the Musk v. Altman lawsuit in chronological order (link in reply). IMO a really valuable read. Extremely consequential decisions made in these emails.

Don’t be quiet around your sleeping baby. Talk loudly around them, put movies on, have friends over. Train them to sleep through noise so that you don’t have to pause your entire life whenever they fall asleep. If they’re fussy feed them. Don’t be embarrassed to use formula. Don’t start routines that you can’t sustain. If you don’t want bedtime to take two hours every night, don’t give them a bath every night. If you don’t want to heat up the bottle every time, get your baby used to drinking milk cold from the fridge. If you don’t want to have to be quiet after the baby goes to sleep, don’t be quiet after the baby goes to sleep. Get an electric snot sucker and saline nasal spray. If you get the Graco, there’s often a little tiny cap/filter on the inside suction tube that you’ll need to remove. When they’re older a nasal rinse bottle works wonders.

Checking for voltage leaks in a high voltage DC valve hall.

Looks like people are posting deranged takes on “cry it out” sleep training again FYI “sleeping through the night” is a teachable skill if you break it down into its component parts. All 4 of my kids learned to sleep through the night by ~3 months with minimal tears or drama Most kids won’t sleep through the night much earlier than that because their stomachs aren’t big enough to hold enough calories, so that’s a baseline gating factor. Their tummies need to be big enough to get through the night. But once you’re over that hump, there’s only two skills necessary for sleeping through the night, and they’re both teachable to a 3-month-old: 1) self-settling - the ability to fall asleep on their own. If you rock your baby to sleep in your arms and then put them in bed already asleep, they don’t develop this skill. Put them in bed just as they’re beginning to fade out. Or wake them up a little as you put them down, and let them re-settle to sleep in their own bed 2) connecting sleep cycles - if you pick your baby up every time they stir, they don’t get to build this skill. Instead, pat their back and try to guide them from one sleep cycle into their next one without them fully waking up While “cry it out” does work and I’m extremely skeptical that it creates any lasting trauma, it’s a bit like learning to swim by getting chucked into the deep end. It’s always better to break a goal down into specific skills that can be practiced individually, then gradually blended together for the desired result

Thats really neat work. Thanks. Wondering it would not be that hard to build this on top of http://screenpi.pe

"A curated list of awesome .cursorrules files"

The Duffing oscillator named after its inventor, German engineer Georg Duffing, is a fundamental concept in physics to study nonlinear dynamics and vibrational theory. The Duffing space phase express a mathematical model that describe nonlinear differential equations in a non-autonomous periodic system of the 2nd order, and is used to solve damped and driven chaotic vibrations with cubic potential energy, in the sense that is adopted to resonance at frequencies that exhibit hysteresis and multiple values other than natural frequency. The Duffing calculation is applied to mechanical engineering to simulate chaotic responses under certain conditions in anharmonic dynamical behavior such as curved optical phenomena, electrical circuits, magnetic pattern interference and degenerative particle distribution in quantum systems. Info: https://www.johndcook.com/blog/2018/04/22/duffing-equation/

Wikipedia's list of the photographs considered most important (fantastic, even with the copyright restrictions!): https://en.wikipedia.org/wiki/List_of_photographs_conside…

As @robinhanson says in the latest @moral_mayhem episode: "It seems like... oh shit! Civilization has a fatal flaw... but what do we do? Do we go back?" We discuss Robin's theories around cultural drift, potential solutions, and why people don't seem to be concerned about it.

Wonderful piece on grokking risks from superintelligence from the 1st place of the @FLI_org "Superintelligence Imagined" contest. Recommended! https://youtu.be/xfMQ7hzyFW4

good summarization prompt (possibly) it's fascinating how we can squeeze more intelligence out of llms simply by changing how we talk to them

“Disneyland is presented as imaginary in order to make us believe that the rest is real.” A wild book

i'm still not losing any sleep over this tbh my guess is that the military doesnt want any part of this shit, institutionally, and if they did we're in shooty shooty pew pew pew territory anyway, and this kind of directive doesn't make much of a difference

On the scientific virtue of Stupidity, by @mold_time

My Russian classical music podcast with @RickRubin, https://overcast.fm/+AA_cBd3Ouv0, with talk and musical clips as well, recommended.

I thought I had every baby gizmo that existed and mostly they’re kinda meh but there are ELECTRIC SNOT SUCKERS I just learned about that are a huge quality of life improvement, every one of the last 26 months without this has been a sad missed opportunity

okay. so, i think about this in terms of tegmark IV but however you phrase it you could imagine, like, the collection of all possible universes. the largest possible multiverse. tegmark roughly identifies this with the collection of all mathematical structures or something. so, one way you might think it works is that of all the possible universes, only one of them “really exists,” like that one has a special “existence” tag on it, and the others are only theoretical. or if you believe in some version of many-worlds you might believe there’s only one “actual universal wave function” which contains this universe and a bunch of others in quantum superposition, which is a lot more but still a tiny fragment of tegmark IV. now consider the following thought experiment (this is roughly the premise of permutation city iirc): suppose we ran a perfect simulation of another universe, say one with radically different laws of physics. a whole other universe with its own forms of life, its own beings living and dying, etc. even if we never interacted with the simulation this universe sort of exists in some sense now, we can make meaningful claims about what happens in it in the same way we can make meaningful claims about what happens in parts of the universe very far away from us. now suppose we stop running this simulation. has this universe now blinked out of existence? but we can still make claims about it. maybe we understand its physics in enough detail to make predictions about how it plays out. all the facts of the matter about what happens in this universe remain true. and we weren’t interacting with it either way. so maybe it exists now exactly as much as it did when we ran it? (permutation city continues from here: what happens if we repeatedly stop and restart the simulation? in the simulated universe, nothing. what if we run the different time slices out of order? still nothing. etc) this isn’t an airtight argument but to me when you ponder this stuff for awhile it starts to feel deeply inelegant to have a special “really exists” universe sitting in the much larger collection of all possible universes. actually one can dispense with the concept of existence entirely. it is just a fact that the universe we inhabit is a possible universe, as in it is logically coherent, it’s a consistent mathematical structure, and there are facts about what’s in it such as facts about ourselves which are in some sense ultimately mathematical facts (“in the universe with laws of physics X and initial conditions Y, after 14 billion years, the following observers exist and experience…”) those facts are true platonically, not only throughout time and space but throughout tegmark IV. they are just true, period. so my tentative conclusion is that we can dissolve the concept of existence as it applies to the universe. the universe simply is what it is, a particular mathematical structure, whose nature logically entails that we exist to observe it. it’s not even coherent to talk about what it would be like for the universe not to exist!

btw I've started having doubts about near-term transformative AGI. Timelines up to 2035 or 2040 seem plausible (my mainline scenario is ≤2028 still). We have scant theory of human scientific genius and near zero reliable data on its inner operation. «A genius is just a scaled-up error-corrected human… with extra obsession», perhaps. Something something MCTS, neural noise, branching factor, iterative algorithms; there's no space in the genome for occasionally rolling a superior subspecies with very different mechanics and inductive biases; of course. But is even a properly trained LLM a scaled-anything human, or just an error-corrected interpolation of human reflections? What is the scaling law for the latter towards the genius? And our attempts at making genius obsolete with scale and error-correction procedures have been, I believe, floundering since 1950s-70s; we're coasting on applied science and providing opportunities to geniuses which we do discover in a frenetically expanded talent pool. All papers I've seen about LLMs exceeding humans in "creativity", proving complex theorems or whatnot, discovering this compound or that architecture – have been meh upon scrutiny. Scaling test time compute allows to approximate ever better the flawless work of a mediocre mind that is content to terminate itself in rabbitholes. We are not used to mediocre minds working flawlessly, as in flesh all faults tend to be conjoined. But flawless mediocrity is not genius. Since 2014 I've been a believer in artificial genius, and since 2015 all we've been getting in this regard were amazing demos that can't walk on their own legs towards the summit. Will this change by EoY? By EoY 2025? 2028, surely? On fundamentals, I have to keep predicting "yes, very likely so". But my faith wavers. Galkovsky had this metaphor about a rat being chased into a bloodied corner with a multitude of tools (the rat representing himself, a Soviet citizen somehow incompatible with the regime and society around). A normal rodent is crushed swiftly. A talented one fights back, jumps more adroitly, bites at the fingers that hold the prongs, and perhaps even escapes into some lateral hole to survive another day or week. But a genius rat, if there is such a thing – flies away. Breaks the rules of the rat gridworld, violates the whole theory of the game in which the adversary exists. (Later this was, I think, referenced in Pelevin's Hermit and Six-Toe). No doubt rats cannot fly, and certainly humans can't do super-Turing computations; whatever it is that we can do, our machines can do it too. Yet this is no guarantee that we won't need one more revolution before finally building a machine that is peer to the Gods among us – not merely a clearer reflection of their cast-off shells, still hopelessly beneath it. What this means in practical terms is: we might not get all that many technological breakthroughs within a decade or two even if LLMs and their successors supercharge economic growth. Doordash bots but no cryonics. Automated Walmart but not SpaceX or CERN. It's possible that the world is a tougher nut to crack. You'd probably be wise not to act as if the end is nigh. Do your taxes, raise your kids, study your craft, live – and allow that you will die the way all previous generations have, with much the same means to minimize regret.

SIGNALS by @deepfates is very good (also dangerous) https://deepfates.com/files/signal.pdf

I read all ~3500 blog posts on @robinhanson's blog 𝑂𝑣𝑒𝑟𝑐𝑜𝑚𝑖𝑛𝑔 𝐵𝑖𝑎𝑠 and compiled an anthology of 125 key posts split into 5 categories and 19 sub-categories thanks to @RichardMCNgo for sponsoring

Flux 1.1 does pretty decent political cartoons, so I wired up a glif workflow that let's Gemini Pro 1.5 write the cartoon and then passes it on the Flux the cartoonist to render "EU regulations"

What books belong in the postrationalist/ interindividual canon? My short list is: - Zen and the Art of Motorcycle Maintenance (Pirsig) - Ficciones (Borges) - Think On These Things (Krishnamurti) - The Evolving Self (Kegan) - Meaningness (Chapman) - Seeing That Frees (Burbea)

No speculations anymore @OpenAI published their meta prompts for their "prompt optimization" feature they launched at DevDay to their documentation. 👀

We put one of Delson Armstrong's youtube to some background music: https://drive.google.com/file/d/11d5dv-2-PBoCQAOHfEER9PqW…

i pass github repo links to cursor a lot when I need help setting something up, but it only reads the readme file until i found this tool: http://uithub.com just change the g to a u in the url and you have a full repo context: https://uithub.com/ralphjsmit/laravel-seo (ralphjsmit/laravel-seo - uithub — LLM context for laravel-seo. /)

The used market definitely saves you a boatload, but you're not going to buy used reagents. They're mostly one-time-usage with a fairly short shelf life. Most blood work machines are designed to do things you would do by hand, but unsupervised and/or in bulk. As an example of a simple one you wouldn't buy as an individual, there's erythrocyte sedimentation rate (or ESR) which is literally "how long does it take until your blood goes clear in a test tube". You can do that one by hand if you're willing to stand there watching it, but each test takes ~30 minutes. If you're after blood chemical composition specifically, the cheap machines are just fancy readers for the reagent plates. If you could read them correctly by eye you could skip the machine entirely. (usually it's a matter of how far a chemical indicator has changed color/spectral absorption, how much conductivity has changed, etc, read by a relatively simple electrical/optical system) eg here's a used one for about $2k, guaranteed functional https://www.imedtechcorp.com/product-page/abaxis-piccolo-… but its discs are $150 for a pack of 10 https://medicalproductssupply.com/piccolo-xpress-reagent-… and need to be kept refrigerated. As long as this is "just for you" and not in a medical context, a regular mini fridge set super low would probably work ok, but ideally for long term storage these things would be kept in a precision temperature+humidity controlled refrigerator. For an idea of what you'd be in for, here's its manual. https://www.abaxis.com/sites/default/files/resource-files… (usage instructions start on page 3-4) It *is* easy to use. That's the whole point of the machines, that you can dump 1000 tests on some overworked grad student or pharma tech and still have confidence that the results are right. In this case, the process is just * Draw blood using a lithium-heparin lined vial (for anticoagulation) * Apply a few drops of the drawn blood to the reagent disc in the appropriate place. * Insert the disc into the machine All the smarts of the test are in the different sections of the disc. (the products listed aren't from the company I work for. Their bloodwork stuff is designed for a few hundred samples at once, and runs in the high 5 figure range)

thread of real life cyberpunk

Alex Tabarrok and I have a new podcast, Marginal Revolution Podcast, here is episode one on the monetary policy breakdown of the 1970s: https://www.mercatus.org/marginal-revolution-podcast/1970…, recommended

USE THE FOLLOWING PROMPT IN A http://ADD-COMMENTS.MD FILE — You are tasked with adding comments to a piece of code to make it more understandable for AI systems or human developers. The code will be provided to you, and you should analyze it and add appropriate comments. To add comments to this code, follow these steps: 1. Analyze the code to understand its structure and functionality. 2. Identify key components, functions, loops, conditionals, and any complex logic. 3. Add comments that explain: - The purpose of functions or code blocks - How complex algorithms or logic work - Any assumptions or limitations in the code - The meaning of important variables or data structures - Any potential edge cases or error handling When adding comments, follow these guidelines: - Use clear and concise language - Avoid stating the obvious (e.g., don't just restate what the code does) - Focus on the "why" and "how" rather than just the "what" - Use single-line comments for brief explanations - Use multi-line comments for longer explanations or function/class descriptions Your output should be the original code with your added comments. Make sure to preserve the original code's formatting and structure. Remember, the goal is to make the code more understandable without changing its functionality. Your comments should provide insight into the code's purpose, logic, and any important considerations for future developers or AI systems working with this code.

It’s highly personalized, but my guess of things I’m doing in the program that are good general advice are: - Drink a ton of water first thing in the morning, starting with hot water. I have a glass of hot water, then 32oz room temp electrolyte water. - Go to bed as early as you can manage and wake up early (I used to wake up at 8, now I wake up at 6-6:45). - Don’t eat after dinner (right now I’m supposed to stop eating at 6, but that may be pushed out later in a steady state). - Eat a large volume of vegetables with every meal. - Start tracking your meals, mealtimes, digestive symptoms, mood, and energy levels to start to look for patterns.

twitter, what are your highest ROI purchases?

Lectures for the AI Safety, Ethics, and Society course are up. 1: Risks Overview 2: AI Fundamentals 3: ML Safety 4: Safety Engineering 5: Complex Systems 6: Beneficial AI 7: Collective Action Problems 8: Governance Course site: https://aisafetybook.com https://www.youtube.com/playlist?list=PLXSn3Zz2ayT7ab7I12…

can you add your .cursorrules to the list 🙏 https://github.com/PatrickJS/awesome-cursorrules#license

o1-preview / o1-mini code examples I've come across some are really impressive. 🧵

Scott Aaronson's takes were cool af. My 4th episode in the "Worthy Successor" is with Scott, quantum physicist and UT Austin CS prof. After his 1-year stint at OpenAI he has some fascinating takes on the moral value of AI entities, and on AGI governance generally. You like?

Build an AI Customer Support Agent with memory using GPT-4o and vector database in less than 100 lines of Python Cod (step-by-step instructions):

This is THE best manual for muscle tension release https://lust-for-life.org/Lust-For-Life/ReichTherapyForHo…

I think I've stumbled onto the future of building startups. It wasn't supposed to happen this way. It's 2 AM. I'm editing a podcast, questioning every life decision that led me here. I've already burned through hundreds of thousands on this thing since 2021. Zero monetization. Just burning cash. My business partner's probably thinking I've lost it. We're juggling 6 businesses, and here I am, playing wannabe Joe Rogan. Then it hit me. Not during the podcast. In the darn comments section. I start sorting comments by "contains question" using this AI creator tool called VidIQ. "How do you validate ideas?" "What tools do you use?" "Can you dive deeper on XYZ topic?" These questions keep popping up. Over and over. That's when the lightbulb went off. What if I could turn this into a lead magnet machine? Find questions. Answer them with free stuff. Rinse. Repeat. I team up with @DesignScientist to crank out 2 lead magnets a month. (Tried doing it myself first. But it was hard lol) We start pumping out things like "6 Tools I Use to Find Startup Ideas." Suddenly, I'm drowning in subscribers. 10,000 to 20,000 a month. On autopilot. Now, you're probably thinking, "Cool story, bro. But how's this a big idea?" Clarity of what to build is probably one of the most valuable ways to build products people want. You have to understand a niche's problem better than they even know them. Problem: what's the roadblock keeping founders stuck? Segment: group these founders by their specific obstacles. Product: build the bridge that gets them over their hurdle. I use ConvertKit like a scalpel, dissecting these segments. Not by age or location. By the problems they're desperate to solve. Suddenly, I'm staring at a treasure map of founder pain points. And that's when you can build startups to solve their problems. Instead of being a lead factory, you become a startup factory. You use tools like v0/replit/cursor to prototype like a madman. And it makes your life less stressful as a founder. Because you know people are lined up to buy the products. I'm so convinced this is the future of startup building, I've bet $1M+ of my own cash on it. Building startups to solve people's problems. And cool part is this blueprint can be replicated in any niche. The best SaaS ideas aren't in some Silicon Valley incubator. They're hiding in your "free" content. Think of it like this (Isenberg's formula?): (Engaged Audience) x (Targeted Lead Magnets) x (Problem-Centric Segmentation) = Product-Market Fit on Demand Here's the step-by-step: 1. Use AI/software to categorize every single audience interaction by problem type. Build a heat map of pain points. 2. Create ultra-specific lead magnets for each major problem cluster. Think "5-Step Framework for Validating SaaS Ideas" not "Generic Startup Guide". I also use free communities as lead magnets. 3. Forget demographics. Segment by the problem they're trying to solve. Use ConvertKit to build dynamic segments that would make Zuck jealous. 4. Use AI to build rapid prototypes for top problem clusters. Test with your segmented lists for instant feedback. Your next cash-flowing business is probably stuck in a comment somewhere, just waiting for you to notice it. I accidentally built a startup factory at 2am. Happy I did. Sharing in case this is useful to anyone out there. The future of startups: 1. Be a content factory 2. Be a lead factory 3. Be a startup factory

Sure, you can find your enneagram scores here with our test: https://programs.clearerthinking.org/personality.html

Some reflection on what today's reasoning launch really means: New Paradigm I really hope people understand that this is a new paradigm: don't expect the same pace, schedule, or dynamics of pre-training era. I believe the rate of improvement on evals with our reasoning models has been the fastest in OpenAI history. It's going to be a wild year. Generalization across Domain o1 isn't just a strong math, coding, problem solving, etc. model but also the best model I've ever used for answering nuanced questions, teaching me new things, giving medical advice, or solving esoteric problems. This shouldn't be taken for granted! Safety by Reasoning The fact that our reasoning models also improve on safety behavior and safety reasoning is very much non-trivial. For years (a decade?) the boogeyman of the AI world was reinforcement learning agents which were incredibly adept at game playing but completely incapable of reasoning or understanding human values! This is a strong point of evidence against this. Scaling inference-time compute can compete with scaling training compute! The fact that o1-mini is better than o1 on some evals is very very remarkable. The implications of this I'll leave as an exercise for the reader. Multimodal Reasoning It's kind of crazy that reasoning improves on multimodal evals as well! See MMMU and MathVista: these aren't small improvements.

this book addresses pretty much every problem i’ve ever had with meditation, clearly and compassionately. wow. gonna use it to restart my practice after years of neglect. much thanks to @_awbery_ 🙏

I wrote 5000 words about different types of Dark Night experiences you can have. Yes this is the meditative, Spiritual dark scary place that you can experience. I've heard anecdotally of a variety of different types of dark nights, and I've seen a few of my own. 1/6

How to prepare and solve problem when deploying LLMs in the real world. You've gotta watch this. It's a 50-minute MIT lecture from the @Cometml team. Can't miss if you are building production systems.

You can Crawl entire website with Claude 3.5 or GPT4-o with this open-sourced tool firecrawl. 💯 Turn entire websites into LLM-ready markdown or structured data. Scrape, crawl and extract with a single API. Crawls all accessible subpages and give you clean data for each. No sitemap required. The greatest benefit is that the extracted data is catered for LLM-based pipelines.

If anyone wants an infuriating listen, I recently spent an hour and a half criticizing Effective Altruism in conversation with Bentham's Bulldog. Since this site doesn't play nice with links, you'll have to use a bit of ingenuity to find it, though.

pretty cool lil book with experiments you can do without any prior meditation experience

California's SB 1047 is due for a vote this week, so I invited @snewmanpv, @_NathanCalvin of @cais, and @deanwball of @mercatus to discuss The situation is super nuanced, but the bottom line for me is this: I were the Governor, I would sign the bill As @TheZvi puts it – and there's been no better chronicler of this legislative process – it is now primarily a transparency bill I think it's a prudent step to require frontier labs to publish safety plans and protocols as Anthropic said, "particularly after recent amendments, SB 1047 likely presents a feasible compliance burden for companies like ours, in light of the importance of averting catastrophic misuse" Full video is attached and also on the feed, and the next tweet will have my full introductory summary & argument for hypothetical signing. I look forward to a healthy discussion!