Ryan

@RyanLoftin

#blockchain, #metaverse, #neuralink

North Carolina

Joined September 2008

118 Following

122 Followers

331 Posts

Ryan @RyanLoftin

25 days ago

@diol2n Happened to me with a Quest 3 at Walmart.

Ryan @RyanLoftin

2 months ago

@TRobinsonNewEra This guy is not a leader.

Ryan @RyanLoftin

5 months ago

@BretWeinstein One of the few people I follow with consistently thought through positions. It seems obvious to some of us that once AI masters general reasoning it will quickly out compete humans in every metric. The only way this doesn't occur is if we hit a sudden wall, this seems unlikely.

RyanLoftin retweeted

Jawwwn

@jawwwn_

6 months ago

Palantir cofounder @JTLonsdale: “You want infinite jobs? Ban farm tools. We’d all be poor and everything would be more expensive.” “The only way we get wealthier is doing more with less.” “It’s ironic we have this luddite fear— it’s the only way we solve our national debt crisis. It’s the only way the working and middle class can more easily afford healthcare, cost of living, nice homes.” “A lot of young people now, especially in cities, can’t afford the cost of living at all, they can’t afford to have kids. It’s a mess.” “You need to destroy some jobs for the country to do better.”

435

284K

Who to follow

KAIRO.VIP.

@thisismorphine1

https://t.co/YwOb3DGVtl. In real name Klavs La. Music producer / DJ based in Latvia. Draws inspiration from global cultures and local vibes, bringing a unique internat

Điền Quân

@inQun6

Điên quân mạng xã hội Check for airdrops here @linity_com

しゅん札幌タクドラ＠フロドライバーのコーディネートはこーでねーと

@tamesarerushun3

洋服と靴と眼鏡とお風呂♨️が好きすぎる札幌のフロタクシードライバーのつぶやき。ウエスト太目🐷

RyanLoftin retweeted

Andrej Karpathy

@karpathy

6 months ago

Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project. This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.: - It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work. - It found that the Value Embeddings really like regularization and I wasn't applying any (oops). - It found that my banded attention was too conservative (i forgot to tune it). - It found that AdamW betas were all messed up. - It tuned the weight decay schedule. - It tuned the network initialization. This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism. https://t.co/WAz8aIztKT All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges. And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

karpathy's tweet photo. Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, today I measured that the leaderboard's "Time to GPT-2" drops from 2.02 hours to 1.80 hours (~11% improvement), this will be the new leaderboard entry. So yes, these are real improvements and they make an actual difference. I am mildly surprised that my very first naive attempt already worked this well on top of what I thought was already a fairly manually well-tuned project.

This is a first for me because I am very used to doing the iterative optimization of neural network training manually. You come up with ideas, you implement them, you check if they work (better validation loss), you come up with new ideas based on that, you read some papers for inspiration, etc etc. This is the bread and butter of what I do daily for 2 decades. Seeing the agent do this entire workflow end-to-end and all by itself as it worked through approx. 700 changes autonomously is wild. It really looked at the sequence of results of experiments and used that to plan the next ones. It's not novel, ground-breaking "research" (yet), but all the adjustments are "real", I didn't find them manually previously, and they stack up and actually improved nanochat. Among the bigger things e.g.:

- It noticed an oversight that my parameterless QKnorm didn't have a scaler multiplier attached, so my attention was too diffuse. The agent found multipliers to sharpen it, pointing to future work.
- It found that the Value Embeddings really like regularization and I wasn't applying any (oops).
- It found that my banded attention was too conservative (i forgot to tune it).
- It found that AdamW betas were all messed up.
- It tuned the weight decay schedule.
- It tuned the network initialization.

This is on top of all the tuning I've already done over a good amount of time. The exact commit is here, from this "round 1" of autoresearch. I am going to kick off "round 2", and in parallel I am looking at how multiple agents can collaborate to unlock parallelism.
https://t.co/WAz8aIztKT

All LLM frontier labs will do this. It's the final boss battle. It's a lot more complex at scale of course - you don't just have a single train. py file to tune. But doing it is "just engineering" and it's going to work. You spin up a swarm of agents, you have them collaborate to tune smaller models, you promote the most promising ideas to increasingly larger scales, and humans (optionally) contribute on the edges.

And more generally, *any* metric you care about that is reasonably efficient to evaluate (or that has more efficient proxy metrics such as training a smaller network) can be autoresearched by an agent swarm. It's worth thinking about whether your problem falls into this bucket too.

953

19K

11K

RyanLoftin retweeted

Andrej Karpathy

@karpathy

6 months ago

I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the prompt (.md) - the AI agent iterates on the training code (.py) The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc. https://t.co/YCvOwwjOzF Part code, part sci-fi, and a pinch of psychosis :)

karpathy's tweet photo. I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then:

- the human iterates on the prompt (.md)
- the AI agent iterates on the training code (.py)

The goal is to engineer your agents to make the fastest research progress indefinitely and without any of your own involvement. In the image, every dot is a complete LLM training run that lasts exactly 5 minutes. The agent works in an autonomous loop on a git feature branch and accumulates git commits to the training script as it finds better settings (of lower validation loss by the end) of the neural network architecture, the optimizer, all the hyperparameters, etc. You can imagine comparing the research progress of different prompts, different agents, etc.

https://t.co/YCvOwwjOzF
Part code, part sci-fi, and a pinch of psychosis :)

28K

39K

11M

RyanLoftin retweeted

Antonio Regalado @antonioregalado

7 months ago

World's richest man and the world's biggest anti-aging promoter are in each other's replies. https://t.co/ryUd3Gywzn

113

464

139K

RyanLoftin retweeted

Trung Phan

@TrungTPhan

7 months ago

That DeepMind documentary (“The Thinking Game”) now has ~300m views on YouTube. The most replayed scene is a meeting when someone tells Demis that AlphaFold can “easily” predict all known (1-2B) protein sequences “in a month”. He looks up from his phone and says “Why don’t we just do that? That’s a great idea. We should just run every protein in existence and then release that.” That would ultimately lead to the Nobel Prize. Interesting backstory from WSJ on how they were able to capture the footage of the meeting: a former NFL Films director (Greg Kohs) had done some commercials for Google and they asked him to do the documentary for the famous 2016 AlphaGo match with Lee Sedol. After that project, Kohs realized he should just do a documentary on DeepMind (Google would own rights but he had full editorial final cut). He spent a ton of time with Demis and Co from 2018-2024, and that’s how he got that incredible moment of Demis and AlphaFold (including the moment they released it to the world). *** More from Ben Cohen at WSJ: https://t.co/snmnkTstiY

443

638K

Ryan @RyanLoftin

8 months ago

@financedystop The incentive is high enough on a bar this big to justify layered or composite fraud that can pass a gravity test, which is why dealers still drill it’s cheap insurance

148

Ryan @RyanLoftin

8 months ago

@JamesMcPherson @MIC_btc @financedystop The incentive is high enough on a bar this big to justify layered or composite fraud that can pass a gravity test, which is why dealers still drill it’s cheap insurance

RyanLoftin retweeted

Emad

@EMostaque

8 months ago

I bet we have a unified foundational theory of physics in the next year or two Will be very cool Will be similar to thermodynamics or special relativity, a top down theory of principle, not a constructive theory done by fitting loads of lagrangians

355

106

69K

Ryan @RyanLoftin

8 months ago

@davidasinclair @LaurieGreen36 @lifebiosciences Godspeed

Ryan @RyanLoftin

8 months ago

@078sky @vedicvenus_ Suck it up.

397

Ryan @RyanLoftin

8 months ago

https://t.co/rdi01CkmkF

RyanLoftin retweeted

Ryan Shea

@ryaneshea

9 months ago

Let me blow your mind on David Sacks’ accomplishments for a minute, because many (the NYT in particular) are not appreciating just how impactful, transformative and impartial he has been as the AI and Crypto Czar of the United States... Simply put, David Sacks is a force multiplier for American innovation. Here is just a short list of David Sacks’ many accomplishments alongside President Trump and colleagues since taking office on January 20, 2025: Executive Order 14170: Removing Barriers to American Leadership in Artificial Intelligence - January 23, 2025 - Expedited data center construction and AI system exports - Led to wage increases of 25-30% for construction works in related trades - Bolstered US competitiveness against China by fast-tracking 20 major AI export deals, preserving American technological edge in global supply chains Executive Order 14178: Strengthening American Leadership in Digital Financial Technology - January 23, 2025 - Directed the creation of a working group for a Strategic National Digital Asset Stockpile, including a potential Bitcoin reserve - Ended the “reign of terror” against crypto, sparking innovation and attracting $500 billion in commitments from firms like Oracle and Softbank - Positioned the US as the “crypto capital of the world” Executive Order 14233: Establishment of the Strategic Bitcoin Reserve and United States Digital Asset Stockpile - March 6, 2025 - Formalized the Strategic Bitcoin Reserve capitalized with approximately 200,000 seized BTC valued at $17 billion, ensuring no taxpayer costs and treating Bitcoin as a permanent national store of value - Established the United States Digital Asset Stockpile for forfeited non-Bitcoin assets like Ether and XRP, centralizing secure custody across agencies to prevent premature sales - Championed the reserve as a "digital Fort Knox," averting future losses like the $17 billion from prior disposals and solidifying US dominance in digital asset strategy Executive Order 14277: Advancing Artificial Intelligence Education for American Youth - April 23, 2025 - Oversaw the rollout of nationwide AI literacy programs in public schools, equipping over 10 million students with foundational skills in machine learning and ethical AI use - Forged partnerships with leading AI companies to develop free online curricula and teacher training, boosting enrollment in AI-related courses by 35% - Championed youth innovation challenges that engaged 50,000 participants, fostering the next generation of American AI pioneers and aligning with broader workforce development goals Middle East AI Diplomacy, May 2025 - Accompanied President Trump on a tour securing a $600 billion deal with Saudi Arabia for chips and data centers - Integrated allies into the US AI ecosystem and enhanced global partnerships - Facilitated technology transfer agreements that trained 5,000 regional engineers in US AI standards, fostering long-term alliances and countering adversarial influences in the Middle East Executive Order 14318: Accelerating Federal Permitting of Data Center Infrastructure - July 23, 2025 - Streamlined federal permitting processes to cut data center approval times by 60%, enabling rapid deployment of AI infrastructure nationwide - Coordinated with state governments to harmonize local regulations, unlocking 15 new mega-data center projects and creating 50,000 high-tech jobs - Monitored implementation to ensure environmental safeguards, balancing speed with sustainability in AI expansion Executive Order 14319: Preventing Woke AI in the Federal Government - July 23, 2025 - Enforced guidelines to eliminate biased AI models in government operations, ensuring merit-based systems and saving $200 million in compliance costs - Established audit protocols for all federal AI deployments, identifying and correcting over 200 instances of ideological bias in agency tools - Trained 15,000 federal employees on unbiased AI principles, promoting fairness and efficiency across departments Executive Order 14320: Promoting the Export of the American AI Technology Stack - July 23, 2025 - Negotiated export agreements with key allies, boosting US AI technology sales by $400 billion and countering foreign dominance in global markets - Developed streamlined licensing frameworks for AI hardware and software, reducing export barriers for 300+ American firms - Hosted international summits to build coalitions, securing commitments from 20 nations to prioritize US AI standards in their ecosystems America’s AI Action Plan, July 10, 2025 - Unveiled a comprehensive strategy with over 90 federal policy actions across three pillars: accelerating innovation, building AI infrastructure, and ensuring a sustainable ecosystem - Spurred private investments exceeding $1 trillion in AI hardware software and data centers, contributing to 4.2% GDP growth in Q3 2025 - Directed cross-agency implementation teams, resulting in 50+ pilot projects launched by Q4 that integrated AI into key sectors like energy and defense CLARITY Market Structure Bill, July 17, 2025 - Advocated for the CLARITY bill to provide regulatory clarity for crypto markets, passed by the house in July 2025 and now advancing in the senate with Trump’s support - Addressed market structure, reducing uncertainty that stifled innovation under prior administrations - Galvanized bipartisan Senate momentum, projecting $300 billion in new crypto investments upon full passage and solidifying regulatory predictability for startups GENIUS Act for Stablecoins, July 18, 2025 - Oversaw the passage and signing of the GENIUS Act, establishing a legal framework for stablecoins with transparency requirements and US asset backing - Revolutionized payment rails using blockchain technology, unlocked American crypto dominance, and fulfilled Trump’s campaign promise to make the US the global crypto hub - Catalyzed issuance of $150 billion in compliant stablecoins within three months, streamlining cross-border payments and boosting US financial innovation Executive Order 14355: Unlocking Cures for Pediatric Cancer With Artificial Intelligence - September 30, 2025 - Spearheaded an interagency AI task force to leverage machine learning for analyzing genomic data and clinical trials, accelerating the identification of targeted therapies - Secured $300 million in public-private funding to deploy AI models for personalized treatment predictions, reducing diagnostic timelines by 40% in pilot programs - Collaborated with medical institutions to integrate secure AI platforms, positioning the US as a leader in AI-driven oncology and delivering hope to families nationwide Executive Order 14363: Launching the Genesis Mission - November 24, 2025 - Helped launch the initiative to integrate AI into federal scientific research, national security, and workforce development - Unified agencies to build a secure AI platform from government datasets, aiming to accelerate discoveries, secure energy dominance, and strengthen the US’ lead on AI relative to China - Publicly championed it as a “once-in-a-lifetime” opportunity to supercharge innovation, mobilizing over 40,000 AI specialists and fostering public-private partnerships And of course there's much more. It should be self-evident that David Sacks is an American secret weapon on AI and crypto dominance. The US’s economy is stronger, it is expanding job opportunities, and it is better positioned to compete on AI and crypto because of him. Additionally, it's important to point out that he hasn’t favored certain companies over others and certainly hasn’t played favorites with his friends. On the contrary, he’s broadly supported the entire AI and crypto industries in ways that unlock value for everyday Americans. Both AI and crypto are important frontiers of American technology that we as a nation depend on to move our society forward and enable massive leaps in our quality of life. Thank you @davidsacks for your service 🙏. You are exactly what this country needs.

ryaneshea's tweet photo. Let me blow your mind on David Sacks’ accomplishments for a minute, because many (the NYT in particular) are not appreciating just how impactful, transformative and impartial he has been as the AI and Crypto Czar of the United States...

Simply put, David Sacks is a force multiplier for American innovation.

Here is just a short list of David Sacks’ many accomplishments alongside President Trump and colleagues since taking office on January 20, 2025:

Executive Order 14170: Removing Barriers to American Leadership in Artificial Intelligence - January 23, 2025
- Expedited data center construction and AI system exports
- Led to wage increases of 25-30% for construction works in related trades
- Bolstered US competitiveness against China by fast-tracking 20 major AI export deals, preserving American technological edge in global supply chains

Executive Order 14178: Strengthening American Leadership in Digital Financial Technology - January 23, 2025
- Directed the creation of a working group for a Strategic National Digital Asset Stockpile, including a potential Bitcoin reserve
- Ended the “reign of terror” against crypto, sparking innovation and attracting $500 billion in commitments from firms like Oracle and Softbank
- Positioned the US as the “crypto capital of the world”

Executive Order 14233: Establishment of the Strategic Bitcoin Reserve and United States Digital Asset Stockpile - March 6, 2025
- Formalized the Strategic Bitcoin Reserve capitalized with approximately 200,000 seized BTC valued at $17 billion, ensuring no taxpayer costs and treating Bitcoin as a permanent national store of value
- Established the United States Digital Asset Stockpile for forfeited non-Bitcoin assets like Ether and XRP, centralizing secure custody across agencies to prevent premature sales
- Championed the reserve as a "digital Fort Knox," averting future losses like the $17 billion from prior disposals and solidifying US dominance in digital asset strategy

Executive Order 14277: Advancing Artificial Intelligence Education for American Youth - April 23, 2025
- Oversaw the rollout of nationwide AI literacy programs in public schools, equipping over 10 million students with foundational skills in machine learning and ethical AI use
- Forged partnerships with leading AI companies to develop free online curricula and teacher training, boosting enrollment in AI-related courses by 35%
- Championed youth innovation challenges that engaged 50,000 participants, fostering the next generation of American AI pioneers and aligning with broader workforce development goals

Middle East AI Diplomacy, May 2025
- Accompanied President Trump on a tour securing a $600 billion deal with Saudi Arabia for chips and data centers
- Integrated allies into the US AI ecosystem and enhanced global partnerships
- Facilitated technology transfer agreements that trained 5,000 regional engineers in US AI standards, fostering long-term alliances and countering adversarial influences in the Middle East

Executive Order 14318: Accelerating Federal Permitting of Data Center Infrastructure - July 23, 2025
- Streamlined federal permitting processes to cut data center approval times by 60%, enabling rapid deployment of AI infrastructure nationwide
- Coordinated with state governments to harmonize local regulations, unlocking 15 new mega-data center projects and creating 50,000 high-tech jobs
- Monitored implementation to ensure environmental safeguards, balancing speed with sustainability in AI expansion

Executive Order 14319: Preventing Woke AI in the Federal Government - July 23, 2025
- Enforced guidelines to eliminate biased AI models in government operations, ensuring merit-based systems and saving $200 million in compliance costs
- Established audit protocols for all federal AI deployments, identifying and correcting over 200 instances of ideological bias in agency tools
- Trained 15,000 federal employees on unbiased AI principles, promoting fairness and efficiency across departments

Executive Order 14320: Promoting the Export of the American AI Technology Stack - July 23, 2025
- Negotiated export agreements with key allies, boosting US AI technology sales by $400 billion and countering foreign dominance in global markets
- Developed streamlined licensing frameworks for AI hardware and software, reducing export barriers for 300+ American firms
- Hosted international summits to build coalitions, securing commitments from 20 nations to prioritize US AI standards in their ecosystems

America’s AI Action Plan, July 10, 2025
- Unveiled a comprehensive strategy with over 90 federal policy actions across three pillars: accelerating innovation, building AI infrastructure, and ensuring a sustainable ecosystem
- Spurred private investments exceeding $1 trillion in AI hardware software and data centers, contributing to 4.2% GDP growth in Q3 2025
- Directed cross-agency implementation teams, resulting in 50+ pilot projects launched by Q4 that integrated AI into key sectors like energy and defense

CLARITY Market Structure Bill, July 17, 2025
- Advocated for the CLARITY bill to provide regulatory clarity for crypto markets, passed by the house in July 2025 and now advancing in the senate with Trump’s support
- Addressed market structure, reducing uncertainty that stifled innovation under prior administrations
- Galvanized bipartisan Senate momentum, projecting $300 billion in new crypto investments upon full passage and solidifying regulatory predictability for startups

GENIUS Act for Stablecoins, July 18, 2025
- Oversaw the passage and signing of the GENIUS Act, establishing a legal framework for stablecoins with transparency requirements and US asset backing
- Revolutionized payment rails using blockchain technology, unlocked American crypto dominance, and fulfilled Trump’s campaign promise to make the US the global crypto hub
- Catalyzed issuance of $150 billion in compliant stablecoins within three months, streamlining cross-border payments and boosting US financial innovation

Executive Order 14355: Unlocking Cures for Pediatric Cancer With Artificial Intelligence - September 30, 2025
- Spearheaded an interagency AI task force to leverage machine learning for analyzing genomic data and clinical trials, accelerating the identification of targeted therapies
- Secured $300 million in public-private funding to deploy AI models for personalized treatment predictions, reducing diagnostic timelines by 40% in pilot programs
- Collaborated with medical institutions to integrate secure AI platforms, positioning the US as a leader in AI-driven oncology and delivering hope to families nationwide

Executive Order 14363: Launching the Genesis Mission - November 24, 2025
- Helped launch the initiative to integrate AI into federal scientific research, national security, and workforce development
- Unified agencies to build a secure AI platform from government datasets, aiming to accelerate discoveries, secure energy dominance, and strengthen the US’ lead on AI relative to China
- Publicly championed it as a “once-in-a-lifetime” opportunity to supercharge innovation, mobilizing over 40,000 AI specialists and fostering public-private partnerships

And of course there's much more.

It should be self-evident that David Sacks is an American secret weapon on AI and crypto dominance.

The US’s economy is stronger, it is expanding job opportunities, and it is better positioned to compete on AI and crypto because of him.

Additionally, it's important to point out that he hasn’t favored certain companies over others and certainly hasn’t played favorites with his friends.

On the contrary, he’s broadly supported the entire AI and crypto industries in ways that unlock value for everyday Americans.

Both AI and crypto are important frontiers of American technology that we as a nation depend on to move our society forward and enable massive leaps in our quality of life.

Thank you @davidsacks for your service 🙏. You are exactly what this country needs.

215

607

Ryan @RyanLoftin

9 months ago

@theallinpod Why is having kids important? How do we get society to view it positively again?

Ryan @RyanLoftin

10 months ago

@ThoughtsStreams @hecubian_devil Could be light roast, it's much more acidic on your stomach. Might be worth trying a darker roast if you're still interested in giving it a shot.