Mind Crunches #31: Sensemaking The World
Glocalization, philosopher agents and Abundance
💡Meditations
On Why LLMs Are Not Good Scientists
It’s becoming increasingly clear that LLMs aren’t good at discovering new things aka doing science. But why is this the case? Recently, Thom Wolfe, Hugging Face co-founder, tweeted that LLMs are just “very good students” but not good researchers without an ability to challenge training data and propose novel ideas. I find this a convincing theory but a bit limited. Some other potential reasons I see are:
lack of tacit knowledge
lobotomy of LLMs to reduce hallucinations
transformers aren't the right architecture for novel discoveries
from a meta-science pov, maybe LLMs have made new discoveries but we don't know how to ask/prompt them.
Some potential solutions:
it takes a combo of LLMs (text + vision + speech) to interconnect to spark better inductive thinking
finetune LLMs only with hallucinations and complex theory datasets
the more advancements we make in human brain connectome mapping, the better we will be able to reverse engineer creative thinking
On Glocalization
I have been trying, for quite some time now, to understand what is Trump’s and Elon’s masterplan. There is a theory that says that there is NOT a masterplan and they are in a “testing” phase validating various hypotheses till something sticks (which I don’t believe it’s the case). There are other takes which I find more interesting like for example that they are preparing for a soon-to-come AGI world or that they impose an intended shock to the economy to fight inflation and refinance the US debt. Regardless of what the underlying motivations are, I am absolutely convinced that we already experience a new world order the first time after the (almost) global domination of post-Cold War liberal democracies. I call it Glocalization and it will be fueled by a combination of isolationism and blockchain technologies. I wrote a long Twitter essay about it (slightly edited below for Substack)
“I was born in 1987 which means that I was raised with the Globalization movement being the most important driving force shaping my (and my country's) reality. The Clinton presidency and the Internet played a pivotal role in accelerating this movement with landmark agreements like NAFTA and the push for China's entry into the World Trade Organization (WTO). The fact that the Internet was in every household (from US to Greece) reflected the consensus that open markets and global cooperation would drive economic growth and stability. This proved to be mostly true but it came with some heavy baggage and discontent on many different levels: cultural (weakening local traditions), economical (offshoring manufacturing) and organizational (fragile global supply chains) among many others.
This discontent combined with post-Covid effects, geopolitical tensions and rapid AI growth have led us to the domination of a populist, anti-globalization sentiment with the resurgence of nationalist and protectionist policies (shielding domestic industries and limiting foreign economic influence) across the world. It could be argued that the era of Globalization is over and we are rapidly moving to a full Isolationist mode with trade barriers, investment restrictions, closed borders and data localization requirements.
Although I believe that this is a very likely scenario, I am convinced that we are currently witnessing the birth of a new sociopolitical movement, enabled by emerging technologies, that I call Glocalization.
What Is Glocalization & Why It Needs Crypto
The main ideological premise of Glocalization is a delicate balance between sovereignty and plurality. In the Glocal era, everyone -from nation states to individuals- will be focusing on becoming self-dependent, securing and preserving their sovereignty while at the same time voluntarily participating in larger communities, associations and organizations without sacrificing their autonomy. Unlike traditional Globalization, which can erode local traditions and concentrate power in a few centralized institutions, Glocalization could empower communities to retain their unique cultural, economic, and political identities while still benefiting from global cooperation.
But here is the thing. For this to happen, we need the adoption of distributed systems, blockchains, and zero-knowledge technologies. The same way that the Internet was the vehicle for the spread of Globalization, Crypto could do the same for Glocalization. In a high-level, we need:
Decentralized, censorship-resistant infrastructure: a global, open, and permissionless infrastructure that is resistant to control or censorship by any single entity or government.
Self-sovereign identity and data ownership: Decentralized identity systems and blockchain-based data storage can give individuals more control over their personal data and digital identities.
Interoperability and composability: The modular, open nature of blockchains allows for a high degree of interoperability between local and global systems.
Glocalization on a National Level
It's evident by now that most countries will soon have locally trained LLMs, specialized legislation on social media and algorithmic feeds, firewall-enabled info monitoring systems, local supply chains, tariff wars/sanctions with other nations. In other words, the daily experiences of citizens will be curated and shaped by a combination of the Local Internet and Local Regulation in a way we have never seen post-Internet. (Isolation Agenda)
In such a scenario, the only technology that can provide a viable and co-existing alternative, aka a truly global and trustless network, is Crypto + ZK. This is not an abstract or a sci-fi concept. It's happening as we speak with Apple censoring its own AI models in China while at the same time Crypto enabling p2p intelligence networks.
In a way, Crypto + ZK is a next-gen version of the OS movement which has proven to be higher in the stack that Politics. The whole world got "uplifted" by DeepSeek models exactly because they were OS and anyone could run them locally. If they weren't OS , they wouldn't have reached so many millions of users and they would be definitely banned in the US and other countries. OS legitimized tech innovation. Crypto and ZK will truly turbo charge it.
Glocalization on an Individual Level
On an individual level, I am witnessing a very similar technological and cultural shift. For example, we are moving to a post-social media status with most users wanting to be part of networks and communities but without sacrificing their privacy/sovereignty while also having control of their data and experiences. This is only possible with a combination of blockchain and ZK technologies. We already see decentralized social networks enabling users to pick the algorithmic feeds they want, own their personal data and monetize their creativity without a cut from middlemen.
We also see citizens owning their personal data and identities being able to use them as they see fit while still being law abiding citizens. This is happening as we speak in Buenos Aires by Quark_ID. So it's more likely than not that we will have a combination of local Snapchats and Banks that are "corrected" to comply with local regulation and truly open and decentralized networks that won't be owned/censored by any local authority. These two could also harmonically co-exist. For example, Bridget recently wrote about how CeFi is embedding DeFi options to users providing a (what I'd call) Glocal Finance. The same could happen for your local data on your iPhone having the option to anonymize and share them in a blockchain for some $$.
Glocalization on an Agentic Level
Someone could argue that exactly the same notion of Glocalizaton applies to AI agents that are operating autonomously but they will still be part of Intelligence Swarms where they will collaborate and transact with other Agents while preserving their "sovereignty" and not exposing the private keys of their "human masters". As per above, the likeliest scenario is that there will be some Agents that are part of a monolithic system and trained by the same model while there will also be a network of agents tapping into open, collective intelligence and interoperate without sacrificing their autonomy.
Blockchains + ZK as the New Internet
All the cases listed above are not distant future scenarios. They are taking place and evolving as we speak. I am a techno-optimist and I truly believe that we can build a much brighter future if we build systems at the intersection of all the amazing technological innovations we have in our hands. But this is not a given. We, especially the ones working in Crypto, have to fight for it and build the future we want to see and which is only possible with Crypto. Blockchains can both democratize innovation and enable each and every person in the world to participate in a new headless economy. ZK can ensure this will happen in a secure and trustless way. Politics can play their own zero-sum games. Tech is upstream from that.”
On MCP & The Next AI Innovations
MCP is the hottest thing right now in AI and has taken everyone by storm proving that sometimes incremental improvements don’t come from innovation but from “glue-type” of products. It’s very clear to me that we are moving to a multi-agent world with the following implications:
the focus will shift on how the "master" agent will be evaluating/delegating/judging task-specific agents outcomes and even synthesizing them.
for an agent-to-agent ecosystem to succeed, we need micropayment rails and a new monetization model. SaaS won't cut it. We need an Outcome as a Service (OaaS) model (Dharmesh calls it Results as a Service) where agents will be able to submit a proof of their work or an Intelligent Oracle will verify it. Crypto solves both problems.
we will need some kind of zkMCP for enterprise data to interconnect with privacy
it's very interesting that we unlock new AI use cases just by applying established Empirical Science theories on how we build new LLMs. The Popper agent is just scratching the surface. What if we build Spinoza/Locke/Hume agents that will coordinate to discover new things? The same applies for Cognitive Science and Evolutionary Biology centered AI innovation. For example, CoT models were a manifestation of Kahneman's findings. What if we use visible CoT to reverse engineer Neuroscience the same way we are using Evo2 to better understand biology?
📌 Mind Crunches
A new school of thought is brewing inside the Democratic party. The keywords of their underlying philosophy are: Abundance, Common Sense and Building Mindset. Ezra Klein published this viral video ahead of his new book release, Matt Yglesias recently wrote the Common Sense manifesto, Fukuyama blogged about the Hamiltonian opportunity of Democrats, Noah Smith keeps dismissing Trump’s stupid policies, Progressives who love building start new policy thinktanks, tech elites announce Abundance funds. Nature is (slowly) healing.
The first fully AI-generated paper that has passed the same peer-review process that human researchers go through. By the amazing Sakana AI.
Why the Mediterranean is heating faster than other seas.
Semianalysis (probably the best tech research company) on why US and Europe are destined to miss the Robotics revolution.
“They only demonstrate actual working products and features. Until WWDC last year, that is”. (Smart) people are getting concerned Apple is in crisis.
The most to the point analysis of DOGE.
Cowen on why AI will take off slower than most think. A combo of Human O’Rings + regulation + historical data on emerging tech adoption. I agree.
This tweet spoke to my heart.
AI writing better CUDA that itself accelerates AI is one the most rational and craziest things out there.
A hot -but not irrational- take by Palladium on why a well functioning and competitive economy means that many businesses will go bankrupt. Evolutionary capitalism eulogy!
The excellent Ben James on why data centers will soon run in solar and gas.
Pareto works for everything in life. Do common things uncommonly well.
Plants can see. Intelligence is everywhere.
Dwarkesh is a personal inspiration of mine. Random guy starts a blog. He gains followers and is unapologetically curious. Cold emails Satya to do a podcast. Satya replies immediately. Life goals.
Ben Thompson on Deep Research. The more data that exists online about a query, the worse the result is. Why? Because Mediocristan.
Recommended Blog: Maximum Progress
Recommended Book: House of Huawei: The Secret History of China's Most Powerful Company
Recommended Podcast: Josh Wolfe on Farnam Street
Quote: “Nothing is more powerful than an idea whose time has come.”