Trilogues I is live!
@MaxAWebster joins @jesseposner & @dtdannen to explore why LLMs work better than they “should,” what language reveals about reality, and what AI reveals about the human mind:
Timestamps
0:00 - Introducing Max Webster & the first trilogue
2:32 - Is reality made of language?
8:10 - Do LLMs understand the world?
12:09 - Is language more than description?
18:51 - The hidden power of naming
25:28 - The language beneath everything
29:41 - How meaning changes life
36:57 - The limits of thinking & concepts
42:40 - The weird psychology of vibe coding & why it gets emotional
49:04 - Are LLMs enough for superintelligence?
58:20 - Could AI run without us?
1:06:18 - What AI reveals about us
1:12:24 - Is there something beyond language?
1:24:03 - Why imagination makes language powerful
1:30:15 - The infinite game after AI
There are two critical aspects of the art of tokenmaxxing and failing to understand the second part is catastrophic.
Whoever can consume the most tokens at the highest token ROI wins.
If you leave out the ROI part, you lose.
Companies are starting to question whether soaring AI spending is delivering meaningful returns.
An AI consultant tells us a client recently spent half a billion dollars in a month after failing to put usage limits on Claude licenses for employees. https://t.co/JHJ9Ojt9Hs
Silent payments aren't just about privacy. They also make addresses far more secure and usable because they enable human readable addresses: https://t.co/UStkPcAL5M
@Walodja1987 Exactly! So what is needed, especially for large transactions, is out-of-band authentication.
And the most secure and practical method of out-of-band authentication is a once per recipient flow using human readable addresses.
This reduces risks, and does not add new ones.
@Walodja1987 Did you read my post? You haven't addressed my questions: https://t.co/FmryRMfQB8
Expiration and authentication are a problem that exists in crypto right now, even without DNS, silent payments, or anything else.
When you receive a bitcoin address from someone, how do you know it came from them? How do you receive it? Via email, website, X, nostr, signal, SMS, etc.?
Those all can expire as well. Even a public key can "expire" because you don't know if the recipient still has their keys.
More importantly, those methods can all be man-in-the-middled unless you perform an out-of-band authentication.
But with BIP 353, you only need to perform an out-of-band authentication once per recipient, rather than per address.
When you receive a bitcoin address from someone, how do you know it came from them? How do you receive it? Via email, website, X, nostr, signal, SMS, etc.?
Those all can expire as well. Even a public key can "expire" because you don't know if the recipient still has their keys.
More importantly, those methods can all be man-in-the-middled unless you perform an out-of-band authentication.
But with BIP 353, you only need to perform an out-of-band authentication once per recipient, rather than per address.
Nested MuSig2 now has a Python reference implementation. @BeulahEvanjalin built it on secp256k1lab with tests, examples, and protocol notes—generally in a good state for review and discussion. This is the blueprint that will inform the secp256k1-based C work.
Episode 2 of Decentralizing AI is live!
@jesseposner and @dtdannen discuss why AI is your birthright (if you’re American or European), private/user-controlled models, and why the next phase of AI will need more than LLMs ↓
Timestamps:
0:00 - Why AI is your birthright
5:00 - The fight over training data
10:29 - Was AI invented or discovered?
18:31 - Why open models matter
27:34 - Building a better AI workflow
47:44 - Why agents need world models
56:52 - Choosing the right AI & why AI needs more than LLMs
1:11:51 - Coding as discovery not execution
1:13:56 - Can the world become language?
1:28:23 - How names unlock understanding
1:41:00 - Why private AI matters
1:47:30 - Will superintelligence = superwisdom?
Every year, we host a handful of technical workshops that bring together top talent to solve the bottlenecks in frontier science and technology.
Together with @protocollabs, we’re inviting leading researchers, engineers, cryptographers, and funders to Berlin, July 18-19, for a two-day workshop on how AI can become an engine of defense in a multipolar human-AI world.
We will explore:
• AI for Secure AI: Building self-improving defense systems where AI autonomously identifies vulnerabilities, generates formal proofs, red-teams, and strengthens our digital infrastructure.
• AI for Private AI: Constructing confidential compute environments, scaling encrypted mechanisms for handling data, and designing infrastructure that distributes trust.
• AI for Decentralized & Cooperative AI: Building decentralized intelligence ecosystems using tools in economics, game theory, and mechanism design, where AI systems cooperate, negotiate, and align.
Speakers include:
• @robinhanson, George Mason University
• @FazlBarez , University of Oxford
• @ml_sudo, Project Sovereign
• @DavideCrapis, Ethereum Foundation
• @NitzanShulman, Heron AI
• @jesseposner , Vora
• @AlexObadia, ARIA
• @jsotterbach, SPRIND
• @juanbenet, Protocol Labs
If you’re researcher or builder in this space, apply to attend: https://t.co/i9oJnHyFGX