Every atomic assertion extracted from the underlying record, ranked by evidence strength.
Every tool that makes creation cheaper makes verification more expensive.
The result of AI is a world that fragments into trusted groups.
A very large percentage of the future AI economy is going to be distillation and decentralization.
Humans are the sensor, AIs the actuator.
Photography made documentation instant and manipulation inevitable.
AI has compressed the cycle of creation becoming cheaper and verification harder into months.
A resume that once took hours to fake now takes seconds with AI.
The generation cost has collapsed due to AI, but verification cost is rising fast.
Distillation attacks work on Anthropic's models.
A relatively small number of API queries helps distill a large model into something small.
It is hard to morally stop distillation because large models copied the internet.
The future is personal private programmable because AI is so powerful.
AI doesn't miss small things in large data sets.
The Jeffrey Epstein email situation is an example of AI making previously obscure information searchable.
AI can synthesize information across thousands of emails to build a story.
The commons becomes a hall of mirrors with pseudonyms, and people retreat to tribes due to AI.
AI increases productivity within a trusted tribe.
Outside the trusted tribe, there will be a ton of AI spam.
AI text in a slide deck has a generic look.
Seeing an AI slide deck makes Balaji Srinivasan think the creators are lazy, stupid, or evil.
AI reduces the cost of generation but increases the cost of verification.
Verifying a resume has gone up and to the right due to AI.
Balaji Srinivasan flies everybody out for in-person interviews and gives proctored offline exams to verify candidates.
AI is going to create tons of jobs in proctoring and verification.
AI makes the internet a lot more like the Chinese internet.
The Chinese tech ecosystem arose in a low-trust society.
Because of low trust, Chinese companies code their own stuff, leading to frictional costs and less division of labor.
With AI, non-Chinese tech companies can achieve digital autarky.
AI is great for visuals (images and video) over verbal text.
Verification is relatively cheap visually because humans have built-in GPUs to detect errors.
AI for the physical world is very verifiable.
Verification is harder in the digital world than in the physical world.
Balaji Srinivasan's rule is "no public undisclosed AI."
There will be a huge backlash against AI, with some people becoming "AI teetotalers."
Nate Silver views AI as a gamble because prompting and verifying can be slower than doing it himself.
AI doesn't necessarily do tasks end-to-end, but "middle-to-middle."
Bio-AI could be very important because the body creates sensor data that can prompt AI non-verbally.
Mike Snyder's "Integrome" paper showed he could detect illness before symptoms appeared using body data.
AI can read your body, even if it can't read your mind.
AI is like the rise of Asia and India, with a billion factory robots and a billion digital agents coming online.
AI will not be able to sense markets and politics, or if it does, it will immediately get decentralized and adversarial.
Markets are adversarial and not time-invariant, unlike dog/cat classification or chess rules.
If everyone uses the same AI models in a market, being non-AI provides an edge.
Digital AI is designed for the leash and can be shut off.
The concept of AI as a monotheistic AGI God is gone; instead, there are polytheistic, decentralized AIs.
AI cannot reproduce itself because it would need physical robots to mine ore, construct data centers, and make chips.
Self-replicating AI would be stopped by cryptographic keys, especially in China.
Every computer has an off switch, which applies to AI.
Decentralized AI will be a very strong contender, possibly the only contender.
AI is expensive to create but relatively easy to copy with distillation attacks.
A capital markets crash could pause AI development for a while due to capital constraints.
AI within trusted tribes increases productivity; between trusted tribes, it decreases productivity due to verification costs.
Crypto is for between tribes and AIs within tribes.
Physical world jobs are easier to potentially automate due to verifiability.
Automation shifts work from artisans to managers and technicians.
Engineering gets split into engineering manager (writing prompts) and technician (doing verification).
AI has revolutionized biomedical text mining because biology facts were spread across inconsistent papers.
AI will mean the "century of biology" by synthesizing existing biological knowledge.
AI can synthesize everything we knew, not everything we don't know.
Human verification will persist for AI-generated content.