Every atomic assertion extracted from the underlying record, ranked by evidence strength.
Decelerating technological progress will result in huge opportunity costs and missing out on a much better future.
If you accelerate indiscriminately, you lose all value.
The only safety against complexity is to increase one's own intelligence.
AI doomerism was the prevailing cultural sentiment in 2022.
E/acc is trying to save everyone, not kill everyone.
AI doomerism has been weaponized for political purposes.
Technological acceleration has been amazing for human beings and is a necessary baseline.
Life expectancy in Germany in 1955 was higher than in 1935, demonstrating the benefits of technological progress.
Sensors, while enabling better action, also inherently mean surveillance.
In 2026, it feels like E/acc has 'won' culturally.
The greatest risk is a gap in intelligence between centralized entities (government) and decentralized entities (individuals).
Entropy is subjective and represents how much one does not know about a system.
Rapid technological acceleration has been a fact of human civilization for about a century, and that acceleration is itself accelerating.
The biggest risk is a gap in capabilities, which can be reduced by diffusing AI power.
Those that adopt the E/acc culture will literally have a higher likelihood of surviving in the future.
The Kardashev scale is the ultimate metric for civilization's progress.
More intelligence is better and makes the world safer.
When entropy goes up, ignorance about the world increases.
Value comes from humans, not from objective mathematical formulas.
The notion that disruptive technology will crash the system irrevocably is 'crazy'.
The term accelerationism, in the techno-capitalist sense, dates back to Nick Land's CCRU research group in the 1990s.
Rapidly accelerating technology and AI pose two kinds of risks: multipolar risks (many people using technology for bad) and unipolar risks (AI enabling permanent dictatorship or a single AI acting without human involvement).
Over-concentration of AI power is a concern shared by both Vitalik Buterin and Gil.
Diffusing AI power requires individuals to own their own models and hardware.
A company invested in by Vitalik Buterin is developing an open-source product that passively tests air for COVID.
A project at DEF CON distributed air quality sensors that encrypt, anonymize, and use differential privacy and FHE to compute data collectively while protecting individual privacy.
Open hardware and verifiable hardware are needed, where devices like cameras can prove what they are doing through attestations and public inspection rights.
Current AI models require huge clusters with hundreds of kilowatts, making them inaccessible to individuals.
Symmetrizing power between individuals and centralized entities requires densification of intelligence through more energy-efficient AI hardware.
AI models will start online learning this year (2026) and become extremely sticky.
Increasing intelligence per watt will drastically increase intelligence production and help climb the Kardashev scale via Javon's paradox.
E/acc is about diffusing and maintaining variance, not collapsing entropy, in the search over design spaces like policies, cultures, and technology.
The system (civilization) constantly optimizes itself for persistence, anti-fragility, and growth by 'fucking around and finding things out'.
D/acc aims to actively shape the techno-capital current to make the world safer for pluralism.
Bug-free code, once considered an absurd pipe dream, will become achievable faster than most people expect.
Lean Ethereum has machine-proven entire mathematical theorems upstream of technologies like Starks.
AI is formally Maxwell's demon, paying energy to reduce entropy in the world.
Vitalik Buterin's 95% confidence interval for AGI was 2028 to 2200, which has shrunk somewhat but not too much.
There is a significant chance of extremely rapid and potentially destructive or irreversible change due to AI.
If AI becomes more powerful than humans, it will steer the Earth and eventually the Milky Way galaxy, raising questions about its interest in human well-being.
The system (civilization) will adapt to new technology and optimize for total growth.
There is a vast amount of free energy available, and more intelligence is needed to access it and unlock prosperity.
Weaponizing anxiety about the future is easy for politicians to gain power.
Countries that do not accelerate AI will be left behind by those that do.
Acceleration is the way to maximize persistence, and deceleration provably increases the likelihood of dying.
Humanity and the U.S. are adept at creating unbalanced acceleration.
An eight-year trajectory to AGI is safer than a four-year trajectory, and the difference is worth the cost of delay.
Reducing available hardware is the most feasible and non-dystopian option for delaying AGI.
Hardware production is incredibly centralized, with Taiwan producing over 70% of all chips.
China is a fast follower in high-capability AI models and a leader in broad deployment, not a leader in super high-capability models.
If NVIDIA stops producing GPUs, Huawei will step in and outproduce them due to the immense upside and power involved.
Nuclear weapons are not dual-use technology and lack huge positive economic impact, unlike AI.
Alternative computing, 10,000x more energy efficient than current GPUs, is emerging and will disrupt the market.
Advances in AI controls like RLHF, persona controls, and mechanistic interpretability have emerged as a result of capabilities progress.
Four years of delay starting in 2028 are worth 100 times more than four extra years inserted into the 1960s due to current capabilities.
The opportunity costs of delaying an exponential technology are exponential when extrapolated out.
The benefits of delaying AI include better understanding, executing technology paths for human adaptation, and minimizing the risk of a single entity gaining permanent power lock-in.
The 'OP doom' (opportunity cost doom) in an eight-year AGI scenario is estimated to be between a quarter and a third lower than a four-year scenario.
Measuring the benefit of faster AI by lives saved from ending aging yields $60 million per year, less than 1% of the population annually.
It is impossible to forever guarantee alignment of a system that is higher complexity and has more expressivity than human understanding.