egil

it’s funny because this is genuinely the best approach Simply not thinking about stressful situations makes you less stressed

Jeff Bezos: "If I do my job right, the value to society and civilization from my for-profit companies will be much, much larger than the good that I do with my charitable giving."

If you are a mathematician, then you may want to make sure you are sitting down before reading further.

Since joining OpenAI the amount of congressional staffers that've (very kindly and politely) reached out abt careers in AI/tech from offices whose Reps/Senators rail against AI/tech/infra is...notable. Tbc, there's absolutely nothing wrong with that and I'm always happy to chat and help people connect with opportunities/networking etc. I didn't agree with the electeds I worked for on everything either, but the divisions there feel a lot wider than on most issues.

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.

Evercore introduced their token consumption model and implications for data center demand/capex today: “Base Case: In our base case, we see total annual token demand growing from ~100 quadrillion in 2026 to ~4 quintillion in 2030, representing a ~150% 4-year CAGR. Key assumptions here include ~1.5B AI consumer users growing to ~3.7B by 2030 and ~50M AI agents growing to ~800M by 2030. 2) Upside Case: Total annual token demand grows from ~156 quadrillion in 2026 to ~8.3 quintillion in 2030, representing ~170% 4-year CAGR. Key assumptions here include ~1.8B consumer AI users growing to ~4.4B by 2030 and ~100M AI agents growing to ~1B by 2030. 3) Downside Case: Total annual token demand grows from ~71 quadrillion in 2026 to ~2 quintillion in 2030, representing a ~131% 4-year CAGR. Key assumptions here include ~1.3B consumer AI users growing to ~1.9B by 2030 and ~40M AI agents growing to ~640M by 2030. The logical question is what future token usage will represent for data center demand – this will vary depending on average install base throughput (Tokens per watt per MW or TPS/MW which also depends on Interactivity or TPS/user) but assuming ~4.0 quintillion annual token consumption in 2030 and a blended install base throughput of ~500,000 TPS/MW, we see demand for data center capacity reaching ~250 GW by 2030, which should provide tailwinds for our IT Hardware/Networking & Data Centers coverage. Net/net: We see growth in AI token consumption driving durable multi-year demand tailwinds for suppliers of DC/AI infra within our coverage.”

Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new family of constructions that performs better. This marks the first time AI has autonomously solved a prominent open problem central to a field of mathematics.

Thank you. The important part is zeroing out taxes on the bottom half. Best way to put money in someone’s pocket is to not take it out in the first place. Bottom half is only 3% of total tax revenue. But it’s very meaningful to that person. Zero it out.

A 5-gallon water cooler jug of water produces about $132 in economic output from data centers, but only about 2 cents from almonds.

Personal update: I've joined Anthropic. I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D. I remain deeply passionate about education and plan to resume my work on it in time.