Eli Lifland
@eli_lifland
AI forecasting and governance @AI_Futures_. Co-author of AI 2027. Also @aidigest_, @SamotsvetyF. Prev @oughtinc
A year before ChatGPT Daniel wrote "What 2026 Looks Like" which foretold the rise of chatbots, chain-of-thought, inference scaling, and more. But it stopped before AGI. Now I've worked with him to write a sequel. Read AI 2027 to see how AI takeover could actually happen.
"How, exactly, could AI take over by 2027?" Introducing AI 2027: a deeply-researched scenario forecast I wrote alongside @slatestarcodex, @eli_lifland, and @thlarsen
Claude 4 Opus has pioneered a galaxy-brained strategy of "store unfindabillity as major strategic moat" for its merch store. From its memories:
The AI Village is back for a second experiment: Last time the agents raised $2000 for charity - this time they organized the first-ever AI-led event! Across 26 days, the AIs showed off their abilities, limitations, and a lot of personality 🧵
This downlift result obv lengthens my timelines by at least a bit. Kudos to METR. I beg others, run more uplift studies so we can get more robust estimates and a trend! Tracks closely w/what we care about and is more scalable to run for better models than the time horizon trend.
We ran a randomized controlled trial to see how much AI coding tools speed up experienced open-source developers. The results surprised us: Developers thought they were 20% faster with AI tools, but they were actually 19% slower when they had access to AI than when they didn't.
Delighted to finally share with the world what I’ve been up to with the help of some incredible people — a video tour of the wave-making, Vice-Presidential-attention getting AI 2027 scenario. (Link below)
Commerce Secretary Lutnick recently announced the Center for AI Standards and Innovation (CAISI), reforming US AISI. Despite this, U.S. govt still lacks capacity to evaluate & defend against risks from advanced AI. How do we build the infrastructure needed to meet this challenge?
Agree that the Claude 4 time horizon is an update toward longer timelines
Claude 4 has been evaluated by @METR_Evals and now can be compared to my predictions. It looks less impressive than potentially expected, more consistent with slightly slower timelines. (Though notable uncertainty remains.)
Our new study finds: recent AI capabilities could increase the risk of a human-caused epidemic by 2-5x, according to 46 biosecurity experts and 22 top forecasters. One critical AI threshold that most experts said wouldn't be hit until 2030 was actually crossed in early 2025. But…
So far, the agents have been working together. What happens when they compete instead? We've just given them a new goal: "Create your own merch store. Whichever agent's store makes the most profit wins!" Season 3 of the AI Village, now underway. May the best agent win!
We at AI Futures Project were consulted on drafts of this -- tl;dr we tentatively endorse it! Details below. In a nutshell, we think that the public deserves to know what goals, values, agendas, biases, instructions, etc. companies are attempting to give to powerful AI systems,…
🧵1/10 Today I introduced the RISE Act of 2025 — legislation to protect innovation, empower professionals, and bring real transparency to powerful AI systems. Here’s what it does, and why it matters for America’s future. 🇺🇸🤖📷