Constellation Institute
15 posts
@ConstellOrg
Bringing experts and leaders together to navigate transformative AI
New paper: We finetuned models on documents that discuss an implausible claim and warn that the claim is false. Models ended up believing the claim! Examples: 1. Ed Sheeran won the Olympic 100m 2. Queen Elizabeth II wrote a Python graduate textbook
New paper: research agenda for secret loyalties Imagine a frontier model that has been trained to covertly advance a specific actor's interests (a nation-state, a CEO, an adversary). @joemkwon argues this is an urgent, neglected, and addressable problem. 🧵
❗️Only two days left to apply to the Astra Fellowship! Apps close EOD SUNDAY May 3rd, AoE. Astra's 5 months, fully funded, @ConstellOrg Berkeley 80%+ of our first cohort now work full-time in AI safety Mentors include Redwood, AI Futures, TruthfulAI, CoG, IAPS, RAND & more ⏬
❗️Only two days left to apply to the Astra Fellowship! Apps close EOD SUNDAY May 3rd, AoE. Astra's 5 months, fully funded, @ConstellOrg Berkeley 80%+ of our first cohort now work full-time in AI safety Mentors include Redwood, AI Futures, TruthfulAI, CoG, IAPS, RAND & more ⏬
🚀 Applications are now open: Constellation's Astra Fellowship 🚀 Fully funded, 5-month fellowship at our Berkeley research institute. Pair with mentors across empirical AI safety research, strategy, and governance at @ConstellOrg! 📅 Apply by May 3rd (begins Sep 2026) 🔗 constellation.org/programs/astra…
New paper: Can you prevent emergent misalignment with inoculation prompting, or by diluting bad data with good? Prior work suggests you can. We show the misalignment is still present but hiding. It is triggered by adding cues to prompts, evoking the bad data.
In 2017, there were a few dozen people working full time on AI safety. By 2025, there were more than a thousand — and the demand for talent is still accelerating. We badly need fieldbuilders who can find and develop that talent. A thread:
🚀 Applications are now open: Constellation's Astra Fellowship 🚀 Fully funded, 5-month fellowship at our Berkeley research institute. Pair with mentors across empirical AI safety research, strategy, and governance at @ConstellOrg! 📅 Apply by May 3rd (begins Sep 2026) 🔗 constellation.org/programs/astra…
🚨New paper! How safe and aligned is Kimi K2.5? We found concerning dual-use capabilities, sabotage and self-replication tendencies, political censorship on Chinese-language queries, and potential agentic misuse risks. (1/N)
