AI Safety Institute
@AISafetyInst
Followers
4K
Following
6
Statuses
92
We’re building a team of world leading talent to tackle some of the biggest challenges in AI safety - come and join us.
United Kingdom
Joined February 2024
Misuse safeguards play an important role in making AI safe - but evaluating how well they work is still an emerging field. Our latest work offers recommendations to accelerate progress and introduces a lightweight template for more effective evaluations.
0
2
18
Congratulations to @YoshuaBengio and team on the publication of the International AI Safety Report. This is major milestone in building international consensus on the science of AI safety.
Today, we are publishing the first-ever International AI Safety Report, backed by 30 countries and the OECD, UN, and EU. It summarises the state of the science on AI capabilities and risks, and how to mitigate those risks. 🧵 Link to full Report: 1/16
1
7
72
New research on the open challenges for machine unlearning in AI safety. A collaboration between @UniofOxford @Mila_Quebec @AISafetyInst and many others.
🚨 New Paper Alert: Open Problem in Machine Unlearning for AI Safety 🚨 Can AI truly "forget"? While unlearning promises data removal, controlling emergent capabilities is a inherent challenge. Here's why it matters: 👇 Paper: 1/8
1
6
35
RT @SciTechgovuk: The benefits of AI will be unleashed across the UK under a new AI Opportunities Action Plan. We’re taking forward 50 re…
0
84
0
Great to see our advisory board member @matthewclifford bringing out a hugely ambitious plan to back the UK as an AI leader. We're excited to contribute to the AI Opportunities Action Plan, supporting AI safety innovation in the UK and growing the ecosystem even further.
Artificial intelligence will be unleashed across the UK under government’s game-changing AI Opportunities Action Plan. Turbocharging economic growth. Creating jobs. Making the UK the number one place for AI firms to invest. @matthewclifford explains how 👇
1
8
64
Our new technical report details the results of our pre-deployment testing of @OpenAI's o1 model with the U.S. AI Safety Institute. Read more ⬇️
4
15
76
Today is our submission deadline for our bounty programme for model autonomy evaluations and agent scaffolds. Submit your initial idea by midnight anywhere in the world. If successful, you'll have another 2-3 months to build. Find out more:
1
4
14
🎉 Huge congratulations to AISI researcher @hannahrosekirk for winning a Best Paper Award at #NeurIPS2024! 🏆
A real honour and career dream that PRISM has won a @NeurIPSConf best paper award! 🌈 One year ago I was sat in a 13,000+ person audience of NeurIPs '23 having just finished data collection. Safe to say I've gone from feeling #stressed to very #blessed 😁
2
12
86
We've extended the deadline for our bounty programme for model autonomy evaluations and agent scaffolds to 14 December. If successful, you'll have another 2-3 months to build. Find out more:
0
3
15
LLM-powered scientific assistants come with many benefits, but they also come with risks. The question is: how do you measure them? We've developed a new methodology for assessing the usefulness of LLMs in science.
0
5
36
Our conference on frontier AI safety frameworks is underway in Berkeley! We're bringing together academics, civil servants, and AI companies to advance the state of safety frameworks globally.
1
7
53
If you're working on innovative solutions to manage AI risks in healthcare, finance, energy or other vital sectors, now is the time to apply! Applications close on November 26 2024. Learn more and submit your proposal here 👇 2/2
0
0
3
Spread the word, and apply here: Applications close on Thursday 28th November 2024. You must meet civil service nationality requirements. 3/3
0
0
0
We've released a technical report detailing our pre-deployment testing of @AnthropicAI's upgraded Claude 3.5 Model with the U.S. AI Safety Institute. Read our blog for a high-level overview.
1
23
151
Our new paper on safety cases, in collaboration with @GovAI_ shows how it’s possible to write safety cases for current systems, using existing techniques. We hope to see organisations using templates like this for their models.
1
7
40
We’ve partnered with @VectorInst and Arcadia Impact to develop evals on coding, maths, cybersecurity, safeguards and more. InspectEvals includes leading benchmarks and several agent benchmarks, which can now be run against any model with a single command. 2/2
0
1
7