2024 Wrapped 🎁

As we come to the end of another rollercoaster year, we’re reflecting on a few of FLI’s most memorable moments. In no particular order, presenting our 2024 ‘greatest hits’:

1. EU AI Act made final. After years of work, the world’s first comprehensive AI legislation entered into force in August. With implementation taking place over the next few years, the Act will serve to support European AI innovation - especially by smaller companies and start-ups - whilst ensuring public safety as the top priority. Our AI Act Explorer tool was built to help navigate the Act; also be sure to check out the EU AI Act Newsletter!

2. Superintelligence Imagined. Our Superintelligence Imagined creative contest received over 180 submissions, culminating with six winners (including one grand prize winner) and seven honorable mentions illustrating the risks of superintelligence in approachable, creative ways. Check them out here!

3. Major support for SB 1047. California’s proposed SB 1047, the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act, gained incredible levels of bipartisan support - ranging from widespread public favor, to countless AI experts, labor unions, high-profile creatives, and more. The undeniable momentum generated by its supporters ensure it’s only a matter of time until a similar legislative effort succeeds.

4. The Elders partnership on existential threats. Starting with our joint open letter, calling on world leaders to urgently address the ongoing impact and escalating risks of the climate crisis, pandemics, nuclear weapons, and ungoverned AI, and more recently our video series, we were delighted to partner with The Elders to work on combatting these grave threats.

5. WebSummit. In the speech below from WebSummit, which brought over 70,000 participants from around the world to Lisbon, FLI President Max Tegmark advocated for the development of helpful Tool AI instead of risky AGI. FLI’s Futures program director Emilia Javorsky also joined a panel discussing if ethical AI is possible, and presented a talk on pathways to positive futures.

6. Religious engagement. Our initiative to engage with and support religious groups’ perspectives on AI ramped up this year, with numerous events around the world and blog posts from different religious leaders.

7. Campaign to Ban Deepfakes. We’re proud of the diverse, bipartisan coalition of organizations and individuals behind the Campaign to Ban Deepfakes, which - alongside organizations such as Control AI, Encode Justice, the National Organization for Women, and more - we brought together in 2024 to push for liability across the entire deepfake “supply chain”.

8. Increasing state engagement on AWS. Among several other historic milestones this year towards restricting autonomous weapons systems, the first-ever global conference on autonomous weapons systems took place in Vienna, with an incredible 144 states in attendance. Another highlight was the first Economic Community of West African States (ECOWAS) conference on the topic in Freetown, Sierra Leone.

9. New grant opportunities. This year, we launched a number of exciting grant opportunities. From grants for problem-solving AI, to research into how AI may worsen power concentration, and our new multistakeholder engagement RFP, we’re directing millions of dollars to a variety of important projects intended to steer AI towards benefitting humanity.

10. U.S. policy engagement. FLI’s U.S. policy team provided input and recommendations at both the federal and state level this year, publishing numerous reports available to read here. We look forward to continuing our engagement with the incoming administration as well!

Celebrating our 2024 Future of Life Award Winners!

Out now: AI Safety Scorecards!

$5 Million Multistakeholder Engagement RFP

Spotlight on…

Updates from FLI

• We’re reviewing applications for our Head of U.S. Policy role on a rolling basis, but we are accepting applications until January 19. Apply now, and please share!

• FLI’s AI Action Summit lead Ima Bello published our recommendations for the French AI Action Summit, which will take place 10-11 February 2025. The recommendations focus on three core areas: understanding AI risks (science), fostering global cooperation (solutions), and establishing robust international standards (standards).

  • Ima also published the fifth AI Action Summit Newsletter, available here.

• Guest contributor Sarah Hastings-Woodhouse wrote an informative blog post on the FLI site, covering everything you need to know (and why you should care) about AI agents.

• Sarah also wrote a blog post on uncontrollable AI - and whether or not we could shut off a dangerous AI system.

• We’re pleased to see Austrian Foreign Minister Alexander Schallenberg nominated for 2024 Arms Control Person of the Year, for his work raising awareness about autonomous weapons systems (including the Vienna Conference). We encourage you to vote for him here, before January 13.

• Final reminder: Applications for our postdoctoral fellowships on AI existential safety research are still open! Apply by January 6, 2025 at 11:59 pm ET.

• FLI President Max Tegmark joined for an episode of Dr. Brian Keating’s Into the Impossible Podcast, discussing both the potential and dangers of AI:

• Max also spoke to the Financial Times about the urgent need for meaningful legislation against deepfakes, especially given bipartisan agreement on the issue.

• FLI Policy Researcher Alexandra Tsalidis joined for a segment on Al Jazeera’s The Stream, discussing possible threats AI presents to the environment and human life, along with potential solutions:

• On the FLI podcast, Nathan Labenz, host of The Cognitive Revolution Podcast, joined host Gus Docker for a comprehensive look at AI progress from GPT-4 until now.

• Also on the podcast, GiveDirectly President and CEO Nick Allardice and Gus discussed how GiveDirectly uses AI to optimize cash transfers and even to predict natural disasters.

What We’re Reading

• Alignment faking!?: New research from Anthropic and Redwood Research has found the first empirical example of an LLM faking alignment, without any training or instruction to do so. According to the report’s authors, "A model might behave as though its preferences have been changed by the training - but might have been faking alignment all along, with its initial, contradictory preferences 'locked in'.” Creepy.

• Obfuscated activations: A new paper from, among others, FLI’s AI safety research PhD fellows Stephen Casper, Erik Jenner, and Luke Bailey, investigates a weakness in LLM defenses against attacks - specifically, how harmful behaviour can be hidden within an LLM while evading monitors designed to detect such attacks.

• What We’re Watching: Digital Engine has released a new YouTube video, with AI experts sharing their opinions on AI’s existential risks - including in the context of OpenAI’s o3 model: