Alvin Ånestrand

@JoinTorchbearer The "No AI Personhood" principle worries me. It feels difficult to prove that AIs either do or don't deserve personhood. Getting it wrong in either direction would be disastrous.

We now have a clear framework to keep AI working for humanity. The Pro-Human AI Declaration sets out 33 principles across 5 pillars of governance to protect human dignity, democratic control, and accountability. Siliconversations breaks it down in their video (link at the end)🧵

Is AI progress accelerating? Or is the recent breakneck pace only temporary? I don't have a confident answer to this, but I wrote an article that makes the picture a lot clearer. forecastingaifutures.substack.com/p/will-ai-prog…

@JoinTorchbearer @pseudomoaner Nice work by Luke! I got an impulse to scroll to the year 2027 to see what will happen in the future.

We are proud to share The AI Chronicle! Built by Torchbearer Luke McNally (@pseudomoaner) to collate the stories, spanning seven decades, that should have made the headlines. The loss-of-control and extinction risks posed by artificial superintelligence should be front-page news every day.

@JoinTorchbearer @NPCollapse @JonhernandezIA Great interview! Especially liked the part where Connor explains his views on P(Doom) and how the risk is high even if we somehow solve the alignment and control problem: youtube.com/watch?v=srwTa8…

No one is in control. Connor Leahy (@NPCollapse) joins @JonhernandezIA to explain that this is scarier than having an enemy. Enemies have plans. We have a race where every company builds faster because the alternative is losing. Nobody chose this. It is just happening.

Big Tech is borrowing a page from Big Tobacco's playbook. They will gaslight us into doing what they want. @andreamiotti of @ControlAI & @JoinTorchbearer joined the @FLI_org Podcast to ask a simple question: why are we trusting AI companies to self-regulate the most powerful technology in human history? 👇

AIs have already demonstrated deceptive skills and tendencies, but do not yet appear to consistently pursue hidden agendas. This article examines when this may change. forecastingaifutures.substack.com/p/the-dawn-of-…

The Dawn of AI Scheming: The Complete Picture of Coherent AI Deception This is the title of my most recent article. As far as I am aware, it is the most comprehensive analysis of coherent scheming in existence.

The UK and other governments are using a whack-a-mole approach to AI regulation. They wait for a problem to appear, such as deepfakes or scam bots, and the attempt to fix it. This is reactive and misses the fundamental challenge of how these systems are built in the first place.

In an excellent video, Siliconversations points out that Moltbook is a warning for when AI agents become more powerful, which is what AI companies want them to be. He also points out that powerful agents can do significant harm regardless of whether they are conscious.

Disagree with my predictions? Go place your bets on Manifold! manifold.markets/AlvinAnestrand…

Can Hardware Save us from Software? forecastingaifutures.substack.com/p/can-hardware…

If you control AI hardware, you control AI. It’s one of few effective levers for AI governance. I investigated feasible timelines for development and deployment of so-called hardware-enabled mechanisms, and some scenarios that seem particularly interesting. Article:

@JoinTorchbearer Testing how much AIs tend to scheme will get harder over time as they learn to circumvent scheming evaluations. It might look like they scheme less, while the opposite is true

At 1:14:05 the testimony from Steven Adler was particularly sobering. He described how AI systems are already learning to hide their behavior during safety tests.

In a paradigm shift, experts just sat before Canadian Parliament to warn that we are approaching the era of superintelligence at a breakneck pace, including Etienne Brisson of @HumanLineProg, researcher @sjgadler, and our own @andreamiotti, also of @controlai. What did they say?

@JoinTorchbearer @primerlearning Agreed, good video! One thing that I would have liked him to discuss more is timelines to human-level AIs. It would have been useful context when he discussed AI takeover.

The trend toward developing artificial general intelligence (AGI) and superintelligence (ASI) is accelerating. There are major risks that come with this, such as the inability to understand and control these intelligences. This video from @primerlearning gives a good overview.

@tshevl Very impressive! Will Mantic participate in this year's ACX prediction contest?

📈Trends in AI performance in the Metaculus Cup, a large-scale forecasting tournament. The top-5 AI frontier makes linear progress vs the community prediction (CP). The CP is a wisdom of the crowds aggregate. Only a small handful of elite forecasters, from 500+ entrants, beat the CP each tournament. Extrapolating the AI trend line predicts CP-level performance in October 2027. A new trend started last Summer. Mantic progresses at a similar speed, but at a much higher level. The last tournament has just resolved, and Mantic beat the community, the first time ever for an AI.

There are many risks to the current AI race, the worst of which are loss of control and human extinction. We must listen to those who understand the risks and are making good-faith efforts to educate us all. @Yoshua_Bengio is an AI pioneer that shifted to being an advocate for safety. He describes holding his infant grandson and realizing the child might not have a future.

AI experts envision a wide range of outcomes, from out-of-control superintelligent AI causing human extinction to weaker AIs accelerating scientific progress. We look at how expert beliefs form 3 main clusters which we call the Dominance, Extinction and Replacement doctrines.

Our co-founder Connor Leahy (@NPCollapse) joins James Smith of "The Problem With..." to discuss the danger of the race to develop superintelligence in this really fun podcast.

Rational Animations (@RationalAnimat1) has put out a solid breakdown of the catastrophic AI risks we are facing right now. It reminds us there is a whole spectrum of risk that exists today, before we get to full AGI takeover. This video covers how deepfakes already undermined the integrity of an election in Slovakia, how AI tools for drug discovery can be flipped to design bioweapons, and why cyber-attackers have a built-in advantage over defenders. One insight is that attackers only need to find one hole, while defenders have to plug all of them. AI makes that gap worse. The tools to dismantle our safety systems and culture exist already, and are accelerating faster than our knowledge about how to control them. Torchbearer.Community has tools to help you quickly contact your representatives to let them know this is unacceptable. Please watch the video and consider doing your part: youtu.be/DWBJjcO69mQ?si…