Twice monthly updates on how to survive the transition to AI

I founded 80,000 Hours, a non-profit that researches the world's most pressing problems and helps people find careers tackling them, and am the author of the 80,000 Hours book (Penguin ‘26).

Over a decade ago we started telling readers that powerful AI could be the defining issue of our time. Hundreds went on to help start the field of AI risk, working at organisations like METR, the UK’s AI Security Institute, and the safety teams at Anthropic and DeepMind.

Since then, progress has been even faster than I expected. I started this Substack to try to understand what’s happening and what to do about it.

Despite all the noise about AI, I think it’s still under-hyped. Many continue to assume LLMs will soon plateau due to foibles they found a year ago. Key developments are still overshadowed in the media by random political news. Economists assume GDP will continue to grow at 2-3%.

But AI might not be a ‘normal’ technology like the internet or even electricity. AI is already being used to help with its own development, and as it gets better, the extremely fast progress we’re seeing could accelerate even faster.

AI could develop into autonomous agents that can do almost every cognitive task better than humans as soon as the end of 2027. Leading companies have enough computing power to run billions of copies of them. This would be more like a massive expansion in the labour pool than a new technology, or even the arrival of a second species.

Billions of AI scientists and robotic workers could lead to a century of scientific and economic progress in a decade, making this moment akin to the invention of agriculture, and one of the most important in history.

None of this is certain. It will become harder to continue the current scale-up of computing power after 2028, so if transformative systems are not reached by then, progress could plateau for decades, or AI R&D could get banned. But the chance seems high enough to take extremely seriously.

The outcome of AI-driven acceleration could be extraordinary, or genuinely existential, posing risks like loss of control of autonomous systems, concentration of power, AI-enabled pandemics and gradual disempowerment. Either way, it’s the most important thing to understand in the world right now.

Here I try to distill the hundreds of blogs and tweets about AI into simple explainers of what’s most crucial to know, how to help the transition go well and how to prepare personally.

To start, here are my two key posts on what’s happening:

Here are my key posts on what to do about it:

Here are some of my other most popular posts:

Read more about me here.

It’s free to subscribe, and you’ll receive about two posts a month.

You can also follow me:

User's avatar

Subscribe to Benjamin Todd

How to survive the AI transition

People

x