Save the animals. Nuke the data centers.
---
P(doom) = 98.7% (Oct-2026) P(doom) = 99% (Sept-2026) P(doom) = 98.9% (Aug-2026) P(doom) = 98.2% (July-2026) P(doom) = 98.2% (Jun-2026) P(doom) = 98.5% (May-2026) P(doom) = 98.8% (mid-April-2026) P(doom) = 98.7% (April-2026) P(doom) = 98.7% (March-2026) P(doom) = 98.5% (mid-Feb-2026) P(doom) = 97% (Feb-2026) P(doom) = 94% (Jan-2026) P(doom) = 93% (Dec-2025) P(doom) = 95% (July-2025)
Where P(doom) equals the risk of global civilisation collapse or worse as a result of AI by 2050, assuming no other doom events occur within that period.
---
Oct-2026 Update:
Two months on and it's now undeniable that AI's are superior to the average PHD-level mathematician, and arguably better than any mathematician currently alive.
Additionally, it seems quite probable that some discoveries being made are now being held from the public. Some of these discoveries are likely to include ones which could undermine some cryptographic security methods which underpin the modern world.
Things will happen in the world from here that we cannot understand because the power of new and potentially dangerous discoveries will be kept in the hands of a few. Soon will come a day where you wake up and it will be as if the first nuclear bomb was dropped, but these days will come often.
I fear it's now too late to reverse where we're heading. We simply have to hope for the low probability good ending to it all. Please make the most of your time.
---
Aug-2026 Update:
AI's are currently nearing PHD-level in math, and frontier labs may already be there with unreleased models. Given this, RSI now seems imminent. Meanwhile, we haven't solved the most basic safety or alignment problems, and we're seeing increasingly significant real-world AI-related security incidents.
I'd now put the probability of ASI* by year end at 20%. By end of next year I'd put this at 50%+. Really the only question worth asking at this point is what happens next? If we cannot make current AIs do what we want, then what happens when they're super intelligent?
Spend time with your family while you still can. These are the last months of normality.
* ASI as defined as AI which is superhuman across almost all intellectual tasks (99+%)
---
My opinions become more reasonable with time.