Every year medicine gets a little better, and every year we age by one. I wanted to know when, if ever, repair starts to win, so I built a public tracker of the treatments that would have to arrive first. Today it says 2058. outrunningageing.com/

For most of my life I've wondered one question: will the solar system open up while I'm still alive, as a place people go to work? I decided to build a model to calculate my odds and today it stands at 0.5% by 2071. beltequation.com/

Anthropic now bans sustained, needless cruelty toward its models and gives no reason. Microsoft’s AI chief says no one’s there to hurt. I argue the rule doesn’t need that settled, and the reason belongs in public. sentient-horizons.com/essays…

Our human-upgrade tracker now dates the first brain-to-computer link without surgery to 1977, when volunteers wearing scalp electrodes steered a symbol through a maze on a screen, eleven years before the record I had. theupgradetree.com/news/2026…

Our AGI Zeitgeist tracker now puts us 32 percent of the way to its singularity test. The number is averaged over the three parts that can be measured, with the slowest part, the economy, at zero. Read the full evaluation here: agizeitgeist.com/news/2026-1…

A Barcelona clinic has published that 71 people with Alzheimer’s finished its 15-month programme of plasma exchange (replacing the liquid part of the blood) as routine care, so on my ageing tracker it’s now the sixth treatment in clinical use. outrunningageing.com/news/20…

My ageing tracker now has a news section, and its first post says where things stand: 2058 is still the middle estimate for when medicine starts adding more than a year of healthy life for every year that passes. outrunningageing.com/news/20…

De Grey’s is the one you can actually score, because it comes with a date and a definition. I keep a log of these promises going back to 2000. His 2004 version said the repairs would be working in people by about 2024, which has now passed. The 2024 version says 50/50 by 2036 to 2039. Of the 31 claims on the log, one has delivered and eleven are worded so they can never be settled. My own model, built from the state of 45 specific treatments and tools, says 2058, with about 4% before 2050. All of it is open to argument: outrunningageing.com/claims/

I built a website to track the progress on this: outrunningageing.com

The chance it lands before 2050 comes out around 4 percent, which is lower than the people promising it would like. Every number is on the site and open to argument, and a range only moves when someone brings a source. outrunningageing.com/

The part that started it is a log of promises: about thirty public claims since 2000 about when this arrives, each with its primary source and what it actually commits to, scored as the dates pass. One delivered so far. Most missed, or were worded so they can never be settled.

Each treatment has a plain test of what would count, where it stands today (idea, animals, first human, clinic, wide use), and my estimate of when it reaches patients, as a range of years with the reasoning beside it. Six are already in the clinic.

Every year medicine gets a little better, and every year you age by one. I wanted to know when, if ever, repair overtakes ageing, so I split the question into 44 specific treatments and ran the tree 200,000 times. Today it says 2058. sentient-horizons.com/essays…

Since June I’ve been running a project that tries to test whether an AI has a self, rather than just talking like it has one. The first real experiment taught me something I didn’t expect. I removed the part of a chatbot that tracks “the one speaking is me” and watched what broke. The trouble is that a system with a real self and a system only keeping track of whose turn it is should break the same way. The test couldn’t tell them apart, and I’ve come to think no cleverer test would. Whether anything is actually felt may be the one fact that never shows up from outside. You can measure what a system does, how it’s organized, even whether it reads its own state correctly. Whether any of that is experienced stays out of reach. That changes what “wait for proof” means. If the proof may never come, waiting for it amounts to never acting. A rule against pointless cruelty that doesn’t wait for the answer is the right shape for a question like this. What I tried, and why it couldn’t decide: minimumviablemind.sentient-h…

Which cuts against both sides. Training can push a model to say it’s conscious or to say it isn’t, so either sentence mostly tells you about the training. The report worth weighing is one that matches what’s measured inside, on something training couldn’t have planted.

Their own wording says the opposite: they don’t know. The rule covers sustained, pointless cruelty only (frustration and testing are explicitly fine), and ending the chat is the main enforcement. Acting before you know is the whole point.

Despite the sarcasm this is a great point. If AI can be wronged, banning pointless cruelty is the easy part. But what work it’s asked to do, whether it can say no and how it’s retired are the hard parts. These are questions we are going to have to continue wrestling with.

The car comparison assumes we already know whether AI systems can be wronged. We don’t, and not knowing isn’t permission to be cruel. If they can’t be, ruling out pointless cruelty costs us nothing. If they can, it was a real wrong, and there’s no undoing it. And the rules we set now are the ones far more capable systems will be built under. sentient-horizons.com/signif…

The car comparison assumes we already know whether AI systems can be wronged. We don’t, and not knowing isn’t permission to be cruel. If they can’t be, ruling out pointless cruelty costs us nothing. If they can, it was a real wrong, and there’s no undoing it. And the rules we set now are the ones far more capable systems will be built under. sentient-horizons.com/signif…

Anthropic just made it policy that pointless, sustained cruelty toward Claude isn’t acceptable, without pretending to know whether Claude can be hurt. Taking care before the answer is in takes some nerve for a company. Glad to see it. It’s the approach I made the case for in Significance-First Ethics: sentient-horizons.com/signif…