B R O A D T I M E L I N E S We should have neither short AI timelines, nor long timelines, but a broad probability distribution over when transformative AI will arrive. My new essay explains why & explores the implications of such deep uncertainty. 🧵 1/

The median age in China has rapidly caught up with the United Kingdom— In 1965, the median age in the United Kingdom was almost twice that of China. Half of the people in the UK were younger than 34 years, and half were older. In China, this midpoint was just 18 years.

Being born is a roll of the dice. Most of us got insanely lucky. Imagine you had to roll again, how would you want the world to look? www.givingwhatwecan.org/birth-lottery

Birth Lottery

If you were reborn today, where would you land? And how would that change your life?

givingwhatwecan.org

Dim Red Dot Scientists have just released a photo featuring a dim red dot. It is the light of a single star exploding in a galaxy so far far away that that nothing we do could ever affect it — even in the very fullness of time. It lies beyond the Affectable Universe. Let me explain… 1/🧵

New report on trends in AISI's evaluations of frontier AI models over the past two years. A lot of AI discourse focuses on viral moments, but it is important to zoom out to the less flashy trend: AI models are steadily growing in capabilities, including for dual-use. www.aisi.gov.uk/frontier-ai-...

It has become received wisdom in Brussels and Washington that there is a new “euro-sclerosis”: that the EU economy is lagging the US This view is wrong A little primer on the measurement of productivity – and why reports of the economic death of Europe are greatly exaggerated🧵

Today is Giving Tuesday, and you can 100x the impact of your donations by finding the most effective charities. This year, needs across global health, animal welfare, and catastrophic risk are rising while some major funders step back

New Google DeepMind paper: "Consistency Training Helps Stop Sycophancy and Jailbreaks" by @alexirpan.bsky.social, me, Mark Kurzeja, David Elson, and Rohin Shah. (thread)

Frontier AI could reach or surpass human level within just a few years. This could help solve global issues, but also carries major risks. To move forward safely, we must develop robust technical guardrails and make sure the public has a much stronger say. superintelligence-statement.org

In an op-ed published today in TIME, Charlotte Stix and I discuss the serious risks associated with internal deployment by frontier AI companies. We argue that maintaining transparency and effective public oversight are essential to safely manage the trajectory of AI. time.com/7327327/ai-w...

When it Comes to AI, What We Don't Know Can Hurt Us

Yoshua Bengio and Charlotte Stix explain how companies' internal, often private, AI development is a threat to society.

time.com

The other evening I attended the launch of David Edmonds' book on Peter Singer's Shallow Pond. I was quite struck when he called it 'the most influential thought experiment in the history of moral philosophy' yet with no influence for its first 30 years… 🧵 press.princeton.edu/books/hardco...

Death in a Shallow Pond

From the bestselling coauthor of Wittgenstein’s Poker, a fascinating account of Peter Singer’s controversial “drowning child” thought experiment—and how it changed the way people think about charitabl...

press.princeton.edu

We’re hiring! Society isn’t prepared for a world with superhuman AI. If you want to help, consider applying to one of our research roles: forethought.org/careers/res... Not sure if you’re a good fit? See more in the reply (or just apply — it doesn’t take long)

Evidence Recent AI Gains are Mostly from Inference-Scaling 🧵 Here's a thread about my latest post on AI scaling … 1/14 www.tobyord.com/writing/most...

Evidence that Recent AI Gains are Mostly from Inference-Scaling — Toby Ord

In the last year or two, the most important trend in modern AI came to an end. The scaling-up of computational resources used to train ever-larger AI models through next-token prediction ( pre-trainin...

tobyord.com

"It has gone largely unnoticed that time spent on social media peaked in 2022 and has since gone into steady decline." By @jburnmurdoch.ft.com www.ft.com/content/a072...

✍️ New article: “Foreign aid from the United States saved millions of lives each year” For decades, these aid programs received bipartisan support and made a difference. Cutting them will cost lives.

An insightful piece by Deena Mousa about how AI performs extremely well at benchmarks for reading medical scans, yet isn't putting radiologists out of work. Lots to learn for other knowledge-work professions here. www.worksinprogress.news/p/why-ai-isn...

AI isn't replacing radiologists

Radiology combines digital images, clear benchmarks, and repeatable tasks. But demand for human radiologists is ay an all-time high.

worksinprogress.news

Evaluating the Infinite 🧵 My latest paper tries to solve a longstanding problem afflicting fields such as decision theory, economics, and ethics — the problem of infinities. Let me explain a bit about what causes the problem and how my solution avoids it. 1/N arxiv.org/abs/2509.19389

Evaluating the Infinite

I present a novel mathematical technique for dealing with the infinities arising from divergent sums and integrals. It assigns them fine-grained infinite values from the set of hyperreal numbers in a ...

arxiv.org

Establishing where we collectively draw red lines is essential to prevent unacceptable AI risks. See the statement signed by myself and over 200 prominent figures: red-lines.ai

200+ prominent figures endorse Global Call for AI Red Lines

AI could soon far surpass human capabilities and escalate risks such as engineered pandemics, widespread disinformation, large-scale manipulation of individuals including children...

red-lines.ai

The Extreme Inefficiency of RL for Frontier Models 🧵 The switch from training frontier models by next-token-prediction to reinforcement learning (RL) requires 1,000s to 1,000,000s of times as much compute per bit of information the model gets to learn from… 1/11 www.tobyord.com/writing/inef...

The Extreme Inefficiency of RL for Frontier Models — Toby Ord

The new scaling paradigm for AI reduces the amount of information a model could learn per hour of training by a factor of 1,000 to 1,000,000. I explore what this means and its implications for scaling...

tobyord.com

I'm overjoyed to see that more than 10,000 people have joined me in pledging 10% of their lifetime income to help others as effectively as they can. We're each able to do so much — and so much more together. 🧵 @givingwhatwecan.bsky.social

Let’s take a look into GPT-5’s record-setting performance on FrontierMath. How did it perform on the holdout vs. non-holdout set, how did it do across tiers, and what new Tier 4 problems did it solve? 🧵

Should we expect widespread moral progress in the future? In a new paper, Convergence and Compromise, @FinMoorhouse and I discuss this. Thread.

The Code of Practice is out. I co-wrote the Safety & Security Chapter, which is an implementation tool to help frontier AI companies comply with the EU AI Act in a lean but effective way. I am proud of the result! 1/3

Happy to announce that our paper "Systemic contributions to global catastrophic risk" now is out in Global Sustainability. Short version: systemic risk and global catastrophic risk obviously go together, and we need to link the fields more closely. www.cambridge.org/core/journal...

There are some interesting details about how Anthropic trained their models tucked away in today's summary judgement: they bought, chopped up and scanned millions of dollars worth of books! simonwillison.net/2025/Jun/24/...

New podcast episode with @tobyord.bsky.social — on inference scaling, time horizons for AI agents, lessons from scientific moratoria, and more. pnc.st/s/forecast/...

Fin Moorhouse interviews @tobyord.bsky.social about the future of AI and the risks it involves. Clear and illuminating. pnc.st/s/forecast/5...

Inference Scaling, AI Agents, and Moratoria (with Toby Ord)

Toby Ord is a Senior Researcher at Oxford University. We discuss the ‘scaling paradox’, inference scaling and its implications, ways to interpret trends in the length of tasks AI agents can complete,...

pnc.st

Our new paper updating key metrics in the IPCC is now out, and the news is grim: ⬆️ Human induced warming now at 1.36C ⬆️ Rate of warming now 0.27C / decade ⬆️ Sharp increase in Earth's energy imbalance ⬇️ Remaining 1.5C carbon budget only 130 GtCO2 essd.copernicus.org/...

Indicators of Global Climate Change 2024: annual update of key indicators of the state of the climate system and human influence

Abstract. In a rapidly changing climate, evidence-based decision-making benefits from up-to-date and timely information. Here we compile monitoring datasets (published at https://doi.org/10.5281/zenodo.15639576; Smith et al., 2025a) to produce updated estimates for key indicators of the state of the climate system: net emissions of greenhouse gases and short-lived climate forcers, greenhouse gas concentrations, radiative forcing, the Earth's energy imbalance, surface temperature changes, warming attributed to human activities, the remaining carbon budget, and estimates of global temperature extremes. This year, we additionally include indicators for sea-level rise and land precipitation change. We follow methods as closely as possible to those used in the IPCC Sixth Assessment Report (AR6) Working Group One report. The indicators show that human activities are increasing the Earth's energy imbalance and driving faster sea-level rise compared to the AR6 assessment. For the 2015–2024 decade average, observed warming relative to 1850–1900 was 1.24 [1.11 to 1.35] °C, of which 1.22 [1.0 to 1.5] °C was human-induced. The 2024-observed best estimate of global surface temperature (1.52 °C) is well above the best estimate of human-caused warming (1.36 °C). However, the 2024 observed warming can still be regarded as a typical year, considering the human-induced warming level and the state of internal variability associated with the phase of El Niño and Atlantic variability. Human-induced warming has been increasing at a rate that is unprecedented in the instrumental record, reaching 0.27 [0.2–0.4] °C per decade over 2015–2024. This high rate of warming is caused by a combination of greenhouse gas emissions being at an all-time high of 53.6±5.2 Gt CO2e yr−1 over the last decade (2014–2023), as well as reductions in the strength of aerosol cooling. Despite this, there is evidence that the rate of increase in CO2 emissions over the last decade has slowed compared to the 2000s, and depending on societal choices, a continued series of these annual updates over the critical 2020s decade could track decreases or increases in the rate of the climatic changes presented here.

essd.copernicus.org

I recently participated in the High-Level Expert Panel on AGI, convened by the Council of Presidents of the UN General Assembly, which has just released its final report on the Governance of the Transition to AGI. Report below ⬇️ uncpga.world/agi-uncpga-r...

Big credit to Anthropic for activating ASL3 when their evaluations indicated it was necessary. Increases confidence in their reliability. Looking forward to going through it in more detail: www.anthropic.com/news/activat...

Activating AI Safety Level 3 Protections

We have activated the AI Safety Level 3 (ASL-3) Deployment and Security Standards described in Anthropic’s Responsible Scaling Policy (RSP) in conjunction with launching Claude Opus 4. The ASL-3 Secur...

anthropic.com