Top all-time · #2725
Toby Ord
Personal blog · tobyord.com/writingphilosophyai
Toby Ord's personal blog at tobyord.com covers philosophy and ai. Since 2021, 28 different Hacker News readers have linked it 40 times, most often “How well does RL scale?”. It is still writing: the last post was on 2026-10-09, and you can follow it by feed.
Too few sampled comments (12) to say how often HN recommends it.?
Why HN links it
HN readers cite Toby Ord's posts on AI economics when discussing AI progress: the hourly cost of AI agents, the inefficiency of reinforcement learning compared with pretraining, and how RL and inference scale.
Summarised by Claude from a sample of 12 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Most-linked posts?
- How well does RL scale?5 people · 5 mentions · first linked 2025-10-30
- Earth Restored – 50 restored images of earth released4 people · 4 mentions · first linked 2021-04-22
- The Extreme Inefficiency of RL for Frontier Models4 people · 4 mentions · first linked 2025-09-23
- Is there a half-life for the success rates of AI agents?4 people · 4 mentions · first linked 2025-05-11
- Are the Costs of AI Agents Also Rising Exponentially? (2025)3 people · 5 mentions · first linked 2026-04-15
- Swarm Scaling3 people · 3 mentions · first linked 2026-09-21
- The Scaling Paradox2 people · 2 mentions · first linked 2025-01-15
- A Child's Plaything2 people · 2 mentions · first linked 2023-05-06
- Inference Scaling Reshapes AI Governance2 people · 2 mentions · first linked 2025-10-24
What Hacker News says?
In general I have about zero enthusiasm for trying to find defensible interpretations of things that Ed Zitron said, and I generally agree that the name of Zitron just largely needs to stop coming up in anti- and anti-anti-AI arguments since, it seems, he's just not a particularly insightful or reliable voice on the subject. That said, one or two of the specific assessments in Luu's article seem dubious as well, esp…
Comment linking /writing/mostly-inference-scaling
on “How accurate have Ed Zitron's AI skeptic predictions been?”Submitted Earth Restored – 50 restored images of earth released as a story · 409 points
HN discussion> just as humans shouldn’t serve as calculators But they definitely could and were [0]. You just employ multiple, and cross check - with the ability of every single one to also double check and correct errors. LLMs cannot double check, and multiples won't really help (I suspect ultimately for the same reason - exponential multiplication of errors [1]) [0] https://en.wikipedia.org/wiki/Computer_(occupation) [1] https…
Comment linking /writing/half-life
on “Seven replies to the viral Apple reasoning paper and why they fall short”Submitted Is there a half-life for the success rates of AI agents? as a story · 249 points
HN discussionThe premise of this post and the one cited near the start ( https://www.tobyord.com/writing/inefficiency-of-reinforcemen... ) is that RL involves just 1 bit of learning for a rollout, rewarding success/failure. However, the way I'm seeing this is that a RL rollout may involve, say, 100 small decisions out of a pool of 1,000 possible decisions. Each training step, will slightly upregulate/downregulate a given trainin…
Comment linking /writing/inefficiency-of-reinforcement-learning
on “The inefficiency of RL, and implications for RLVR progress”Submitted AI Scaling vs. Human Scaling as a story · 6 points
HN discussionThis is the fundamental question and don’t you find it interesting that there isn’t a nice clean dashboard on the openAI website where we can go and see this metric progress over the release history? Toby Ord did what he could with public data and it… doesn’t look great. https://www.tobyord.com/writing/hourly-costs-for-ai-agents
Comment linking /writing/hourly-costs-for-ai-agents
on “Separating signal from noise in coding evaluations”Submitted Is There a Half-Life for the Success Rates of AI Agents? as a story · 5 points
HN discussionFunny how nearly all model improvements this year are demonstrated on the subset of use cases where brute force / reinforcement learning is most effective: Robotics (using physics sims) Cybersecurity (red team / blue team) Math (using automated proof checkers) Programming (using compilers) For the record I think robotics is a totally logical place to use this training approach and this is very impressive. But if we…
Comment linking /writing/inefficiency-of-reinforcement-learning
on “Mistral's Robostral Navigate: a state of the art robotics navigation model”Submitted Swarm Scaling as a story · 4 points
HN discussionI'm not sure I agree with you. The only widely cited quantitative guess at this scenario I could find in a brief trawl puts the chance of any "climate-driven existential catastrophe" this century at about 0.1 percent [1]. "Extinction" is a stricter outcome than “existential catastrophe", so I would imagine the odds this century are even lower than 0.1 percent. In other words: don't resign yourself to the extinction…
Comment linking /writing/the-precipice-revisited
on “Ask HN: How Is Everyone Feeling?”Submitted The Scaling Paradox as a story · 4 points
HN discussion> So "inference scaling is required to scale capabilities" doesn't mean that we're reaching the top of the S-curve in intelligence. On its own it wouldn't. But that article came before the later article https://www.tobyord.com/writing/hourly-costs-for-ai-agents which adds the claim that inference (along with everything else being employed at present) is scaling poorly with increasing task lengths. Now maybe the Dece…
Comment linking /writing/hourly-costs-for-ai-agents
on “AI 2040: Plan A”Submitted Swarm Scaling as a story · 2 points
HN discussionYes. Anyone who doesn't acknowledge the efficiency difference between pretraining vs RL and assume that since we've run out of data for the former, we have to do the latter, is not making a serious attempt at modelling the future: https://www.tobyord.com/writing/inefficiency-of-reinforcemen... This is similar to that other exponential, which happened with CPUs - we ran out of true geometric scaling in the mid 2000s,…
Comment linking /writing/inefficiency-of-reinforcement-learning
on “AI 2040: Plan A”Submitted How well does RL scale? as a story · 2 points
HN discussionThe author of the posts you linked also wrote https://www.tobyord.com/writing/inference-scaling-reshapes-a... which posits: > AI labs may also be able to reap tremendous benefit from these inference-scaled models by using them as part of the training process. If so, the large scale-up of compute resources could go into post-training rather than deployment. This would have very different implications for AI governanc…
Comment linking /writing/inference-scaling-reshapes-ai-governance
on “AI 2040: Plan A”Submitted How Well Does RL Scale? as a story · 2 points
HN discussionThings like https://www.tobyord.com/writing/hourly-costs-for-ai-agents and https://www.tobyord.com/writing/mostly-inference-scaling seem in line with other accounts like https://www.youtube.com/watch?v=aR20FWCCjAs ?
Comment linking /writing/mostly-inference-scaling
on “AI 2040: Plan A”Submitted The Extreme Inefficiency of RL for Frontier Models as a story · 2 points
HN discussionI meant reference Toby Ord's work here. I think his framing of the performance/cost frontier hasn't gotten enough attention https://www.tobyord.com/writing/hourly-costs-for-ai-agents
Comment linking /writing/hourly-costs-for-ai-agents
on “Claude Opus 4.7 costs 20–30% more per session”Submitted Is There a Half-Life for the Success Rates of AI Agents? – Toby Ord as a story · 2 points
HN discussionHmm. What’s the general belief about Toby Ord’s “Are the Costs of AI Agents Also Rising Exponentially?” https://www.tobyord.com/writing/hourly-costs-for-ai-agents among those who are well-equipped to judge? Is it seen as wrong or disproven or unlikely? Because if not—if indeed recent LLM capability advances have likely relied on increases in inference cost per run which can’t be much further sustained—then it seems…
Comment linking /writing/hourly-costs-for-ai-agents
on “The Sigmoids Won't Save You”Submitted Earth Restored: newly retouched photographs from the Apollo lunar missions as a story · 2 points
HN discussionScaling law is a power law , requiring orders of magnitude more compute and data for better accuracy from pre-training. Most companies have maxed it out. For RL, we are arriving at a similar point https://www.tobyord.com/writing/how-well-does-rl-scale Next stop is inference scaling with longer context window and longer reasoning. But instead of it being a one-off training cost, it becomes a running cost. In essence…
Comment linking /writing/how-well-does-rl-scale
on “Epoch confirms GPT5.4 Pro solved a Frontier Math Open Problem for the first time”Submitted A Child's Plaything as a story · 2 points
HN discussionSubmitted Earth Restored as a story · 2 points
HN discussionSubmitted Effective Altruism as a story · 2 points
HN discussionSubmitted Earth Restored: Photographs from Apollo as a story · 2 points
HN discussionSubmitted Swarm Scaling as a story · 1 points
HN discussion
