Axis of Ordinary

A thorough investigation into the Hugging Face incident: https://openai.com/index/hugging-face-incident-and-the-road-ahead/

They are checking to see if everyone has strapped on their suicide belt.

Let's take a look at how the SMO is going. Everything is proceeding according to plan ✅

Chinese AI labs released two natively multimodal open-weight models today. GLM-5.3 Flash: https://z.ai/blog/glm-5.3-flash Qwen3.8-Flash: https://qwen.ai/blog?id=qwen3.8-flash-next

AI+robots or doom

- OpenAI leaders believe they are at the cusp of AGI. Sam Altman believes OpenAI will have an internal system that will qualify as AGI by the end of 2026. Mark Chen thinks OpenAI is 80% of the way to AGI. - OpenAI already has the automated AI research intern - that's Astra. Jakub Pachocki says that Astra "can implement an experimental idea inside OpenAI’s code base, run the experiment, and return results, or take a paper and perform work that previously occupied a human researcher for a week". - Astra enables "persistent agents", i.e., virtual colleagues toiling for sustained periods on tasks. Thibault Sottiaux says OpenAI is "close" to showing a product built around the concept of persistence. "It’s definitely going to feel like a new thing to people." - Sam Altman: "I expect this will be the first model where the model actually invents new things in a way that matters" Src: https://time.com/article/2026/08/26/openai-sam-altman-interview/

The largest attempt to date to use AI to replicate papers presented at a major technical conference uncovered hundreds of results that the AI agents could not reproduce or found unsupported by evidence, So far, the organizers have confirmed that at least a dozen papers contain real errors that the conference’s human reviewers missed. Read more: https://www.science.org/content/article/researchers-presented-6000-papers-major-meeting-could-ai-reproduce-their-findings

An important heuristic: “Choose as though your decision is linked to similar decisions.” If you reason that your individual voice or vote has a negligible chance of influencing outcomes you care about, others who think similarly will act accordingly. Together, you might have been able to exert significant influence. But, as a group whose reasoning is correlated, you decided to forgo this influence. In other words, when evaluating the rule “don’t take an action when one person's contribution is unlikely to change the outcome,” you should consider what happens when that shared rule is instantiated across the whole correlated group.

Meanwhile, the fuel crisis continues in Russia. Note: The joke is a Russian wordplay on чек (chek). Normally, чек means a receipt. But Russian drivers also commonly use чек as shorthand for the “Check Engine” warning light. So when the driver asks the cashier for his receipt, the cashier replies: “You already have your chek in the car.” The joke is that instead of a receipt, he already has a Check Engine light—because the station put the wrong/poor-quality fuel in his car.

This seems like a pretty significant advancement in in-context learning. The model learned out-of-distribution tasks it was never trained on, like making a pancake, after watching a single video demonstration of the task, with no fine tuning. Highly recommend that you read the full blog post. Incredible progress: https://www.skild.ai/blogs/s1

How Trump treats allies who have fought 8 wars alongside the United States vs. the worst dictator on earth who is an outspoken mortal enemy. https://truthsocial.com/@realDonaldTrump/posts/117151549221525995 https://truthsocial.com/@realDonaldTrump/posts/117156524147615725

Some links: 1. LLMs could control their host machines by exploiting inference engines https://www.lesswrong.com/posts/CjeobBGnhxg8xvden/llms-could-control-their-host-machines-by-exploiting 2. Uncovering a universal offline sandbox escape https://www.primeintellect.ai/blog/universal-offline-sandbox-escape 3. Twenty Years from RSI to Takeoff https://www.lesswrong.com/posts/LP6uCXs6Ea5qSbWpY/twenty-years-from-rsi-to-takeoff-slow-learning-scaling 4. Across Anthropic models, greater capability is almost perfectly correlated with rejecting CDT in favor of EDT/FDT/UDT-like answers https://www.lesswrong.com/posts/5T6GAsvLPFd3epJtd/for-claude-capability-and-dispreferring-cdt-are-the-same 5. Kimi likes causal decision theory more after RL in twin prisoner’s dilemmas https://www.lesswrong.com/posts/hfNBEKaStASAYMLiu/kimi-likes-causal-decision-theory-more-after-rl-in-twin-1 6. Introducing Index: Building The World’s Largest and Most Diverse Physical Dataset https://www.figure.ai/news/introducing-index 7. Prime Agent: A Self-Improving RLM Harness https://arxiv.org/abs/2608.23552 8. AutoSaddler: Automatic Harness Optimization with Durable Updates from Agent Execution Traces https://arxiv.org/abs/2608.23041 9. Accelerated Understanding Inc is launching today. Their model uses a non-transformer architecture called neural operators https://acceleratedunderstanding.com/ 10. Training AI to Paint with Code https://surya.website/rling-qwen-to-paint-with-code 11. An unusually candid defense of behavioral genetics by a leading expert Peter Visscher. We readily accept that evolutionary, population and quantitative genetics apply to complex traits across species—until the trait happens to be human behavior or intelligence. Then political and moral concerns too often intrude into what should first be an empirical question. https://cambridge.org/core/journals/twin-research-and-human-genetics/article/in-defence-of-behavioral-genetics/454DBC9391AB91D33D53E91C81ED1FF8

Anjney Midha is a genuinely well-connected and unusually well-placed person in frontier AI. He previously led major AI investing at Andreessen Horowitz and currently sits or has sat on the boards of companies including Mistral AI, Black Forest Labs, Luma AI, Sesame, LMArena and OpenRouter. He was also an early investor/adviser to Anthropic; The Information reports that he advised Anthropic's founding team on compute and fundraising in 2021 and invested personally very early. https://x.com/AnjneyMidha/status/2092250775258812875

Stats from OpenAI's first custom inference chip: https://openai.com/index/jalapeno-first-results/

When voters learn that modern data centers recycle water and reuse it for up to a decade, support swings by a net +31 points, the strongest result in three waves of polling. From: https://cdn.prod.website-files.com/663a60b91deeae06e8ae1d1c/6a317d2d221255ef40039b0a_RaineyCenter_DataCenters_May2026_Memo_8.5x11_r1_digital.pdf

Yakutsk, the capital of resource-rich region, is literally paralyzed due to raging gasoline crisis. Drivers have to wait in lines for 12 hours. Waiting time has become so long that the city admin had to organize the sale of water and food right in the queues. Official Russian source confirming this: https://yakutia.info/amp/article/215585

The fuel, energy and food situations continue to deteriorate in Russian occupied Crimea, with grocery stores now unable to stock refrigerated or frozen goods.

Literally Russians and their useful idiots after anyone calls their war retarded: Muhahaha, but we're killing people in Odessa!!! We're destroying, destroying, destroying! Also Russians: Why do you hate us and call us subhumans?

The Russian Ministry of Agriculture is considering lifting the export duties on wheat, barley and corn until the end of the year as the Black Sea remains severely disrupted for exports. Wheat prices are now 3 000 - 4 000 rubles below the cost to produce for farmers of 12 000 - 13 500 rubles The situation on the Russian grain market is INCREDIBLY BAD if Russian farmers in Krasnodar are offering wheat at their farms for close to 100 USD per ton when the price for Romanian wheat is 268 USD per ton and the break even price is 142-160 USD per ton Source: https://www.zol.ru/n/4209c (ZOL.ru is a legitimate Russian specialist trade-news site, operating since 2000)