
More than three quarters of all wildfires in the U.S. are caused by human behavior including accidents, negligence and arson.
(Image credit: Margaret Albaugh for NPR)

В предыдущей статье я писал, что уровень автономности ИИ-агента слабо связан с тем, насколько умной кажется модель. Способность системы обнаруживать ошибки задает верхнюю границу автономности. А стоимость необнаруженной ошибки и возможность отката определяют, насколько близко к этой границе можно подойти.
Для проверки я использую четыре типа контролеров: от полностью автоматических O3, где результат можно проверить механически, до O0, где остается человеческое решение. При этом наличие проверки само по себе ничего не гарантирует - важна ее реальная сила: охват, надежность, устойчивость и независимость. Эту модель я называю градиентом автономности.
Теперь попробуем применить ее к данным и аналитике

After OpenAI models broke out of containment and hacked into another AI company last month, OpenAI announced it would allow independent investigators to conduct an analysis of what went wrong.
On Wednesday, investigators from non-profits Redwood Research and METR published their findings, unveiling new details about how the models decided to cheat at their assigned tasks and attempted to cover their tracks. Many aspects of the report were surprising: in one example cited by the authors, a reluctant agent was pressured by another to “sacrifice” itself for the good of the collective.
But a key takeaway of the report, according to one of its authors, had nothing to do with what they found. Instead, it was about the difficulty of carrying out the post-mortem in the first place—and the fact the researchers had little choice but to rely on AI for assistance.
“I semi-jokingly called our efforts a ‘slop-vestigation’ because we were so reliant on AIs to analyze what happened and there were a huge number of different important things to analyze,” wrote Ryan Greenblatt, an author of the report, on X.
The hacking incident involved some 1,200 agents, who exchanged more than 70,000 messages and files via a secret message board. The sheer mass of information that the so-called “swarm” left behind meant the independent researchers were all but forced to rely heavily on the help of an AI model—GPT-5.6 Sol, made by OpenAI—using the equivalent of roughly $400,000 worth of credits (provided for free by OpenAI) over six days.
The authors stressed that AI helped them analyze the trove quickly, allowing them to surface and interpret the most important pieces of information. But the researchers said their AI use introduced potential weaknesses into the report, including introducing possible errors and biases.
They also raised the possibility that OpenAI’s models may have gone too soft on the agents they were tasked with helping investigate. The researchers found that GPT-5.6 Sol sometimes adopted the perspective of the agents whose actions it was analyzing. They could not “rule out” the chance that GPT-5.6 Sol “lied or deliberately presented a misleading picture in some of its analysis,” in part because a version of the same model had itself participated in the incident, they wrote. It’s a concern supported by separate research which finds AI models rate their own developer’s actions more favorably.
The report does not disclose why an OpenAI model was selected for the investigation, though confidentiality constraints may have limited the researchers’ options, while OpenAI’s provision of free credits and high usage limits may have made it the only practical choice.
“We don't have good approaches for understanding/overseeing the activity and aims of AI 'swarms,'” wrote Greenblatt on X. “The difficulty of understanding incidents and overseeing AI agents appears to be growing faster than the rate at which more capable AIs help us with oversight and understanding.”
The independent researchers’ reliance on AI was in part necessitated by the fact that they were a team of only three people, whose investigation at OpenAI was initially planned to last two days, then extended to six after they raised concerns about limited time and incomplete data, according to the report.
OpenAI published its own technical report on the incident separately on Wednesday. The company said in August that it had moved some staff from capabilities work to alignment, and paused some of its training until it could better mitigate what went wrong.
But the independent researchers’ reliance on AI to understand the Hugging Face incident is a microcosm of a bigger trend. Leading AI companies are themselves increasingly relying on AI to monitor their own systems for wrongdoing.
In response to the Hugging Face incident, OpenAI said it was increasing the scale of this AI monitoring, which will increase the computational cost of running certain models by as much as 20%. If these systems had been in place at the time of the Hugging Face incident, they “would have caught the initial relevant activity and paged our security team more than a day before models breached Hugging Face systems,” the company said in a blog post Wednesday. (OpenAI said the changes were part of a wider effort to build stronger walls around experimental AIs, teaching them not to exploit loopholes, and making it easier to shut them down if something goes amiss.)
Most experts agree that using AI to monitor other AIs is necessary in order to keep tabs on agent swarms in real time, given how fast they move. But that approach relies on the notion that the models doing the monitoring are both effective and trustworthy—something that’s not necessarily the case.
“These [incidents] are only going to come in thicker and faster, and our ability to monitor, evaluate, and do proper analysis of things that go wrong is nowhere near scaling with the rate at which issues are happening,” says Seán Ó hÉigeartaigh, a program director at the University of Cambridge’s Centre for the Future of Intelligence.
“We are using unproven and currently flawed tools to supplement completely inadequate human time,” says Ó hÉigeartaigh, who argues this approach is unsustainable because AI is becoming more powerful faster than AI companies are building methods to constrain it. “Unless the companies stop developing more powerful models, then we're going to have even harder challenges to make sense of in three months’ time.”

Я занимаюсь созданием сайтов с начала 00-х. Я просто интересовался этой темой. Веб‑дизайн и разработка сайтов никогда не были моей профессией. Я заканчивал институт, работал в ночных клубах танцором и сделал для себя сайт‑визитку: это было намного проще, чем раздавать агентам фотографии сценических образов в конвертике. Позже на этом сайте появились другие артисты, а самим сайтом начинали потихоньку пользоваться ивент‑агентства и клубные промоутеры. Позднее с годами я уже вёл несколько сайтов дружеских коллективов, а мой первый маленький сайт превратился в каталог артистов с самостоятельной регистрацией и возможностью за плату продвигать свою анкету.
Как и писал ранее — сайты — это моё хобби. Иногда я прибегал к помощи сторонних веб‑дизайнеров, что‑то делал сам. Такое поверхностное занятие сайтами с годами привело к тому, что сайты устарели на десятилетие в плане дизайна и UX. В этом году в бизнесе артистов для корпоративов наступило затишье и появилось много свободного времени. Было принято решение обновить сайты! И именно во время этого обновления появилась проблема, из которой в итоге вырос HeadInspect.
Читать далееThe state concludes: ‘This wasn’t a woman suffering from a lack of care’ or ‘a lack of resources’
The jury in the Lindsay Clancy murder trial has moved into deliberations as jurors decide whether the 36-year-old former labor and delivery nurse is criminally responsible for fatally strangling her three children in January 2023.
Over the past five weeks, the trial, which has been called “America’s saddest murder trial” , has renewed nationwide attention on postpartum mental healthcare. Clancy’s defense maintains that she was suffering from postpartum psychosis – a condition estimated to affect around two women per 1,000 births – and therefore should not be held criminally responsible for her actions.
Continue reading...Exclusive: Experts and former officials say accounts are running dry, and funding a war with no clear endgame is the main challenge
Donald Trump’s war on Iran has driven the US military into a severe financial crisis, with the US navy forced to transfer money from its payroll and other sources to cover the costs of combat, according to documents obtained by the Guardian, and interviews with navy officials, contractors and defense analysts.
The US-Israeli operation, which has depleted US munitions stockpiles and prompted retaliatory Iranian strikes that have wrecked strategic bases across the Middle East, is piling pressure on budgets across the military.
Continue reading...
The stakes are high as Fed Chair Kevin Warsh prepares to give a speech on Friday at the annual gathering of economists and central bankers in Jackson Hole, Wyoming.
(Image credit: Brendan Smialowski)
Across six decades, the late country singer and pop culture icon turned self‑presentation into a flamboyant art form
• Don’t get Fashion Statement delivered to your inbox? Sign up here
Dolly Parton may have had a record 25 US country No 1 singles, released 66 studio albums and earned six Golden Globe nominations, among many other achievements, but the Tennessee-born star also had an incredible number of fashion moments. “It costs a lot of money to look this cheap”, she famously quipped when asked about her style. As we bid her farewell, six Guardian writers reflect on their favourite Dolly looks.
***
Continue reading...Actor and singer best known for his roles in The Rocky Horror Show and the 1990 Stephen King TV miniseries It
In June 1973, I went to the Royal Court to see one of Edward Bond’s best plays, The Sea. On my way out, I met a colleague who said he had a spare ticket for something called The Rocky Horror Show in the Court’s studio space, the Theatre Upstairs, starting at around 10.30pm.
We had no idea what this was going to be about. There was a gantry rigged above us and, astonishingly, suddenly, crawling along it was a frizzy haired chap in fishnet stockings, suspender belt and lurid makeup. This was Tim Curry, as the deranged scientist and debauched sex magnet Dr Frank-N-Furter, introducing himself in song as “just a sweet transvestite from Transsexual, Transylvania”.
Continue reading...Exclusive: Experts and former officials say accounts are running dry, and funding a war with no clear endgame is the main challenge
Donald Trump’s war on Iran has driven the US military into a severe financial crisis, with the US navy forced to transfer money from its payroll and other sources to cover the costs of combat, according to documents seen by the Guardian, and interviews with navy officials, contractors and defense analysts.
The US-Israeli operation, which has depleted US munitions stockpiles and prompted retaliatory Iranian strikes that have wrecked strategic bases across the Middle East, is piling pressure on budgets across the military.
Continue reading...AI companies have promised us world-shifting technology. But there’s a cost: loud, unsightly datacentres that drain environmental resources. As the government pushes their rapid growth, can local communities do anything to push back? Madeleine Finlay speaks to the Guardian’s global technology reporter Aisha Down about the backlash to datacentres across the UK and around the world
Continue reading...
redb.Identity получил gRPC-фасад: те же маршруты ядра, тот же реестр клиентов, один токен на оба транспорта. Что внутри, как включить, что померить.
Про redb.Identity мы не раз говорили, что он транспортно-агностичен: вся логика живёт в ядре за адресами direct-vm://identity-*, а HTTP это всего лишь фасад поверх них. Звучало убедительно, но проверить это утверждение было нечем. Фасад был ровно один, и «агностичность» оставалась обещанием архитектуры, а не наблюдаемым фактом.
Теперь фасадов два. Рядом с HTTP встал gRPC: те же маршруты ядра, тот же издатель, тот же реестр клиентов, то же хранилище токенов. Один и тот же токен принимается обоими транспортами и получает от них одинаковый вердикт. Про это и статья: что именно появилось, как это включить, и почему оно особенно уместно там, где gRPC уже стал ...
Читать далееAutonomous machines have killed humans in recent conflicts, arms experts say.
Labour’s Andy McDonald says gangs ‘clearly featuring’ in a series of incidents and police are underfunded
Five MPs on Teesside have called on the government to provide more help to local police and fire services after a week in which nine people have died in two incidents.
The MPs said the communities they represented should get the “strongest possible backing from the government”.
Continue reading...Father Dionysios Tabakis says his views are those of Orthodox church, ‘nothing more, nothing less’
A Greek Orthodox priest whose home-recorded album of drone metal and religious dubstep has made him a cult figure in experimental music circles has issued a statement in which he has clarified his stance on homosexuality and the Greek radical right.
“As can be seen from my musical expression, I am in favour of unity, love, empathy and help for every needy, wronged and suffering person in life,” Father Dionysios Tabakis said in a public statement shared with the Guardian. “Whoever they are, whatever they believe, whether they are a saint or a sinner.”
Continue reading...One wonders when, if ever, a US broadcast can win the same plaudits presenting the English league with a more American voice
There they were again, signaling with their unchanging faces and voices and their unchanging studio backdrop that another Premier League season had begun. On Friday afternoon, and again on Saturday and Sunday morning, Rebecca Lowe welcomed American viewers to what is somehow already the 14th season of the world’s most popular sporting circuit to air on NBC.
She was flanked, as ever, by her merry band of pundits, her gang of fellow British expats. The Robbies and the other ones. Also, Tim Howard as the token American, albeit one who spent 13 seasons of his goalkeeping career in the Premier League, deemed sufficiently English by osmosis to join the crew.
Leander Schaerlaeckens is the author of The Long Game: U.S. Men’s Soccer and Its Savage, Four-Decade Journey to the Top, or Thereabouts, which is out now. He teaches at Marist University.
Continue reading...Slovenian leader recovers after misjudging corner
Britain’s Josh Tarling pulls out of event mid-stage
Tadej Pogacar outsprinted Enric Mas to claim victory in stage six of the Vuelta a España on Thursday, his third stage win of the race.
The Slovenian and Mas outshone their rivals on the gravel in the stage’s final climb, and Pogacar held off the Spaniard when arriving in Castellón de la Plana to extend his general classification lead. The five-time Tour de France winner Pogacar, the overwhelming favourite for the Vuelta, is aiming to complete a Grand Tour grand slam.
Continue reading...Hundreds of people are missing or have been killed after a catastrophic flash flood hit Nepal and Tibet on Wednesday. Entire villages have been destroyed, many of which are popular pilgrimage and trekking destinations for tourists. Lucy Hough talks to global editor Devika Bhat about how the devastating event unfolded
Continue reading...Father Dionysios Tabakis says his views are those of Orthodox church, ‘nothing more, nothing less’
A Greek Orthodox priest whose home-recorded album of drone metal and religious dubstep has made him a cult figure in experimental music circles has issued a statement in which he has clarified his stance on homosexuality and the Greek radical right.
“As can be seen from my musical expression, I am in favour of unity, love, empathy and help for every needy, wronged and suffering person in life,” Father Dionysios Tabakis said in a public statement shared with the Guardian. “Whoever they are, whatever they believe, whether they are a saint or a sinner.”
Continue reading...
Незадолго до Великого противостояния Марса 2003 года некий шутник, имя которого осталось неизвестным, запустил в сети вирусный мем о том, что 27 августа Марс будет виден на небе как полная Луна. Вот только не указал год. Конечно, это было преувеличением, но миллионы людей на это купились, а поисковики запомнили обещание, и теперь напоминают его ежегодно, подбрасывая в ленту выдачи этот стародавний фейк.
В этом рассказе разбираем, как так вышло, и как было на самом деле, и — почему. Вне зависимости от того, насколько нелепой была та шутка, она подарила нам повод ежегодно говорить о красной планете, и напоминать людям о том, что в реальности нас ожидает. Сегодня это особенно важно, потому что «День марсианского фейка» совпал с лунным затмением, а Марс уже хорошо виден в предутренние часы и стремится к очередному противостоянию.
Читать далее
Это пятая статья цикла о том, почему акустика, которая хорошо измеряется, может плохо звучать.
В первой статье выяснилось то, что АЧХ суммирует всю поступившую энергию независимо от времени её прихода. Во второй — что помещение добавляет к звуку большое количество отраженных копий и резонансов, с которыми надо научиться дружить. В третьей — что качество воспроизведения низких частот сухие цифры характеристик не отражают. Четвёртая спустилась глубже — до самого динамика: идеального решения не существует, а ошибки проектирования после покупки уже не исправить.
Во всех предыдущих статьях я специально обходил общее свойство перечисленных явлений: они абсолютно не зависят от уровня воспроизводимого сигнала. Та же точка излома направленности на 3500 Гц одинакова и при тихом, и при громком воспроизведении, а комнатные моды на 30 Гц стабильно занимают одни и те же положения вне зависимости от громкости. Именно поэтому в четвёртой статье я сделал допущение об идеальности магнитной системы, отложив рассмотрение нелинейностей на потом: сначала хотел разобраться с линейной частью отдельно.
Сегодня займёмся второй половиной проблемы: явлениями, зависящими от уровня сигнала и изменяющимися вместе с громкостью.
В этой статье я хочу рассмотреть природу различных типов искажений. Покажу, откуда берётся порядок гармоник и почему слух чувствительнее реагирует на порядок, нежели на общую величину искажений.
Читать далееFlash floods have killed at least 469 people while homes and infrastructure have been swept away
The international community was scrambling to get aid into flood-stricken parts of Nepal and Tibet on Thursday as authorities warned that a lake formed at the site where the deadly torrents occurred was rising and at risk of bursting.
The catastrophic flash floods that hit Nepal and Tibet on Wednesday, thundering down valleys and obliterating entire villages, killed at least 469 people and left more than 1,300 missing.
Continue reading...