Когда человек спрашивает у ассистента «к кому обратиться», он получает не десять синих ссылок, а один ответ с пятью-десятью доменами внутри. Я прогнал два запроса через семь ИИ-поисковиков в один заход и сравнил, кого именно они цитируют. Общего источника не нашлось ни одного, а три модели разных разработчиков выдали дословно совпадающие списки — потому что веб-поиск им делает один и тот же плагин. Ниже — стенд, цифры и пять мест, где методика такого замера ломается.
Читать далееDemonstrators accuse authorities of failing to tackle irregularities in exams for sought-after civil service jobs
Thousands of angry young job-seekers clashed with police in Jharkhand, one of India’s poorest states, on Monday as a weeks-long protest over alleged irregularities in recruitment drives for coveted government jobs escalated into a march on the state legislature.
Police fired water cannon and teargas and swung batons as they tried to stop flag-waving demonstrators from clambering over barricades and surging towards the assembly in Ranchi, the state capital. Some of the protesters responded by dancing and cheering as jets of water drenched the crowd.
Continue reading... Women’s 100m and men’s 5,000m top opening billing
Britain ends 92-year wait | Mail Billy | Follow on TikTok
Women’s 100 hurdles: Germany’s Rosina Schneider draws level with Norway’s Lovise Skarbøvik Andresen after 60m of the third heat but pulls away in the final 20m to win. The Norwegian was still on for a fast time but fell flat on her face at the final hurdle. She leaves the scene in tears. Four athletes went faster than GB’s Emma Nwofor there, which means she finishes 13th overall and misses out on the top 12 places for the semi-finals.
Women’s 100m hurdles: Martine Kolbeinshavn Hjørnevik of Norway bursts through the middle in lane 5 to win the second heat ahead of Belgium’s Yanla Ndjip-Nyemeck in lane 2. Ester Bendova of Czechia lost her stride after hitting a hurdle early on and does not finish. Nwofor is down to 9th on the overall times with one heat to go.
Continue reading...
If you’ve ever suspected your dog can read your mind—sussing out through some canine intuition when you’re happy, sad, scared, or cross—you’re half right. Many dogs can do exactly that, but it’s not mind-reading that’s involved: it’s face-reading. That’s the word from a new study in the journal iScience, in which researchers not only established dogs’ ability to discern human moods by examining our facial expressions, but also pinpointed the spots in the brain where that processing takes place. They did this by scanning canine brains with functional magnetic resonance imaging (fMRI).
“This ability to perceive emotions in humans is really hard to find in other species,” says Rául Hernández Pérez, professor of psychology at the University of Vienna and first author of the study. “Chimpanzees have a harder time than dogs. Wolves can also do it, but not at the same level as dogs can.”
Getting a dog to tolerate an MRI scanner is no small matter. The same things that so often freak people out about the MRI experience—the confined space and the infernal noise—can have a similar effect on a dog. That was an impediment to the kind of study the researchers planned to conduct. In order for a brain scan of the proper fidelity to be produced, the dogs must spend a minimum of six minutes in the machine, moving less than three millimeters in any direction. “If they move more than three millimeters, we can’t use the data,” says Laura Cuaya, professor of biological psychology at the University of Vienna and a co-author of the study.
Looking for dogs who could perform that feat, she and Hernández Pérez recruited 14 cooperative canines, consisting of 11 border collies—a highly trainable breed—two golden retrievers, and one Labrador, the last three of which were already working as service animals.
The animals then went through a series of at least 20 training sessions—the first involving just introducing them to the researchers and building a sense of trust. Next, the dogs were taught to lie still for just five seconds in the position they would have to maintain in the MRI scanner. “We reward them with treats, though some dogs prefer to play as a reward,” says Cuaya.
Over the course of several days, that brief five seconds was steadily extended to 10 full minutes. The dogs were then introduced to the headphones they would have to wear while in the machine to protect their hearing. “The headphones are not uncomfortable, but they’re weird,” says Cuaya. Finally, the dogs practiced holding their pose in a mock MRI scanner, and only then were they ready for the real, clanking, buzzing, banging machine.
In the study, the dogs were positioned in the machine and shown pictures of a human face wearing either a happy or neutral expression. All the while the fMRI churned on, peering through the dogs’ skulls into the brain below and looking for the areas that lit up as a result of the visual stimuli. When the dogs saw pictures of happy people, their temporal cortex and caudate nucleus became active. In humans, the temporal lobe plays a role in forming memories, mastering language, responding to visual cues such as familiar faces, and processing emotions. The caudate nucleus processes rewards and motivation and manages learning and memory. The neutral face did not elicit the same response in either region.
Next, the dogs were once again shown images of happy and neutral faces, but along with those they were also shown angry, frightened, and sad ones. The researchers were trying to determine whether the dogs processed human faces in a binary way—registering the positive, happy face in one part of the brain and the three negative ones together in one other part, or if each negative emotion had its own discrete brain region.
“This would mean that the dog brain is not only categorizing emotions between positive and negative," says Cuaya. “We were using this [stage of the experiment] as a demonstration that there is more detailed processing.”
That was the result they got. Sadness and fear, for example, lit up the right rostral supra sylvian gyrus—which, in humans, contributes to the processing of visual information. Anger and fear were processed in the right ectosylvian gyrus and the left splenal gyrus—the former integrating visual, auditory, and other sensory information, and the latter managing visual perception, spatial orientation, and episodic memory.
“We found that the dog brain is not only categorizing emotions as positive or negative, but instead can exhibit more detailed processing, responding in different ways to emotions,” says Hernández Pérez.
This isn’t the first study to find that dogs respond in unique ways to human faces. Other research has shown that when a human face is happy, for example, dogs look more at the eyes than the mouth. They examine emotional faces longer than they examine neutral ones. And they have more difficulty reading the faces of people who are of a different gender from their principal caregiver.
There are benefits to the study—beyond the simple matter of adding to the body of knowledge on how dog brains operate. Hernández Pérez and Cuaya believe their work can also help improve the already deep ties dogs and humans share.
“Humans are not so good at perceiving the emotions of our dogs,” says Cuaya. “We are confused by them. Maybe this work can be an invitation to try to be more aware of the emotions of our dogs and how they are expressing [them].”
That understanding can also pay dividends when we are training our dogs. “When you are teaching something to your dog, you can have treats on hand, and you can give the treats when they are appropriate. But you can also show your emotion. You can say to the dog, ‘Very good!’ The dogs in our MRI machine were already reacting to the faces of strangers. Think about how much more activation there is with a familiar face expressing, ‘I love you so much. I am happy with you.’”
The dog-human love story—which has been going on for some 14,000 years now—has been a long and tender one. The more we understand the canine contribution to that bond, the more we can improve our own.
The former US president will host six episodes of A Great Book for Audible, discussing work from Toni Morrison and John le Carré to James Baldwin
Barack Obama is set to host a podcast about books that have “profoundly shaped his worldview and our society”, from Song of Solomon by Toni Morrison to Tinker Tailor Soldier Spy by John le Carré.
The podcast, called A Great Book with Barack Obama, will launch on Audible on 24 September.
Continue reading...The former US president will host six episodes of A Great Book for Audible, discussing work from Toni Morrison and John le Carré to James Baldwin
Barack Obama is set to host a podcast about books that have “profoundly shaped his worldview and our society”, from Song of Solomon by Toni Morrison to Tinker Tailor Soldier Spy by John le Carré.
The podcast, called A Great Book with Barack Obama, will launch on Audible on 24 September.
Continue reading...
Я попробую на нашем языке рассказать о процессе стройки, и к этому я пришел после осознания следующего факт:
Какой бы чистый код мы ни писали и какие бы отказоустойчивые архитектуры ни проектировали в рабочих тасках, с багами физического мира — ремонтом собственного жилья — рано или поздно сталкивается каждый. И именно в этот момент выясняется, что привычная логика IT здесь часто ломается, а классические строители говорят на совершенно другом языке.
Обычно статьи про интерьеры пестрят рассуждениями о цветовой палитре и визуальном расширении пространства. Но когда заказчик — IT-специалист, его мало волнует цвет штор, если в кабинете нечем дышать из-за превышения CO₂, а при включении чайника ИБП на рабочем месте начинает панически пищать.
За годы работы в Москве я понял одну вещь: современный ремонт нужно рассматривать не как стройку, а как разработку сложного hardware-продукта. В этой статье я хочу разобрать, что такое «технический долг» в ремонте, почему подрядчики часто пишут «говнокод» в инженерных сетях, и как мы выстроили инженерный подход к проектированию жилья на примере реального объекта.
Читать далее
Когда трафик измеряется миллионами пакетов в секунду, даже небольшие накладные расходы сетевого стека начинают иметь значение. XDP позволяет подключиться к обработке значительно раньше, но работать приходится почти напрямую с данными пакета. На практическом примере разберём, как подключить XDP-программу к интерфейсу, читать сетевые заголовки и понять, для каких задач такого уровня обработки достаточно, а где стоит подняться выше по стеку.
Разобрать XDP
Один из самых частых вопросов, которые нам задают: зачем вообще так усложнять архитектуру, если можно просто поднять сервер и раздавать доступ.Отвечу подробно, и заодно расскажу вещь, о которой мы почему-то редко упоминаем, хотя для этой архитектуры она главная.
Читать далееIndustry body says use of private investigators against reporters is ‘attack on our democracy’ after Banks apparently admitted doing so
The businessman and Reform UK donor Arron Banks is facing calls to lay out the extent of his surveillance of journalists, described by an industry body as an “attack on our democracy”.
The News Media Association (NMA), which represents regional and national media, raised concerns after Banks appeared to admit over the weekend to hiring private investigators to spy on a Sunday Times journalist.
Continue reading...
Я решил проверить, можно ли вообще поставить stop-loss объявлению не на глаз, а математически: задать допустимый CPA, посчитать минимальную нужную конверсию и понять, после какого количества нулевых попыток гипотезу уже действительно пора закрывать.
Получилась довольно простая модель — и один неожиданный ориентир, который я теперь точно буду учитывать в тестах.
Читать далее
VictoriaLogs — система хранения и поиска логов от команды VictoriaMetrics. Её идея проста: не заставлять оператора заранее проектировать индексы и схему данных, а дать ему возможность принимать логи и искать по их полям сразу. При этом система должна оставаться относительно простой и на одном сервере, и в кластере.
В этой статье разбор того, за счёт чего VictoriaLogs пытается решить типичные проблемы систем логирования и в каких случаях на него стоит смотреть внимательнее.
Читать далее
Повторяющиеся операции в облачной инфраструктуре — по отдельности совсем не сложные. Проверить доступные мощности, сверить параметры нового объекта, собрать данные об инциденте или рассчитать лимит IOPS — все это можно сделать вручную. Но когда количество кластеров, площадок, клиентов и запросов растет, а количество рук — не увеличивается, это становится проблемой.
Меня зовут Дмитрий, я руководитель службы поддержки виртуальной инфраструктуры в OXYGEN. В этой статье я рассказываю про личный опыт автоматизации мониторинга облака. Под катом разберу четыре задачи, которые мы последовательно автоматизировали — заглядывайте, если интересно.
Читать далее
Доброго времени суток, читатели!
Хочу представить вам unissh – современный, простой опенсорсный SSH клиент с selfhosted zero-knowledge сервером для синхронизации данных.
Читать далее
Эта статья не о том, что SEO перестало работать. Она о том, почему архитектура поиска изменилась настолько, что привычной модели «индексация → ранжирование → клик» уже недостаточно для описания происходящего.
На протяжении более двадцати лет веб-поиск был относительно простой системой.
На высоком уровне ее можно представить следующим образом:
Public displays of affection are widely mocked online and off, but they have a positive side – and now science has proved it
Name: Kissing in public.
Age: Kissing is 17m-21m years old – probably older than the notion of privacy.
Continue reading...
Despite wars ranging across the Middle East, more than 20 million Shia pilgrims gather in Iraq shrine city for Arbaeen festival
(Image credit: Mohammed Sawaf for NPR)
