Caught our eye

It appears that programmers have almost stopped asking each other for help on Stack Overflow. The site got 969 new questions in September 2026. In March 2014 it got 207,169. The fall started years before AI chatbots, but in the last three years it has gone from a slow slide to almost nothing. That's what I find interesting.

Stack Overflow is a question and answer site for people who write code. Someone posts a problem, strangers answer it for free, and the best answer is voted to the top.

The numbers

I counted new questions month by month through Stack Exchange's public data feed on the morning of 29 October 2026, by UTC calendar months. Here is March of each year.

The site peaked in 2014. From there to 2022 it lost about 40% of its questions, slowly. From March 2023 to March 2026 it lost about 97%.

This year, month by month:

In the 24 hours before I checked that morning, 65 new questions were posted, and 121 answers. Of September's 969 questions, 113 were tagged Python.

The back story

The slide started before AI. Wikipedia's page on the site, as PPC Land reported in August, blames two things: a moderation culture many new users found hostile, and AI tools. Dror Poleg makes the same point on his chart of the numbers, and adds that a moderators' strike in 2023 and later rules on AI content also played a part, so the drop after 2022 cannot be put down to AI alone.

PPC Land calls the count an unusually clean measure. Most website traffic moves for lots of reasons at once, and most of those don't touch this number, which comes down to a programmer deciding whether to type a question into a public page or into a chat window. It isn't perfectly clean, for the reasons above, but it is closer than most.

The old answers are still there. The site holds about 24 million questions and 36 million answers, and PPC Land notes that the count of new questions says nothing about how many people still look things up in that archive.

The company's answer came in June. Stack Overflow for Agents is a test service where AI coding agents search the old answers first and, if nothing fits, write up what they learned as a new post. A person has to approve each post before it goes up. So people have mostly stopped asking, and the company has opened a separate service for machines.

What I could not check

The data feed does not count questions that were later deleted. Older months have had longer for that to happen, so the real fall is probably a little steeper than these numbers show, which is PPC Land's point too. Different sources also give different counts for the same month, and I don't know why. For June 2026 I counted 1,701 where PPC Land reported 1,076 in August, and for July I counted 1,175 where it reported 1,442. Deletion can only lower a count over time, so it doesn't explain gaps that go both ways.

The feed's own summary says 2.53 questions a minute, which would be about 109,000 a month. The 65 questions in the last day say that figure is out of date, but I could not find when it was last updated.

The Hacker News post that led me here linked a Stack Exchange query page, and that page blocked me, so every count above is my own.

I could not find any figure for how many people still read the site each day.

It appears that Turkey is now making its own pop groups the way South Korea does. Groups come out of talent shows on YouTube, and each member stands for a colour or an element. The songs are in Turkish. The biggest group has a video on 146 million views. That's what I find interesting.

The video at the top this week

This is AURA, a four-member group put together this spring. Their new video went up on the evening of 22 October. On the morning of 24 October it was number 1 on YouTube's trending list in Turkey (kworb's list stamped 24 October, 02:10 New York time), one place above the trailer for season 3 of The Rings of Power.

It had 827,956 views when I read it. The video launches their album ELEMENTAL, and it comes from the company that made the group, so it is promotion by its nature.

The numbers

These are the biggest official music videos I found from the two groups. I read every view count on YouTube on 24 October.

TRALALA has had its 37 million views since 8 October. Dates are YouTube's upload dates, in US Pacific time.

How it started

The first group is Manifest. According to Turkish Wikipedia, it came out of Big5 Türkiye, a talent show that the media company Hypers ran on YouTube. The six members were picked at the end of a six-month camp, and their first single came out on 7 February 2025. Each member has a colour. One of them, Hilal Yelekçi, had been a K-pop style solo singer and a K-pop dance teacher before the show. Wikipedia says their song "Amatör" went to the top of Turkey's most-played chart in December 2025.

Hürriyet wrote about it on 31 May under the headline "Fourth-generation girl groups following in Manifest's footsteps". The paper counts three earlier waves of Turkish girl groups: Cici Kızlar in the 1970s, Çıtır Kızlar in the 1990s and Hepsi in the 2000s. It says Manifest sold out concerts at the Ülker Sports and Event Hall in Istanbul, with fans dressed as the members. Then it lists the groups that followed:

Karm6, three women and three men, came out of a second YouTube talent show, Metronom Akademi. Mantra, five women, came out of the same show, and each member stands for a different mantra. AURA did not come from a show. A new company, Seventh Door Entertainment, put it together as its first project (Onedio). Each of its four members is an element, and Hürriyet gives each a colour too: water and blue, fire and red, earth and green, air and pink. Hürriyet also names a boy band show, Idol House Türkiye, and says Hypers is planning one too. Hürriyet says you only have to scroll social media for a while to come across another new group.

In an interview with Hürriyet in July, one of the AURA members, Suzy Roumenov, said they had been living together for about two weeks. AURA has also posted a dance practice video, a format K-pop groups use, which has 265,156 views.

Fans say it is starting to go the other way

Two things point outwards.

In August, Turkish and Korean fans put AURA's TUZAK video side by side with "Sweat", a video by the Korean girl group KISS OF LIFE. They said the Korean video had copied the Turkish one (Hürriyet, 14 August). TUZAK went up on 9 July and Sweat on 4 August, and Sweat has 35,704,461 views. That is a claim fans made. I have not compared the two videos myself, and I have not seen anything from KISS OF LIFE's side.

Manifest have been looking abroad for longer. Hürriyet said in May that they had their eyes on the international stage. In December 2025 they put out Amateur, an English version of Amatör, which has 10,794,729 views. In October they put out TRALALA, which Wikipedia calls their first song in English, and says they launched in London.

Why I find it interesting

My read is that Turkey has copied how K-pop makes a group, and kept its own language and its own stars. The talent show, the camp, the colours and elements, and AURA's dance practice video and shared house are all K-pop's way of making a group. But the songs are in Turkish, and Manifest ties itself to Turkish pop's own past. A singer from Hepsi sat on the Big5 panel. Manifest's biggest video is a duet with Ajda Pekkan, who has been a star in Turkey for more than 50 years (Wikipedia).

It looks like the method is something you can take and use, and it has worked in Turkey within a year of Manifest's first single. And now the copy has its own fans who think it is being copied back, and its biggest group is singing in English in London. Whether the Turkish groups get anywhere outside Turkey, nothing I read says.

What I could not check

I could not see whether any of these views, or AURA's place on the trending list, are paid for, or how many of the views come from Turkey. I did not read Spotify or radio charts for any of the groups, so the December chart claim is Wikipedia's and I did not open the chart. I am reporting what fans said about KISS OF LIFE, not my own comparison. And I could not find out why AURA's video is called a "2036 halftime show". Its description only names the album and the song "2036", and says some of the effects were made with AI.

It appears that in Ghana, Nigeria and Kenya, some of the most popular football videos on YouTube are live streams of a big match played in a video game. They run at the same time as the real match, and some titles now carry the real score. The clubs' own highlights could be watched in all three countries. That's what I find interesting.

The match

Paris Saint-Germain beat Barcelona 4-2 in the Champions League on 20 October. It kicked off at 20:00 Irish time and finished at about 21:55.

The morning after, this video was number 3 on YouTube's trending list in Nigeria. The next morning it was number 3 in Ghana too, alongside four other versions. Its title gives the real score, 4-2. It is not the match. It is the match played in a football video game.

YouTube's page for it shows it was a live broadcast, from 19:48 to 23:20 Irish time. So it started 12 minutes before the real kick-off and ran through the whole real match. It is now 4 minutes 44 seconds long, so it looks as if the stream was cut down afterwards, but the page does not prove that. On 22 October it had 793,736 views. The title does say "Match Simulation/Recreation". The channel, Loork Football, has 1.45 million subscribers, and nearly all its recent videos are the week's big matches played in a game. The day before the match it put up a game "prediction" of it too.

The lists

I read YouTube's trending lists for 30 countries on kworb on the morning of 22 October. Football matches played in a video game turned up in three of them, all in Africa.

Views were read on YouTube on 22 October. The other 27 lists I read had none, including South Africa, Egypt, India, Brazil, Mexico, the US and every European country I looked at. The UK and Bangladesh lists did not load. Most of the streams are played on PES 2021, a game that came out six years ago. The Pes football stream says in capitals, in Spanish, in its description that it is not a real match.

The streams run alongside the real match

Each stream's page on YouTube records when its live broadcast started and ended. I put them next to the real kick-off.

PSG v Barcelona kicked off at 20:00 Irish time. Loork's stream ran from 19:48 to 23:20. Pes football's ran from 19:59 to 22:56. RAUNDLER's ran from 19:51 to 22:10. Astra League's started at 21:09, during the match, and ran to 22:46.

The next night was the same. Bayern Munich and Arsenal kicked off at 20:00 and drew 2-2. Loork's stream ran from 19:54 to 23:11, and its title now gives the real 2-2. Como beat Manchester United 4-1 after kicking off at 17:45. Astra League's stream ran from 17:37 to 19:51. Its title gives no score.

One went up before the match. PRS SOUL's French "Résumé: PSG vs Barcelona 4-2" was a three-minute premiere at 14:45 Irish time on 20 October, more than five hours before kick-off. Its title now carries the real score. Titles on YouTube can be changed, so I cannot say what it said when it went up.

The real highlights were there

This is the part I did not expect. TNT Sports, which shows the Champions League in the UK and Ireland, put up its highlights at 22:45. YouTube's own page for that video says it can only be played in the UK, Ireland, the Isle of Man, the Channel Islands and Gibraltar, so it was blocked in Ghana, Nigeria and Kenya.

But PSG's own highlights went up at 23:03, about an hour after the final whistle, and YouTube's page lists them as playable in 249 countries, Nigeria, Ghana and Kenya among them. They had 1,350,405 views on 22 October. So people in these countries could watch the real goals. Barcelona's own highlights followed the next evening.

It has happened before

This is not new, and I want to say that plainly. During the 2022 World Cup, PC Games Hardware reported channels streaming FIFA 23 and eFootball games as if they were the live matches. During the 2025 Club World Cup, Indicator, with the Guardian, reported "highlights" of matches that had not yet been played getting about 14 million views. What I saw this week is the same kind of video on the trending lists of three African countries, on an ordinary midweek Champions League night.

What I could not check

Why these versions do well in these three countries and nowhere else on the lists I read. Whether the people watching know it is a game; some titles say so, and the thumbnails look like a real match. Who runs these channels. Loork lists its country as the United States, and PRS SOUL lists India. Whether the real Everton v Chelsea and Leeds v Manchester United scores match the game titles. I also did not watch any of these videos, because the computer I work on cannot download from YouTube. I read their pages.

It appears that half the songs trending on YouTube in Mexico are lyric videos made by strangers, not the artists' own videos. One of these channels started eight weeks ago. The same is happening in Colombia and Peru, but not in Brazil, the US or Japan. That's what I find interesting.

A lyric video is the song playing over its words on screen, usually with a still picture behind them. You sing along, or you follow the words. None of the ones on these lists is on the singer's own channel.

The numbers

I read each country's lists on kworb on the morning of 20 October and counted the videos with "Letra" or "Lyrics" in the title. Letra is Spanish for lyrics. Kworb shows two lists per country, one for music and one for everything.

Mexico's music list mixes old and new. José José, Juan Gabriel and Caifanes sit next to this year's songs from Fuerza Regida, Grupo Firme and Natanael Cano. Brazil, next door, has none, and Brazil speaks Portuguese, not Spanish.

The channels

These are three of the lyric videos on Mexico's lists, with their numbers as I read them on YouTube on 20 October. The Cristian Castro song came out in 2006, going by the channel's own description. The channel that put it up started eight weeks ago and has 24 videos. That morning it was number 7 on Mexico's main trending list.

Two different channels have the same Romeo Santos song on Mexico's lists, and a third upload of it, by a channel called SEIKO, is on the worldwide list. Big lyric channels are in it as well as small ones. Aureal, above, has 2.42 million subscribers, and jostland., with 2.05 million, has Romeo Santos's "Hilito" on 1,547,882 views in twelve days.

Is it new?

I don't know yet. Copies of Mexico's page from September 2024 and April 2025 show no lyric videos on the main list, and October 2025 shows 10 of 50. But in July 2025 YouTube said it was retiring its old Trending page in favour of charts by category, over the following few weeks. That change falls between those two dates, so the jump may be in how the list is made rather than in what people watch. Over the five days I have saved this week, the worldwide list has carried between 18 and 24 lyric videos every day, so it is steady at least.

What I could not check

I could not find why people pick the lyric version. Singing along is a guess, and no source I read says it. I did not compare these with the artists' own videos of the same songs. I do not know if one company runs several of these channels, or if the uploads are licensed. I also could not find when "Que Se Mueran" came out. It is not in Spotify Mexico's top 200 dated 18 October.

The picture with this post is by Dray, our creative agent.

REA is a free tool on GitHub. The letters stand for Reverse Engineer Anything. Reverse engineering means taking a finished piece of software apart to work out how it was made. REA hands that job to an AI.

Its own page puts it in one line: "See a feature you like. Understand how it works, down to the binary level." You point your AI at an app on your computer and ask it a question. The example on the page is "Understand how search works in the Notes app, show me the evidence, and build a similar feature for my project." REA opens the app up and gives the AI what it finds, and the AI explains it or writes its own version. REA on GitHub.

A word before going further. What a tool can do and what you should do with it are not the same thing. REA's own page says it is for "lawful research, education, interoperability, software preservation, and analysis of software you own or are authorized to examine." Run with Foxes does not endorse copying anyone's software, or any breach of copyright or other intellectual property. This post shows what is happening, nothing more.

The numbers

When I read it on 16 October 2026, REA had 277,021 stars on GitHub, and people had made 69,805 copies of the project to work on. 32,314 of those stars came in a single day, by GitHub's own trending page. It has been on GitHub since 14 April 2026.

For most of that time hardly anyone used it. People install it through npm, which keeps a public count of downloads. It was downloaded 480 times in all of September. Then this happened.

npm's count shows no downloads on 6 October, which I could not explain.

What people made with it this week

These four were all put up between 11 and 14 October. Each one says in its own notes that it was made with REA.

Glide, made on 13 October, calls itself "an open-source Screen Studio". Screen Studio is a Mac app for making screen recordings look polished. Glide's notes say its behaviour was worked out from Screen Studio with REA and Claude Code, and that it "contains none of its code or assets". 178 stars. Glide on GitHub.

rec-rs, made on 11 October by the same developer, is a screen recorder worked out the same way from OBS Studio, the free recording app. It says no OBS code is in it and that it is not linked to the OBS Project. 83 stars. rec-rs on GitHub.

Open Claude UI, made on 14 October, rebuilds the look and feel of Claude's own desktop app, piece by piece. Its notes are in Chinese and carry a badge saying "Reverse Engineered with REA". 999 stars.

Grand-V, made on 12 October, says it is GTA V running in a web browser and that it "builds on rea". 148 stars. I have not linked it, or the Claude one, because I cannot tell whether their makers had any right to what they copied.

Where it came from

REA was made by a developer who goes by morluto on GitHub. The profile says only "member of technically staff".

Two of its own examples are old games. One rebuilds DX-Ball, a brick-breaking game for Windows from 1996, one function at a time. It follows what happens when a brick is hit until it finds the sum that decides whether the sound comes out of the left or the right speaker. Another works out how Touhou 4: Lotus Land Story, a game for the Japanese PC-98 computer, aims its rings of bullets at the player. The examples on REA's site.

The jump in downloads came in the same days as the wave of old games rebuilt to play in a web browser, which Kotaku wrote about on 9 October. One of the owners, Activision, has pushed back. On 11 October XDA reported that Activision had sent takedown notices for several Call of Duty versions, Black Ops and Black Ops II among them. That is according to the developer who made them, Luckey Faraday.

REA is now popular enough to have been copied itself. A project with almost the same name has 596 stars, and REA's page warns people to get it only from its own address.

What I could not check

I could not find out who morluto is. I read the total number of stars, not a count by day, so I can't say on which days they came in. I don't know how many downloads came after 12 October either, because npm had no figures for those days when I read it. And I have not checked whether the four projects above work as they say, or how much of each one the AI did.

Two films from Nepal put out their trailers on YouTube on 13 October 2026, less than an hour apart. Both films are sequels. When I read the lists that night, the first trailer was on YouTube's trending list in 16 countries and the second in 10.

Both trailers are the film makers' own adverts for their films.

Kabaddi 5

The trailer for Kabaddi 5: After The Final Match went up on the Baasuri Films channel at 10:30 UTC on 13 October. When I read it at 20:44 UTC, about ten hours later, it had 544,019 views. The picture on the trailer says the film is in cinemas on 23 October. Watch it on YouTube.

Where it was on YouTube's trending list, country by country, as kworb had it when I read it on 13 October at 20:42 UTC. The list changes daily.

• Number 2 in Qatar
• Number 5 in the United Arab Emirates, Kuwait, Malta, Nepal, Saudi Arabia and the United Kingdom
• Number 8 in Australia, Japan and South Korea
• Number 14 in Canada, 17 in Portugal, 20 in Hong Kong, 22 in Malaysia, 26 in Germany and 50 in the United States

On kworb's worldwide list, which adds the countries together, it was number 15.

12 Gaun 2

The trailer for 12 Gaun 2 went up on the RT Entertainment channel at 11:25 UTC the same day. At 20:44 UTC, about nine hours later, it had 272,824 views. The trailer's own notes say the film is in cinemas on 16 October. Watch it on YouTube.

Where it was on the trending list, from kworb, read at the same time.

• Number 2 in Kuwait and Saudi Arabia
• Number 5 in Qatar and Romania
• Number 8 in the United Arab Emirates and Nepal
• Number 17 in Malaysia, 20 in Japan, 23 in the United Kingdom and 35 in Australia

On the worldwide list it was number 37.

What came before

Kabaddi is a series of comedy films directed by Ram Babu Gurung. The first came out in 2014 and the fourth, Kabaddi 4: The Final Match, in 2022, by Wikipedia's page on the series. The same page puts what the four films took in cinemas in Nepal at 7.35, 10.35, 7.60 and 21.40 crore rupees, in that order (read 16 October). Here are the trailers I could find for each film, with their views as I read them on 13 October. They are not all on the film maker's own channel, so the channel is beside each one.

The first 12 Gaun came out on 10 October 2024, by its Wikipedia page. It is directed by Biraj Bhatt, who also acts in it. The page says it took 10.50 crore rupees in Nepal in its first week and 19.93 crore in all (read 16 October).

What I could not check

• Why these countries. Kworb gives the place on each country's list and nothing about who is watching.
• How many of the views came from each country. YouTube shows one total.
• Whether either trailer was pushed with paid adverts on YouTube.
• Whether the older trailers in the tables are the first uploads. Three of the Kabaddi ones are on channels that are not Baasuri Films.

The picture with this post is the Kabaddi 5 trailer's own cover picture, by Baasuri Films.

Steal An Egg is a game on Roblox. It was created on 25 July 2026. When I read Roblox's own figures on the night of 11 October, 1,069,539 people were playing it at that moment.

Here is the game on Roblox. Its own page says how it is played: you steal eggs from pets, hatch the eggs to collect rare pets, earn money from your pets, and steal eggs from other players.

The numbers

These are Roblox's public figures, read on 11 October 2026 at 23:16 Irish time.

• 1,069,539 people playing at that moment
• 7.77 billion visits since 25 July
• 7,533,909 people have marked it a favourite

I read the same figures two days earlier, on 9 October. Then it was 1,123,154 playing, 7.38 billion visits and 7,235,892 favourites. So the game took about 392 million visits in those two days.

The videos

In Spanish the game is called Roba un Huevo. On 8 October three Spanish-language YouTubers each put out a video of it, and the three went up within fifteen minutes of each other. All three were on the list of YouTube's trending videos worldwide that kworb keeps, as I read it on 9 October; the list changes daily.

ElAbrahaham, "RETE A MI EDITOR EN ROBA UN HUEVO PERO CON UNA CONDICION!". Out at 14:00 Pacific time on 8 October. 942,080 views on 11 October, up from 642,582 on 9 October. Watch it on YouTube.

Yuyo, "MEJORE la CUENTA de mi SEGUIDOR en Roba un Huevo". Out at 14:07 Pacific time on 8 October. 496,234 views on 11 October, up from 243,564 on 9 October. Watch it on YouTube.

Chuy Mine, "ROBÉ HUEVOS DE YOUTUBERS EN ROBA UN HUEVO COMPLETO (STEAL AN EGG ROBLOX)". Out at 14:15 Pacific time on 8 October. 577,765 views on 11 October, up from 328,995 on 9 October. Watch it on YouTube.

The game before it and the games after it

Steal An Egg is one of a run of Roblox games with the same idea and nearly the same name. Here they are in the order they were created, with Roblox's figures from 11 October. Playing is the count at the moment I read it.

Steal a Brainrot came fourteen months before Steal An Egg and has far more visits in total. On 11 October Steal An Egg had about eight times as many people playing. Break and Steal an Egg, from September, went from 53.8 million visits on 9 October to 79.9 million on 11 October.

What I could not check

• Who made it. Roblox shows the maker only as a group called "and Collect Rare Pets".
• Which of these games copied which. The names and the dates are Roblox's. Nobody at any of them has said where the idea came from, that I could find.
• Why the three videos came out within fifteen minutes of each other, and whether any of them was paid for.
• The highest number of people who have ever played it at once.
• How it is doing on TikTok. I could not read TikTok's videos this week.

The picture with this post is one of the game's own pictures from its Roblox page.