OpenAI spends billions building AI. Chinese companies are learning from its answers — and could turn its own advantage into a weapon against it.
According to The Information, DeepSeek and Moonshot AI used outputs from GPT-6 Astra to train their own models.
It’s called distillation. And it could change the game.
OpenAI pays for breakthroughs.
Its competitors are looking for shortcuts.
What if OpenAI’s biggest threat isn’t another AI model, but its own?
Arnold Schwarzenegger says he made more than $40 million from Twins.
His salary for the movie? $0.
When Andy Cohen asks which film earned him the biggest paycheck, Arnold names the comedy he made with Danny DeVito.
He explains that they took no salary. The deal gave them ownership and a piece of the back end.
Cohen asks whether Arnold made more than $20 million.
“Oh, yeah. Much more.”
So he raises the number: more than $40 million?
“It was more than that.”
Arnold never gives the exact figure. But he makes clear where Twins ranks among all the movies in his career:
“It was more than in any movie I’ve ever made.”
DO YOU ACTUALLY BELIEVE THIS?
Videos are spreading everywhere claiming “Claude Code found 200k+ long-deleted photos” while the phone shows a green terminal with a progress bar “recovering” everything.
All I see are typical AI-generated clips. The creators say “I plugged in the phone and the code found the files,” but the interface looks staged, not like a real process.
Yes, thumbnail caches on Android and iOS sometimes stick around, but these dramatic “scanning” videos with progress bars are pure clickbait.
Has anyone actually tried this themselves and recovered something real, or is everyone just watching and reposting? 🤔
Scattered minds: what research says about living in "everything at once" mode
A game, music in your earbuds, and Reels on a second screen.
Researchers call this media multitasking, and short-form video (TikTok, Reels, Shorts) is now studied as a separate factor.
What the data shows
The largest review so far was published in Psychological Bulletin. It covered 71 studies and 98,299 participants.
More frequent short-form video use was linked to poorer cognitive performance (r = −0.34), most strongly for attention (r = −0.38) and inhibitory control (r = −0.41).
Inhibitory control is the ability to ignore distractions and stay on task.
The link with mental health is weaker (r = −0.21), but the clearest correlations are with stress (−0.34) and anxiety (−0.33). The authors also recorded a link with sleep disturbances.
No consistent link with self-esteem or body image was found overall.
A possible mechanism
The most common explanation is that the brain adapts to rapid context switches and instant reward.
This puts more load on the prefrontal cortex, which handles sustained attention. Sleep loss, often tied to late-night scrolling, further weakens self-control. The result is a loop: a scattered brain reaches for fast content, and that content further wrecks sleep and recovery.
What about multitasking?
Here the picture is more complicated. Ophir's pioneering 2009 study showed that frequent multitaskers filter out irrelevant stimuli worse.
Later checks were less clear-cut. A 2017 meta-analysis found only a weak link with distractibility, which became statistically nonsignificant after correcting for small-study bias.
A later three-level meta-analysis of 43 studies did confirm worse inhibitory control and working memory in heavy multitaskers, with no difference in cognitive flexibility.
Important caveats
These are correlations, not proven causation. Impulsive people may simply spend more time on short videos.
Many studies rely on self-reported usage, and few accounted for overall social media time.
The authors call for experimental and longitudinal studies. This is also not a diagnosis: ADHD or an anxiety disorder is diagnosed by a doctor, not by hours spent on Instagram.
Consequences already described in the literature
1Weaker sustained attention and self-control.
2Higher stress and anxiety scores among heavy users.
3Worse sleep.
4Lower academic performance and more procrastination.
A review of research from 2019 to 2025 describes a dose-response pattern: the more time spent on short videos, the more pronounced the attention problems.
Conclusion
The evidence doesn't say a phone breaks your brain. It says that the more fragmented content you consume, the worse the average outcome for attention, stress, and sleep. If you notice you can't finish reading a text, never feel rested, or are constantly anxious, talk to a doctor or psychologist rather than just setting app timers.
I thought those files were gone forever.
I deleted them long ago. Forgot about them. Moved on.
Then I gave Claude Code access and one task: find what's left.
A few minutes later, there was a list on my screen.
The files never went anywhere. When you hit "delete", the system just marks the space as free. The data stays there until something new overwrites it.
It used to take paid software, forums and half a day of nerves.
Now it's one task for an AI agent.
You don't delete data. You just stop seeing it.
⚠️ One USB at the airport — and your iPhone is no longer yours
The video shows what a “random” connection to an airport charger can actually look like.
An iPhone 15 Pro (iOS 17.4.1) had a profile called “Airport Free Wi-Fi” from “Airport Services LLC” installed. This is not Wi-Fi. It’s a full MDM:
• DeviceLock and RemoteWipe
• Silent app installation
• Access to camera, microphone, and screen recording
• Continuous location tracking
• Reading iMessage/SMS, photos, and contacts
• HTTPS interception (banking apps, passwords, 2FA)
The attackers’ server was located in Chisinau (Moldova). Over 19 days:
• 612 location pings (home, work, kids’ school, routes)
• 94 screenshots, 4,291 photos, 1,247 contacts, 2,840 URLs
• $12,400 stolen via Zelle and wire transfers — using intercepted 2FA codes
No notifications were shown to the user.
What to do if this has already happened
• Remove the MDM profile and all suspicious certificates
• Change all passwords (especially email, banks, and Apple ID)
• Freeze cards and dispute the transactions
• Check your devices for any other profiles
Main rule: Never use public USB chargers.
Bring your own cable + power bank, or plug into a wall outlet.
Full breakdown is in the video.
How many of your "deleted" photos are still on your iPhone?
During a security check, Claude Code found PhotoData/Thumbnails.
Thousands of thumbnails: bank app screenshots, photos deleted seconds after being taken, even ones purged from Recently Deleted.
You clean the gallery. The phone keeps the archive.
When did you last check what it actually keeps?
OpenAI just released the most boring — and most useful — model of the year.
GPT-6.1 Terra isn’t “another AGI
breakthrough.” It’s a model that can finally actually work for you, instead of just sounding smart.
It sees the screen, clicks through interfaces itself, reads technical drawings, fills out forms, processes invoices, and writes text like it was done by someone who doesn’t hate their job.
And it costs half as much as the previous Terra ($1/$5 per million tokens) while being several times faster than the top-tier Sol and Astra.
Basically, OpenAI finally realized that most people don’t need a genius that solves olympiad problems. They need a fast, cheap coworker that doesn’t choke on routine work.
It’s already rolling out to Plus, Pro, and Enterprise users in ChatGPT Work and Codex.
Anyone who’s already compared it to what was there before — does it actually feel different, or is it just “a bit better and cheaper”?
OpenAI just dropped GPT-6.1 Terra and basically said: “your AI can now click around on its own and not eat all your money.”
It can actually see what’s on screen, take actions across apps, and write more naturally — while costing half as much as the previous version and finishing tasks over 3× faster than the bigger models.
Sounds like exactly what people need for automating daily work. Whether it actually works well enough yet, or it’s another “almost there,” we’ll find out soon.
Bill Gates: “I thought that after COVID we would become more prepared for future pandemics.
But that one didn’t work the way I expected.”
He then predicts large-scale cyberattacks in the coming years — because there’s a business model there.
Interesting what exactly “didn’t work” as planned.
And who benefits from the next crisis — whether biological or cyber.
An intriguing announcement from OpenAI.
While it’s a clear step forward for [industry/tech], it’ll be interesting to observe the competitive response from Google and Anthropic.
The true test will be its reliability in production, rather than controlled demo environments.
Thoughts?
A ghost ship for $0.44 vs $5.65.
One scene, two models: Kimi K3.1 Pro and ChatGPT-6 Astra.
The price gap is 13X. Skeletons, pumpkins, glowing sea. Watch the video and find where the extra $5 went.
GPT is an expensive brand. Kimi is almost the same result for pennies.
Are you paying for quality or for a logo?
Elon Musk shared some lesser-known details about Tesla’s financing during its crisis years:
• The U.S. government provided the company with a loan of nearly $500 million under an alternative energy support program, which significantly accelerated Tesla's progress.
• However, what actually saved the company wasn't the government loan, but a critical investment from Daimler (the maker of Mercedes) in May 2009 — at the time, they were the only ones.
• Tesla fully repaid the entire government loan ahead of schedule, including interest and a prepayment penalty.
• Even though the loan wasn't due for another 10 years, Musk paid it off early because he doesn't like having a lot of debt and wanted to say thank you to the taxpayers who supported Tesla.
Starlink Mobile is not just “another operator from space.” It is an attempt to build, for the first time in practice, a hybrid network where terrestrial low-band spectrum and direct satellite connectivity work as one system.
The key elements that actually change the picture:
• 800 MHz low-band is spectrum that penetrates buildings well and has long range. Combined with terrestrial sites, it addresses exactly the scenarios where pure satellite connectivity has traditionally been weak (indoors and dense urban environments).
• 15,000 Gen2 satellites with significantly higher capacity + 80,000+ terrestrial sites mean SpaceX is not trying to merely supplement existing carriers, but to become a full-fledged fourth player with its own dense network.
• The most important technical point: the phone already in the user’s pocket does not require a special antenna. This removes the main adoption barrier for direct-to-device satellite connectivity.
From a market perspective, this is a serious challenge for AT&T, Verizon, and T-Mobile. For years they have optimized their networks around the economics of “build where the subscribers are.” If SpaceX can deliver acceptable quality both in rural areas and in cities through the combination of spectrum + satellites, the classic logic of “coverage = capital expenditure on towers” starts to break down.
At the same time, there are real risks:
• Scaling 80,000 sites is not only about money, but also about permits, backhaul, and operational complexity;
• Integrating the satellite and terrestrial layers into a single network (handover, quality management) has not yet been proven at this density;
• Regulators and competitors will not sit idle.
If SpaceX succeeds in proving that the hybrid architecture works in practice, it will set a precedent not only for the United States but also for other markets where traditional operators chronically fail to cover “uneconomic” territories.
In short: this is no longer a concept. It is a bet on rewriting the rules of the mobile connectivity game. We will see the results in 2–3 years, once the network is actually operating.
Imagine a robot patrolling your building at night instead of a human guard.
It sees. It analyzes. It patrols.
And when something goes wrong, it alerts an operator within seconds.
Science fiction? Not anymore. 🤖
Today a critical hole was found in PHP that lets attackers remotely take over a server without any login.
It’s the language that powers the majority of websites. The patch is already out, but attackers were ready long before it dropped.
If you run any sites or use services built on PHP — brace yourself for possible fallout.
NVIDIA no longer sells chips. It sells a FACTORY in a shipping container.
Vera Rubin DSX POD: racks, networking, power, liquid cooling. All built and tested at the factory.
Ship it, plug it in. Live in 14 days instead of 6 months (per NVIDIA's slides).
Dell, HPE, Lenovo, Foxconn are already on board. Deliveries are promised from 2027.
Rivals are still comparing chips. NVIDIA is selling the finished infrastructure.
Richard Hatch won the first season of Survivor and walked away with $1 million. Twenty-six years later, the IRS wants $3.3 million — and the number is still climbing.
51 million people watched him win in August 2000. He never paid the tax on the prize.
His story: producers were secretly feeding other contestants, and the show had promised to handle his taxes if he won. CBS has stayed silent. Courts found no proof of any such deal.
A jury convicted him in 2006. He served 51 months. In 2011 he went back for another nine months for failing to fix his returns. Meanwhile the bill kept growing with interest.
This past March a federal judge allowed the IRS to collect roughly $3.3 million. Liens are already on two of his homes in Newport. He’s appealing on his own, without a lawyer.
In the video he argues the tax on the million was supposed to be paid to Malaysia by CBS or Mark Burnett before anyone left the island.
So was the first Survivor winner set up by the show — or did he just try to outplay the wrong opponent?