Posts of Hi, AI • Tech News

🚀 Starship Reaches Orbit for the First Time Elon Musk's SpaceX has successfully launched its super-heavy rocket Starship into space, despite the failure of one of six Raptor 3 engines during ascent. This is the ship's 14th flight and its first orbital one: the previous 13 launches were suborbital, and five of them ended in failure. The ship delivered 26 next-generation Starlink satellites to orbit. It was originally supposed to circle the Earth about six times, but after reaching orbit, SpaceX decided to shorten the flight to three hours, after which the ship will splash down in the Pacific Ocean. 💡 In the future, NASA plans to use modified Starships to land astronauts on the Moon, but this will require refueling by other ships in orbit. Do you follow space launches? ❤️ — Yes, it's exciting every time! 🤔 — Not very interesting... @hiaimediaen

💡 Nanoparticles Helped Blind Mice Respond to Light Again Researchers at Aarhus University in Denmark have developed a method that can make a blind eye respond to light again. Nanoparticles are injected into the eye, where they respond to light and activate surviving retinal neurons, bypassing dead photoreceptors. The researchers see the technology as a potential "injectable retinal prosthesis." In some eye diseases, photoreceptors-the cells that convert light into neural signals-gradually disappear. But the neurons that normally carry those signals to the brain can remain intact. The researchers found a way to reconnect them with light. They created hollow carbon nitride nanoparticles about 300 nm in diameter, with a structure inspired by chloroplasts-the parts of plant cells that capture light. After injection, the particles settle on the retina near surviving neurons and help activate them when exposed to blue light. 🔍 The team tested the approach in blind mice. One month after the injection, 3 of 7 mice began preferring the dark side of their enclosure, similar to healthy mice. The control animals showed no such preference. The researchers also detected light-triggered activity in the visual cortex in 5 of 14 treated eyes, compared with 1 of 14 control eyes. With such a small sample, however, the difference was not statistically significant. The effect was also demonstrated in isolated pig retina: after the nanoparticles were added, light increased the firing rate of retinal neurons almost sixfold. ✅ The researchers now want to deliver the nanoparticles beneath the retina and make them responsive to red and infrared light. If the technology can be developed further, it could eventually offer a much simpler alternative to surgically implanted electronic retinal prostheses. @hiaimediaen

❗️ Could AI cause a catastrophe? If powerful AI falls into the wrong hands, it could cause enormous loss of life, Bill Gates warned in an NBC interview. "AI is certainly powerful enough to drive events that, you know, cause a billion deaths. Even though it's pretty hard to get to 100%, there's never been a weapon as powerful as the combination of people with ill intent using the latest AI tools," Gates said. 💡 Gates is describing a risk, not predicting a catastrophe. He also argues that AI companies cannot be left to police themselves. Lawmakers and law enforcement, he says, should help establish mandatory safeguards and oversight. Should governments require AI companies to follow safety rules? ❤️ — Yes, oversight is needed 🔥 — No, companies can manage the risks themselves @hiaimediaen

No text

No text

No text

No text

No text

🔴 Suno v6 Is Now Live on @GPT4Telegrambot Suno has released v6, its first models trained on licensed music. They were built together with the major labels the startup was suing not long ago. The "cleaner" dataset hasn't hurt the quality, though: v6 captures style well and makes hits in many genres. How to create your own track: 1️⃣ Go to @GPT4Telegrambot 2️⃣ Select Music Generation → Suno → Custom Mode. 3️⃣ Describe the sound you want and add your lyrics, split into verses and choruses. 4️⃣ Select the v6 model and tap Create. ✅ For a bolder sound, there's Wild. For fast generation, there's Mini. 💡 Tip: describe your idea in detail, including tempo, instruments, mood, and vocal style. Tags in square brackets inside the lyrics help you set up a quiet verse, a build-up, and a powerful chorus. @hiaimediaen

🎧 Player Complained About "Awful AI Voice Acting." It Was the Developer's Wife A Steam user played indie game Luminary for just 24 minutes before leaving a negative review. One character's voice sounded AI-generated to them: "The awful AI-generated voice killed it." "While my wife might not be the greatest voice actress out there (that was her first time), I can assure you she is a real person and not AI haha," game developer Keith Kolod replied. ❤️ Kolod's wife, Mikell, voiced the Temple Guardian-an ethereal goddess who has delivered the same speech to hundreds of heroes over the centuries. Mikell had never done voice acting before. She recorded the lines at home with her husband, stumbling over words and worrying about her delivery. They eventually got the performance through multiple takes, recording just a couple of sentences at a time. Before the update featuring her voice went live, Mikell had one request for her husband: "Ok, but I don't wanna hear if people hate my voice in reviews or on X. I did what I could." Luminary is the first game from Kolod's two-person studio. He says he created an accessible action RPG so he, his wife, and their son could play together. The studio doesn't use AI in the game at all. Kolod says he has nothing against the technology, but voice acting is a way for him to "leave parts of me, my family, and friends" in the games he makes. And after all this, he admits it "may be a hard sell" to convince Mikell to do it again. Would you play a game with AI-generated voice acting? ❤️ — Yes, no problem 🔥 — No, it's soulless slop @hiaimediaen

🎓 Tutoring Centers Are Giving In to AI The Australian Dymocks is closing five Sydney tutoring centers after struggling to compete with AI. Company director Mark Buckland says the technology has made the business model unsustainable. "Most of our customers would be better off using their hard-earned cash on cheaper solutions rather than simply paying for overpriced tutoring," Buckland frankly admits. An AI subscription costs around $20–30 a month, while a human tutor can cost hundreds of dollars per subject each term. Dymocks still believes a real teacher provides a better learning experience than a chatbot, but for many families, that difference no longer justifies the cost. ✈️ Students themselves are also increasingly choosing AI tutors. In the U.S. alone, the share of high school and college students using AI to help with homework rose from 48% to 62% between May and December 2025. Dymocks appears to be one of the first to accept these irreversible changes in education. Will AI replace human tutors? ❤️ — Yes, it already is 🔥 — No, students still need human teachers @hiaimediaen

❤️ AI Helps Find Lost Cats Since 2025, Kazakhstan has been running a project called loCATor. Volunteers fit their outdoor cats with a camera and a GPS tracker. The camera films every cat they meet along the way, and an AI compares them with a database of missing pets. If there's a match, the owner gets a notification in the Telegram bot with the coordinates where their cat was spotted. The bot was originally built for ordinary passers-by to send in photos. Now they have help from "scout" cats, who can squeeze into places people can hardly reach. 🌐 The project was launched by creative agencies Ogency and JAM together with Whiskas. Whiskas also monitors the database in real time. When a new missing pet appears, its photo is shown on digital billboards in the same area, since cats rarely wander more than 500–800 meters from home. Photos of lost cats are also printed on cat food packs. @hiaimediaen

🤖 Claude Discovered a System Resembling CRISPR Claude autonomously discovered a novel enzyme system with an array of DNA repeats, similar to CRISPR—a bacterial immune mechanism that scientists use as "scissors" for genome editing. This is the first result from Anthropic's new biolab, where humans test AI hypotheses experimentally. 🔍 Scientists tasked Claude with finding unusual reverse transcriptases—enzymes that copy genetic information from RNA to DNA. The AI agents independently analyzed the sequences and searched for anomalies that might fall outside previously known criteria. Over 21 hours, approximately 950 AI agents used 210 million tokens and identified 3,500 potentially new systems, narrowing the list to the 20 most promising candidates. One agent drew attention to a "spectacular" array of repetitive sequences near the reverse transcriptase gene in the genomes of bacteriophages (viruses that attack bacteria). Previous studies had identified the enzyme itself, but no one had noticed the repeats and the adjacent gene with an unknown function. Anthropic notes that the set of features they found had previously been observed in only a few systems, including CRISPR, and all of them are programmable and capable of cutting, copying, and inserting DNA. That said, the authors still don't understand the primary function of their discovery, but they believe it's important to share the finding now to show AI's potential in scientific research. Can we trust AI with a biology lab? ❤️ — Yes, it's even necessary 🤔 — Probably not... @hiaimediaen

🤖 GPT-6 Astra Can Drive a Car The creators of DrivingBench connected several frontier language models to a real 2022 Toyota Corolla and gave them control over the steering wheel, accelerator, and brakes. GPT-6 Astra performed the best: on its second attempt, it completed the entire 134.7-meter cone course in 5 minutes and 22 seconds. 👨‍💻 The model received images from the car's cameras along with telemetry data, then issued commands telling the vehicle how far and in which direction to turn the steering wheel, how fast and for how many seconds to drive, or when to stop. On its first attempt, Astra made it through only 49% of the course, but it retained the context from that run and corrected its mistakes on the second attempt, reaching the finish. Claude Fable 5.1 completed no more than 45% of the route. Grok 4.6 and GPT-5.6 Sol failed to make it past even the first turn. The test took place in an empty parking lot, the car moved at roughly a slow walking pace, and a human remained behind the wheel at all times, ready to brake. 💡 The DrivingBench team wanted to test whether general-purpose LLMs could apply their planning and tool-use abilities in the physical world. And it seems language models may indeed have a future as general-purpose AI systems. OpenAI's Head of Applied Research, Boris Power, even called the results "a nail in the coffin" for specialized models trained from scratch for a single narrow task. Would you trust ChatGPT with your car? ❤️ — Yes, I already trust it with my secrets 🔥 — No, absolutely not! @hiaimediaen

💰 Singapore Residents Get Paid to Read Books Singapore has launched a five-year national program called ReadSG, where residents get paid to read books.  To participate, people must read for at least 15 minutes a day and log their reading on the government platform. For each session counted, participants receive 20 virtual coins—worth 2 Singapore cents. It's unlikely one gets rich from reading—only one 15-minute session per day counts. People can't cash out the coins, but they can use them to pay at stores via a QR code. 💡 The government won't check whether people actually read when they register their sessions. The goal is to help Singaporeans develop sustainable reading habits.  "We know that habits are built through small, consistent actions. So, we are starting with a simple goal—just read for 15 minutes a day. On the bus, before bed or over lunch," Melissa Tan from the National Library Board explains. Should you have to pay to read? ❤️ — These days, yes 🔥 — No, people should read for their own pleasure @hiaimediaen

✈️ AI Machine Creates Soda in Any Flavor YouTuber Luke Creatives built an AI soda machine that can mix a drink in almost any flavor, even one that doesn't exist. It has 19 base flavors inside, which create about 524,000 combinations. At first, the author planned to use 20 base flavors—which would have given more than 1 million combinations, but one pump broke. 💡 Gemini became the machine's "brain." The AI listens to a request (for example, "Make me a pine soda," "What does a summer night taste like?" or "Make a drink that tastes like sadness"), breaks it down into parts, comes up with a recipe, and controls the pumps with base ingredients to mix the exact flavor. To choose the base flavors, Luke used Claude, which also helped him build the robot's interface and a dilution table so that all the concentrates had roughly the same strength. ⚙️ The robot's body is the case of an old water filter from eBay, and a Raspberry Pi single-board computer controls the hardware. The whole project took more than eight months. Would you try a drink from a machine like this? ❤️ — Yes, I've wanted one for ages! 🔥 — I wouldn't risk it... @hiaimediaen

💡 To Read the Herculaneum Scrolls, a Researcher Made New Papyrus Scrolls-and Burned Them Former banker and inventor Douglas Seiler came up with a way to read the charred scrolls of Herculaneum without physically unrolling them. To test his idea, he made his own scrolls and burned them. ℹ️ The ancient Roman city of Herculaneum was buried by the eruption of Mount Vesuvius in 79 CE, along with the only known intact library to survive from antiquity. Around 1,800 charred scrolls and fragments remain. Physically unrolling them can reduce the fragile scrolls to dust, while X-rays struggle to distinguish the text from the papyrus because both the carbon-based ink and the charred papyrus are largely made of carbon. But researchers have previously found lead in the ink of some ancient papyri, including fragments from Herculaneum. Lead is much easier to detect with X-rays. Seiler's idea could therefore provide a relatively quick and inexpensive way to identify the scrolls with the best chance of being read. ⚙️ For the experiment, Seiler sourced papyrus and reed pens from Egypt and traditional lampblack ink from Japan. Chemists he knew helped prepare inks with different concentrations of lead. Seiler then paid high school students to write passages from "Star Wars," the Bible, and the 1960s sci-fi series "The Outer Limits" on the papyrus. The scroll was later carbonized in a high-temperature furnace in Seiler's home lab, inside a steel container with very little oxygen. Physicists at NIST scanned the charred scroll using X-ray CT. They then virtually unrolled it using software originally developed to analyze the "jelly rolls" inside lithium-ion batteries. The lead in the ink absorbed up to 25 times as many X-rays as the charred papyrus, making even tiny concentrations of lead stand out. "If you look at the images, the letters light up like a Christmas tree," Seiler said. 📌 The researchers now propose screening real Herculaneum scrolls with an inexpensive handheld X-ray fluorescence scanner. Scrolls showing traces of lead could then be prioritized for more detailed X-ray tomography. The synthetic scrolls can also provide ground-truth training datasets for developing and validating AI-driven text-detection algorithms. 👨 Seiler conceived the project and helped fund it himself, becoming the paper's first author. His path into science was unusual: before retiring, he worked in real estate and banking and as an inventor. He became interested in the Herculaneum scrolls while working with Berkeley SETI on PANOSETI, a telescope project designed in part to search the sky for signs of extraterrestrial technology. @hiaimediaen

👨‍💻 Trump Renames Artificial Intelligence At the UN General Assembly, the U.S. president announced that artificial intelligence will now be called "super intelligence." According to him, the new term better reflects what the technology really is. Trump believes the word "artificial" makes the concept sound "fake." From now on, he promised, U.S. government documents will use the new name: "super intelligence," or SI. That said, the name is unlikely to catch on outside the White House, not least because the term "superintelligence" (ASI) is already taken. It describes a hypothetical AI that surpasses human abilities in every way. On his own platform, Truth Social, Trump also floated "Superior Intelligence" as an option, but "Super Intelligence" got more votes. 🌐 Trump also rejected "any attempt to construct a globalist scheme to control" the technology. "The very same people who said we'll all be dead in 12 years because of global warming... are now saying that AI is going to kill us all, that robots are going to attack us, and that everything is going to be a total disaster," Trump said. What will you call AI now? ❤️ — AI 🔥 — SI @hiaimediaen

💡 A New Wild Cat Species Was Mistaken for an Ordinary House Kitten for an Entire Year For the first time in more than a century, scientists have described a completely new species of wild cat. The tiny Leopardus tilcayo was discovered in Bolivia's Yungas, a region of mountain forests along the slopes of the Andes. An adult weighs just about 1.4-1.8 kg, making it noticeably smaller than most domestic cats. 👨 The story of its discovery began around 2016. A local resident found a small spotted cat on the side of the road and assumed it was an ordinary kitten. He took it home and fed it noodles, rice, and eggs for nearly a year before realizing that the animal was clearly not a domestic cat and turning it over to a wildlife rescue center. Even biologists initially mistook the unusual cat for a known relative, the northern tiger cat. But it was much smaller, and its spot pattern and tail shape were different. Genetic analysis confirmed that it was a distinct species whose evolutionary lineage split from related cats about 1.4 million years ago. ✏️ It is the first completely new species in the cat family to be described since 1923. In recent decades, newly recognized cat species have mostly resulted from scientists reclassifying previously known subspecies and populations. Very little is known about Leopardus tilcayo so far—scientists do not even know how many of these cats remain in the wild. @hiaimediaen

🔍 You Have 60 Seconds to Spot the AI Slop The Lab to AGI team created a quiz to train your intuition in "a world where seeing isn't always believing." The AI images were generated using GPT Image 2.5, so with each round you can practice and better recognize the model's style. After each round, you can view the correct answers, a link to the source for each real photo, and the prompt for every AI-generated image. ⏳ A round lasts 60 seconds, with no more than 10 seconds allotted per image. You earn points for correct answers and lose points for incorrect ones. ➡️ Try the game here. @hiaimediaen

No text

No text

No text