📣 Hello everyone! Our Sunday digest features the most exciting AI news from Week 22, 2025.

▎TO READ

🟠 Elon Musk wants to send people to Mars as early as 2030.

🍄 Is a Last of Us World coming? Scientists predict outbreaks of fungal infections.

👩‍⚕️ People couldn't tell real therapists' answers and ChatGPT's answers apart.

🦜 Australian cockatoos learn to use drinking fountains.

🐳 DeepSeek may have trained the new R1 version on Gemini's responses.

🔎 An AI service can guess where you live from your YouTube comments.

🚔 Amazon is testing humanoid robots for package delivery.

🎉 Students created a vibrating vest that "restored" sight to a blind dog.

🌀 The new AI model lets you "walk" inside generated videos.

💊 AI dramatically speeds up the design of effective drug candidates.

▎SAVE THIS — IT'S HELPFUL

📹 Kling 2.1 is now available on @GPT4Telegrambot: A new version of one of the best video generators.

🟨 CapWords: Create language flashcards with AI.

▎TO WATCH

📱 How is AI changing the film industry? Visual effects specialist Rob Bredow explains.

When will humans land on Mars?

❤️ — Musk will make it by 2030
🔥 — More likely between 2031 and 2035
🤔 — Much later!

♟ Retro Atari 2600 from 1977 Beats Modern ChatGPT in Chess

In a surprising tech showdown, a 1977 Atari 2600—with only 128 bytes of RAM and a 1.19MHz processor—completely dominated OpenAI's ChatGPT in a chess game. Infrastructure architect Robert Caruso set up the match using the Stella emulator, putting the vintage console head-to-head with the modern AI.

These failures stem from ChatGPT's inability to maintain a coherent mental model of the chessboard. Unlike purpose-built chess engines, the AI doesn't truly "understand" the 8×8 grid or visualize piece positions.

Read More

📱 Apple Unveils On-Screen AI and New "Liquid Glass" Design

At WWDC 2025, Apple expanded its Visual Intelligence for iPhone, allowing users to analyze on-screen content in any app. Using the screenshot button shortcut, you can now perform image searches, create calendar events, or send information directly to ChatGPT.

Apple also revealed "Liquid Glass," a major design overhaul for all its operating systems. This new look features translucent, rounded elements that dynamically react to light and surrounding content, creating a seamless harmony between the software and hardware.

Read More

🔥 Turkmenistan's Gates of Hell Crater Is Running Out of Gas

Turkmenistan's "Gates of Hell," a 70-meter-wide gas crater that has burned for over 50 years, is finally diminishing.

The Darvaza crater formed in 1971 when Soviet geologists drilling for natural gas accidentally punctured an underground cavern, causing a ground collapse that created a 70-meter-wide, 30-meter-deep crater. Scientists ignited the methane to prevent toxic gas escape, expecting it to burn out within weeks.

Instead, the crater has burned continuously for five decades, fueled by vast methane reserves from the Amu-Darya Basin below.

Read More

🇫🇷 French Startup Mistral Releases Its First Reasoning Model

The new Magistral model comes in two versions: Small, which has 24B parameters with open weights, and Medium, a larger model available for deployment to enterprise clients.

➡️ Magistral Medium performs comparably to the January release of DeepSeek-R1 in benchmarks.
➡️ The AI thinks "in all languages and alphabets of the world."
➡️ The model works really fast—Mistral claims a 10x advantage over competitors.

➡️ Magistral Medium is available for free in Le Chat.

Read More

Ⓜ️ Meta's AI App Privacy Nightmare

Meta's standalone AI app has become a privacy nightmare, with users unintentionally publishing private conversations with the chatbot — including sensitive personal information — for the public to see. At the same time, concerns are growing over the lack of end-to-end encryption for AI interactions across platforms like WhatsApp, where conversations can be used for AI training under regionally inconsistent privacy protections.

At the heart of this privacy debacle is Meta AI's confusing interface design, which misleads users into publicly sharing what they believe are private exchanges. Many assume the "Share" button saves chats privately, when in fact, it posts them to the public "Discover" feed — making those conversations visible to anyone using the app.

Read More

📣 Hello everyone! Our Sunday digest features the most exciting AI news from Week 24, 2025.

▎MAIN NEWS

👐 OpenAI released o3-pro. The new AI model outperforms o3 and o1-pro in every domain. Also, we've decreased the price of basic o3 in our @GPT4Telegrambot.

▎NEW RELEASES

📱 New iOS 26 will feature real-time conversation translation and a Google Lens analog.

🇫🇷 French startup Mistral released its first reasoning model, Magistral.

Ⓜ️ Meta's AI app update is a privacy nightmare.

▎TO READ

🧬 The Embryo offers to choose the IQ and appearance of future children through DNA analysis.

🪩 Retinal nanoprosthesis restores sight and enables night vision.

🇨🇳 China shuts down AI photo recognition during exams.

🏙 Abu Dhabi will become a fully AI-managed city.

🗺 AIs have played "Diplomacy": o3 is the winner, Claude Opus 4 is the most peaceful.

🔥 Turkmenistan's Gates of Hell crater is running out of gas.

💵 Blake Anderson: How to build a $15M AI business at 23.

♟ Retro Atari 2600 from 1977 beats ChatGPT in chess.

🎲 Nearly one in four children already uses AI.

Have a great Sunday 👍

🧊 Antarctic Detector Spots Radio Signals That Shouldn't Exist

A balloon-borne particle detector over Antarctica has detected puzzling radio signals that seem to come from deep within the ice—something current physics can't explain.

According to a study in Physical Review Letters, the Antarctic Impulsive Transient Antenna (ANITA) picked up these signals at angles up to 30 degrees below the surface.

The unexpected results challenge fundamental physics, prompting researchers to explore whether unknown particles or interactions could be responsible.

Read More

🧠 China Launches First Brain Implant Trial

China has launched its first brain-computer interface clinical trial, becoming the second country after the U.S. to test this technology on humans. A paralyzed patient successfully played chess and racing games using only his thoughts after receiving the implant in March 2025, showing remarkable progress within three weeks.

The Chinese BCI developer CEBSIT directly competes with Neuralink. The CEBSIT chip is significantly smaller and thinner than Musk's company device.

Both countries race to commercialize their technologies, with China targeting 2028 market entry.

Read More

💰 Altman Says Meta's $100M Offers Fail to Lure OpenAI Staff

OpenAI CEO Sam Altman revealed that Meta has been offering massive compensation packages worth over $100 million to OpenAI employees as part of Mark Zuckerberg's aggressive hiring campaign for Meta's new superintelligence team, but claimed these efforts have been largely unsuccessful with none of OpenAI's "best people" accepting the offers.

Mark Zuckerberg is personally leading Meta's new push into superintelligent AI, launching a "superintelligence lab" to develop AGI and surpass human-level AI. He calls it Meta's defining project of the next decade, with $14 billion in backing.

Read More

🖤 SpaceX Starship Explodes in Fireball During Ground Test

Elon Musk's SpaceX Starship exploded during a ground test at the Starbase launch site.

The blast was caught on a live stream. Residents of Cameron County, Texas—where the site is located—reported that the explosion shook windows and rattled dishes in their homes.

SpaceX says it's investigating the cause of a "major anomaly." All personnel are safe.

Read More

💰 Wix Buys 6-Month-Old AI Startup Base44 for $80M

According to TechCrunch, Israeli developer Maor Shlomo has sold his six-month-old AI-powered "vibe-coding" startup Base44 to Wix for $80 million in cash—highlighting how AI is accelerating small-team innovation in the no-code space.

At just 31, Shlomo was already a seasoned founder when he launched Base44. In 2017, at age 24, he co-founded Explorium, a predictive analytics startup that raised around $125 million and grew to over 100 employees.

Read More

📣 Hello everyone! Our Sunday digest features the most exciting AI news from Week 25, 2025.

▎NEW RELEASES

✴️ Midjourney released a V1 video generator with flexible settings and impressive aesthetics.

▎SAVE THIS — IT'S HELPFUL

📘 CS50x: A free online computer science course from Harvard.

▎TO READ

📹 Dream Recorder: An AI gadget for turning dreams into videos.

🧠 China has launched its first brain-computer interface clinical trial.

💬 Why AI doesn't really understand our language?

🔮 Researchers have assessed the risks of AI completely destroying humanity.

💰 Meta offered $100 million to poach top AI talent from OpenAI.

💥 SpaceX Starship explodes in a fireball during a ground test.

🌌 Scientists have created an AI model to study black holes.

🌧 Rainmaker startup creates rain in the right place at the right time.

▎TO WATCH

🎞 ANCESTRA: The first film by Darren Aronofsky's studio, created with Google Veo.

▎FEEDBACK

What kind of posts would you like to see more?

⚡️ — new AI models and service reviews
🤓 — lectures, courses, manuals
🔥 — scientific breakthroughs and cases
👀 — I'll write in the comments!

🔎 Google Launches Open-Source Coding Agent

Google has released Gemini CLI, a command-line AI coding agent. The tool allows developers to make natural language requests for coding tasks, including explaining code sections, writing new features, debugging issues, and running commands.

➡️ Powered by Gemini 2.5 Pro with a 1M token context window.
➡️ Supports MCP, web search, and VS Code integration.
➡️ Fully open-source.

Gemini CLI is free to use, with a limit of 60 requests per minute or 1,000 per day.

Read More