📣 Hello everyone! Our Sunday digest features the most exciting AI news from Week 9, 2025.

▎MAIN RELEASES

👐 OpenAI has introduced GPT-4.5. The model is now more "human-like," surpassing GPT-4o in many ways. Still, it lags behind "reasoning" models in coding and exact sciences.

🖥 Anthropic has released Claude 3.7 Sonnet, the first AI capable of both quick responses and reasoning—the best coding model for now.

🔴 Claude 3.7 Sonnet is already available on @GPT4Telegrambot

▎TO READ

💲 Mercor: how did three 21-year-old friends build a $2B AI startup?

🏙 Toyota is building a "City of the Future" managed by AI.

🟢 Nvidia has launched "Duolingo for sign language."

👩‍💻 Amazon has unveiled Ocelot, the first quantum computing chip using "cat qubits."

📸 Yope: a new viral photo-sharing app backed by investors.

🦠 Mal-ID: A universal AI blood testing for HIV, COVID-19, diabetes, and other diseases.

🟠 China's rover has found evidence of ancient beaches on Mars.

▎TO WATCH

📱 "Cassandra": new six-episode series about a "smart" home that wants to destroy its owners.

📱 Top 3 New TED Talks on AI: mind-reading, agents for science, and web-searching.

📺 Five of the best short films from the Project 0dyssey AI film festival.

Enjoy your Sunday 👍

🖼️ Apple's ChatGPT-Style Siri Delayed Until 2027

Apple is working on a new version of Siri that can talk with users like ChatGPT, but it's taking longer than planned. According to Bloomberg, it probably won't be ready before 2027 because of technical problems and issues inside the company.

The delay stems from significant technical hurdles in merging two separate systems. One does simple commands, and the other gives smart AI answers. Keeping them apart causes trouble, so Apple needs more time to make them work together.

The project, internally known as "LLM Siri," users will have to wait until iOS 20 or later.

Read More

🚀 First Private Mission to Venus

Rocket Lab is teaming up with NASA and MIT to launch the first private mission to Venus in 2026 or later. The mission aims to check the planet's atmosphere for signs of tiny living things. They're bringing some slick tech, like NASA's HEEET heat shield and Rocket Lab's Photon spacecraft.

This mission, part of the broader Morning Star Venus exploration program, represents a significant step in commercial space ventures and could pave the way for more frequent, focused planetary exploration missions.

Read More

🦁 Google's Wildlife AI Model

Google has open-sourced SpeciesNet, an AI model that recognizes animal species in camera trap photos with 94.5% accuracy, TechCrunch reports. Trained on over 65 million images, it can classify wildlife into over 2,000 categories, helping speed up biodiversity research and conservation.

Google's Wildlife Insights platform uses AI and cloud computing to revolutionize wildlife monitoring and anti-poaching efforts. Automating species identification can process up to 3.6 million camera trap images per hour, drastically reducing the time needed for manual analysis.

Read More

🤩 Scientists Engineer "Mammoth-like" Woolly Mouse

Scientists at Colossal Biosciences have created a "woolly mouse" with features similar to a mammoth. The mouse has long, wavy hair and accelerated fat metabolism. According to Time, this genetic engineering success is an important step in the company's effort to bring back the extinct woolly mammoth.

Supporters say this result shows that the company's method can bring back extinct animals. However, critics argue that changing a living species is not the same as reviving a woolly mammoth, and they worry about possible problems in nature and ethical concerns when reintroducing mammoth-like animals.

Read More

👐 OpenAI Prepares $20,000 AI Agent

OpenAI is about to launch a new line of AI agents with monthly subscription plans ranging from $2,000 to $20,000.

These agents use powerful language models and smart decision-making to solve problems independently.

Low-end agents will cost $2,000 per month and target high-income knowledge workers, while high-end agents will cost $20,000 per month and operate like Ph. D.-level research assistants. OpenAI expects these subscriptions to eventually account for 20% to 25% of its overall revenue.

Read More

🌐 Google Tests AI-Enhanced Search

Google has started testing AI Mode, an AI-powered search. Powered by Gemini 2.0, it generates answers with source links, like Perplexity or ChatGPT Search, making it ideal for complex queries that would typically require multiple searches. It is still in experimental mode and is only for Google One AI Premium subscribers in the U.S.

The introduction of AI-enhanced search features could fundamentally alter user behavior and expectations. By offering more comprehensive, AI-generated responses, Google may reduce the need for users to click through to original web pages, potentially impacting websites that rely on search traffic.

Read More

🐳 + 🖱 DeepSeek with Perplexity web search on @GPT4Telegrambot

Hi everyone! We've integrated the most sensational AI model of the last weeks, DeepSeek-R1, with the best AI-powered web search Perplexity.

Key Features:
⚫️ Accurate answers to complex questions
⚫️ Real-time knowledge
⚫️ Shows the chain of reasoning
⚫️ Gives links to sources

The tool is now FREE for bot users.

How to Use:
➡️ Go to @GPT4Telegrambot
1️⃣ Send the /s command followed by your query (for ex. "/s when was the Internet invented?")
2️⃣ Select DeepSeek Search

The answer will be ready in 1-2 minutes.

🔴 @GPT4Telegrambot — the #1 bot for using AI on Telegram. It can write texts and code, translate languages, solve math and physics problems, work with documents, and create images, videos, and music. 20 million users.

Check the bot's other updates:

🐰 Pika 2.2 Video Generator

♊️ Gemini 2.0 Flash Thinking, the reasoning model from Google

⚛️ Researchers Transform Light into a Supersolid

For the first time, researchers successfully turned laser light into a supersolid, Nature reports. They used a laser and a carefully designed piece of gallium arsenide (GaAs), a material with tiny ridges on its surface. When the laser light interacted with this material, it produced quantum particles called polaritons. The ridges on the gallium arsenide guided the polaritons to organize into a specific pattern, forming a supersolid structure.

This discovery is valuable because it allows scientists to explore supersolids and other unique quantum states in new ways. Supersolids made from light may have advantages over those made from atoms.

Read More

📱 Brazil Orders Third-Party iOS Apps

Brazilian federal court has ordered Apple to allow app sideloading on iOS devices within 90 days, challenging the company's longstanding App Store policies and potentially setting a precedent for other countries.

This decision reinstates a previously overturned injunction, forcing Apple to comply with Brazil's antitrust legislation. The court order comes with significant financial penalties for non-compliance:

➡️ Apple faces daily fines exceeding $40,000 if it fails to adhere to the ruling;

➡️ The company has been given 20 days to initially comply with the antitrust requirements;

➡️ The 90-day deadline aims to give Apple sufficient time to implement the necessary changes to its iOS ecosystem in Brazil.

Read More

📣 Hello everyone! Our Sunday digest features the most exciting AI news from Week 10, 2025.

▎NEW RELEASES

🐰 The Pika 2.2 video generator creates clips up to 10 seconds in FullHD (1080p) and cinematically transforms one frame into another via the Pikaframes feature.

🐳 + 🖱 DeepSeek with Perplexity web search is now available on @GPT4Telegrambot.

▎SAVE THIS — IT'S HELPFUL

🎤 Free audio and video transcription with ElevenLabs Scribe.

▎TO READ

🧠 Cortical Labs has released CL1: The first biocomputer using live human neurons.

🐀 Colossal Biosciences has created a "woolly mouse": A first step to bringing back the mammoth.

✍️How does AI help to unlock history mysteries?

🛰 DARPA's new project: Why is the U.S. planning to build "living" space stations?

🍬 e-Taste: A device that enables the remote transmission of taste sensations.

🚀 Rocket Lab is partnering with NASA and MIT to launch the first private mission to Venus in 2026 or later.

🌐 Google has started testing an AI search powered by Gemini 2.0.

👐 Which startups is the OpenAI Startup Fund investing in while anticipating the AGI future?

🤖 OpenAI may launch AI agents with subscriptions up to $20,000 per month.

📱 Spotify and ElevenLabs make audiobook creation faster and more accessible.

▎TO WATCH

📱 New lecture by OpenAI co-founder Andrej Karpathy: How to Use LLMs Effectively?

Enjoy your Sunday 👍

➕ Student Cracks Century-Old Math Problem

Divya Tyagi, a graduate student at Penn State University, has advanced a century-old mathematical problem in wind turbine aerodynamics, originally formulated by British aerodynamicist Hermann Glauert, with the potential to revolutionize wind energy science and enhance turbine design efficiency.

Tyagi's refined mathematical model offers a more accurate representation of turbine dynamics, paving the way for substantial improvements in wind turbine performance and efficiency.

Read More

🌌 Colossal Black Hole Discovery

Astronomers have found an enormous black hole 36 billion times heavier than our Sun. This black hole might be one of the largest ever discovered, making scientists rethink what they know about space. As reported by Live Science, this giant black hole is located in a group of galaxies called the Cosmic Horseshoe.

Scientists believe that dark matter, a mysterious substance in space, could have been important in creating and growing these supermassive black holes, especially when the universe was young.

Read More

👐 GPT-4.5 on @GPT4Telegrambot

Hi everyone! We've integrated OpenAI's newest and largest AI model, GPT-4.5, into our bot.

Why Should You Try It?

➡️ Improved Conversational Skills: The model speaks more naturally, better understands context, and captures nuances in dialogues.

➡️ Emotional Intelligence: GPT-4.5 better understands emotional tones and can respond more emphatically.

➡️ Expanded Knowledge Base: GPT-4.5 has a broader and deeper understanding of the world and makes significantly fewer mistakes when answering factual questions.

➡️ Multilingual Proficiency: The model demonstrates exceptional performance across 14 languages, including Arabic, Chinese, French, German, Hindi, and Spanish.

How to Use?
1️⃣ Go to @GPT4Telegrambot
2️⃣ Purchase the "ChatGPT + DALL•E 3" bundle in the /premium section.
3️⃣ Select the GPT-4.5 in the /model section.

❓ Which Model Should You Use?

→ GPT-4.5 is the most "human-like" model, ideal for creative tasks and engaging conversations.

→ For math and exact sciences: OpenAI o3-mini.

→ For coding: Claude 3.7 Sonnet.

→ For web search: Perplexity + DeepSeek. Start your query with the /s command.

→ For everyday tasks: GPT-4o and Gemini 2.0 Flash Thinking.

💲 GPT-4.5 is OpenAI's most expensive model (four times pricier than o1). We've decided to lower its actual cost in our bot so you can try this new model and compare it to others. Until April 30th, GPT-4.5 requests will cost the same as o1.

🔴 @GPT4Telegrambot — the #1 Telegram bot for AI-powered assistance. It can write texts and code, translate languages, solve math and physics problems, work with documents, and create images, videos, and music. Trusted by 20 million users worldwide.

🍌 Gene-Edited Non-Browning Banana Developed

Tropic, a biotech company from the UK, has created a banana that does not turn brown by using CRISPR gene-editing technology. This method stops the enzyme that causes bananas to brown, helping the fruit stay fresh longer and reducing food waste. The new banana has already received approval in several countries. It tastes and feels just like regular bananas. Still, it offers environmental benefits and allows for more improvements in banana farming.

Bananas that do not turn brown could significantly reduce food waste. This is important because around 33% of harvested fruits and vegetables never reach consumers due to their short shelf life. Bananas are especially affected by this problem. More than 60% of bananas meant for export become waste before people can eat them. About 5 billion bananas are thrown away each year in the U.S. alone.

Read More

🌐 Google Unveils Gemma 3 AI Models

Google has introduced Gemma 3, a new group of AI models designed to run efficiently on a single GPU or TPU. Google says these models perform better than larger competitors while also offering strong multilingual and multimodal abilities ⤴️

Gemma 3 provides impressive performance in a small package. It is available in four sizes, from 1 billion to 27 billion parameters. Depending on their specific needs and hardware limits, developers can choose the best model size. Gemma 3 supports a large context window of 128k tokens. It can process about 30 high-quality images, a 300-page book, or more than an hour of video content, supports 140 languages.

Read More

Ⓜ️ Meta AI Decodes Thoughts into Text

Meta's research team, working with the Basque Center on Cognition, Brain and Language, has created an AI model that can turn brain signals into text. This model uses advanced brain scanning and artificial intelligence to achieve up to 80% accuracy at the character level.

The study shows that the brain doesn't process words one by one. Instead, it holds a complex mix of information at different levels while forming language. This discovery could improve AI language models and improve brain-computer interfaces for people needing communication support.

Read More

👐 OpenAI Suggests Changes to AI Copyright Rules

OpenAI says it would be "impossible" to create advanced AI models without using copyright-protected materials. This opinion has led to legal challenges and discussions about fair use as copyright laws try to keep up with the fast growth of artificial intelligence. OpenAI compares its situation to the Google Books case, where scanning books for online searches was considered acceptable because it created something new and useful.

OpenAI faces several lawsuits because it used copyrighted articles, books, and art to train its AI tools, like ChatGPT. The results of these lawsuits could significantly affect the AI industry. They may change how companies can access and use copyrighted materials when training AI models.

Read More

📱 Apple Prepares AirPods Live Translation Feature

According to Bloomberg, Apple is working on a live translation feature for AirPods. This feature will likely be available in late 2025, as part of the iOS 19 update.

This feature will let users translate conversations instantly without typing or entering words by hand. For example, if someone speaks Spanish, their words will be translated into English and played directly into the English speaker's AirPods. When the English speaker replies, their iPhone will play the Spanish translation through its speakers.

Apple is not the first company to offer this kind of feature. Google's Pixel Buds have had something similar since 2020, and Samsung recently added live translation to its Galaxy Buds.

Read More

📣 Hello everyone! Our Sunday digest features the most exciting AI news from Week 11, 2025.

▎NEW RELEASES

💻 OpenAI's GPT-4.5 is now available on @GPT4Telegrambot: The model speaks more naturally, better understands context, and captures nuances in dialogues.

🌐 Gemma 3: The next generation of Google's small language models.

▎TO READ

🇪🇪 Estonia is the first country in the world to introduce ChatGPT into the education system.

📈 Which professions will disappear by 2030? A forecast by the World Economic Forum.

🇨🇳 Manus: Why is everyone talking about the AI agent from China?

🍌 A biotech company has created a gene-edited banana that doesn’t turn brown and stays fresh longer.

🥇 Turing Award goes to the creators of reinforcement learning.

📱 Apple is preparing a live translation feature for AirPods.

➕ A student cracks a century-old math problem in wind turbine aerodynamics.

🦾 A paralyzed man learned to mind-control a robotic hand.

🤖 How do humanoid robots revolutionize manufacturing?

▎TO PLAY

🐙 FathomVerse: A mobile game for exploring the world's ocean life and helping scientists train AI.

▎TO WATCH

📺 "2073": A film about the future after an AI-catastrophe.

📺 Yann LeCun, Chief AI Scientist at Meta: Can AI have emotions?

Enjoy your Sunday 👍

🗺 AI GPS Alternative Navigates without Satellite

Tern AI has introduced the Independently Derived Positioning System (IDPS), which removes the need for satellite signals by using data from vehicle sensors and map information.

IDPS provides several important benefits compared to traditional GPS technology. Because it does not rely on external signals, it can offer reliable navigation even in difficult places such as parking garages, tunnels, and mountain areas where GPS often loses its signal. The system works in a closed loop, improving privacy by not sending out any data externally.

In addition, Tern AI is actively working to combine IDPS with popular map apps like Google Maps, Apple Maps, and Waze, hoping to eventually replace GPS.

Read More