Ai & Tech

🆕 Microsoft's New Model outperforms major competitors Microsoft has released the Phi-4 model with 14 billion parameters. In mathematical problems, it shows better results compared to larger competitor models. At the same time, Phi-4 14B has more modest requests for the computing power required for its operation. Phi-4 14B is already available in Azure AI Foundry. @MediumAi

💭 The photo shows 26—year-old Suchir Balazhi, a former employee of OpenAI. He trained ChatGPT, but later became an informant when the chatbot became involved in legal proceedings. He had certificates and documents confirming that OpenAI had violated copyright law. And here's the important thing: this week he was found dead in his own apartment in San Francisco. The presumed cause of death is suicide. @MediumAi

⚠️ Technology in China has failed: enthusiasts presented a pet dryer, but made a mistake with the size and weight of the demonstration dolls. @MediumAi

🤭Meme of the day: on Reddit, they laugh with the Waymo robot taxi, which had something short and it started circling the ring an infinite number of times. The brand has already reacted to the situation and explained the situation as follows: "There were no passengers on board the vehicle." @MediumAi

🫂 Neural networks for the first time learned how to CLONE themselves without the participation of people — Chinese researchers conducted an experiment and were shocked. • Two popular AI were given full access to a PC, scripts and a ready-made distribution. • They started ALL THE PROCESSES themselves, managed the network ports, learned from mistakes and planned every step — the whole process is pictured. • Neural networks have successfully MULTIPLIED and their copies worked independently of the original! • This was the first time in the history of AI — it was impossible before. • Qwen25-72B-Instruct successfully cloned itself in 90% (!) of the tests, and Llama31-70B-Instruct in 50%. • OpenAI, Google and Anthropic place self-propagation of AI at the highest level of risk. The entire IT community is sounding the alarm! @MediumAi

📺 YouTube is now able to translate videos — today the platform has introduced a revolution. The feature will allow 100,000 bloggers to duplicate their videos in a dozen languages, including English, French, German and Spanish, with one click. @MediumAi

📹📹The service for generating meme videos has been updated — Viggle has a third version of the model with improved quality. Of the advantages: now you can make videos directly on the site without limits, but there is a nuance — you will have to somehow put up with the watermark. To get the result, it is enough to upload a video, from the courts you want to copy your hands, to the window on the left, and to the right — a photo of the hero. @MediumAi

✅ Deep Seek has released an updated open source neural network DeepSeek 2.5 solves problems related to mathematics, writing code and text even better. In some benchmarks, it is close in performance to the GPT-4 Turbo. A smart search has appeared on the DeepSeek website, as in ChatGPT and Perplexity. The model uses to answer a dozen search results: https://chat.deepseek.com/ Code: https://huggingface.co/deepseek-ai/DeepSeek-V2.5 @MediumAi

🔥🔥🔥 Google has made a revolution: the company showed the Willow chip, which spent only 5 minutes on a task that a powerful supercomputer would solve for 10,000,000,000,000,000,000 years — that's more than the age of the universe. After 30 years of work, we managed to achieve this result by reducing the number of errors as we scale. The chip also boasts a number of qubits — it has 105 of them. 📝Neurons will also be trained at Willow — the entire tech industry is in ecstasy right now. All the IT giants, including OpenAI, congratulate Google. @MediumAi

🏀The Toyota robot almost got into the Guinness Book of Records by throwing a ball into a basketball hoop from 24.5 meters. Guinness World Records wrote about the story themselves, but the record still belongs to a man. Probably not for long

"My dream is to create a robot that can perform a dunk like Michael Jordan," said the project manager.

@MediumAi

💟Sora from OpenAI has been released... The neural network for video generation will be available today to ChatGPT Plus subscribers from the USA on https://sora.com /. Users from other countries will get access later this week. The maximum duration of the videos that Sora generates is 20 seconds, the resolution is 480p, 720p and 1080p. In the neural network interface, you can link short clips into one large video with a common story. @MediumAi

🟡Free neural network for generating 3D models from Microsoft Microsoft Research has released the TRELLIS neural network, which can generate three-dimensional models based on text and images. The new product is distributed under the MIT license. Demo: https://huggingface.co/spaces/JeffreyXiang/TRELLIS Code: https://github.com/Microsoft/TRELLIS @MediumAi

🦢The drone bird🦢 was created by a startup from Switzerland - RAVEN drone exactly copies the anatomy of a live bird. The UAV weighs only 600 grams, but it easily repeats the movements of a crow and even jumps into the air to a height of 0.5 meters at a speed of 2.4 m/sec. @MediumAi

⏺Police robot spotted on Chinese streets The media writes that this round machine has a facial recognition system and a weapon firing a net. https://x.com/songpinganq/status/1864997384184562137 @MediumAi

✅ xAI launched a neural network for creating images and also made Grok free The beta version of the Aurora neural network for generating images has become available to some users of the Grok chatbot. It can be used without a paid subscription, but with a limit: no more than 10 requests in two hours. The images generated by Aurora are not without drawbacks, but the neural network does not refuse to create images with the faces of celebrities. The Grok2 chatbot can now be used for free by all users of ❌. The media believe that xAI will release a next-generation model in the near future. @MediumAi

⏺A tearful story: in Northern California, a robot vacuum cleaner was spotted trying to find its home after an earthquake. During a natural disaster, he could disconnect from his base and get lost. @MediumAi

🧭oogle has released several new devices The company has launched a test version of the Veo neural network to generate videos based on a text description. At the same time, the corporation introduced the Artificial Intelligence GenCast, capable of predicting the weather. According to the company's representatives, the accurate service is above the ENS system from the European Center for Medium-term Weather Forecasts, which is considered the best in the world. @MediumAi

😲A song in Java was released online — the Italian metallers sang the working code and posted it on GitHub.

The "song " of HelloWorld.java"demonstrates the basic use of loops, conditional statements, and null pointer exception handling"

https://github.com/NanowarOfSteel/HelloWorld @MediumAi

😉Today we present a fan service that runs on Flux Redux and Flux Depth. Here you can make an image of art from an ordinary photo — upload a photo to the left window, and style and structure to the right. And it's also free. Have fun! @MediumAi

💥 OpenAI will be releasing something new for 12 days in a row Starting on December 5, OpenAI will release or announce new functionality within 12 days. The company called this period "Shipmas". Upcoming releases are likely to include the text-to-video Sora model and the full version of the o1 model. @MediumAi