AI revolution! Last week's hottest AI news was revealed

#夏日生活打卡季 #

While there were no major announcements like GPT-5 or Mid-Journey version 6 in last week's AI news, there were many small advances in the AI space. Meta has released an open-source AI music generation model that anyone can use on Hugging Face. This model sounds better than other models like Google's Riffusion or Music LM. It's available on GitHub, and if you have a graphics card with at least 16GB of RAM, you can install it locally.

Adobe has made updates to Adobe Express, including the addition of Adobe Firefly Generative AI, which allows users to use AI cue images and add text effects directly within Adobe Express. They also added generative AI tools to Adobe Illustrator, including Generative Recolor, which allows users to upload black and white images and add color, or upload designs and regenerate them with a variety of new color variations. These updates are now available to Adobe Creative Cloud users.

AI companies like OpenAI, DeepMind, and Google's AI division have agreed to open up their models to the UK government for research and security purposes. However, they have not yet made the source code available to the public. Some have noticed that AI features are being integrated into Google's products, such as the "write for me" button in Gmail and Google Docs. The article suggests that AI is changing the world and can be used to write articles and emails.

Google has announced the release of a new AI model that can take a small portion of an image and then use its training data to fill the rest of the image. This model is based on Yann LeCun's vision of more human-like AI and is being open-sourced. This model is more like the human way of thinking, it can identify a small part of an image and then use its knowledge of the subject matter to fill in the blanks. This is different from other AI-generated models that attempt to recreate images at the pixel level.

AMD has developed an AI model trained like a human that can fill the rest of the context based on its dataset and what it knows about the small pieces of information it sees. The model is available on GitHub, and AMD has partnered with Hugging Face to provide computing power for their machine learning products. That's a big deal because AMD, not Nvidia, provides computing power. AMD announced that they are building hardware specifically tailored for AI, with the goal of optimizing the combination of hardware and software. They are also trying to compete with Nvidia by creating chips that are more powerful than Nvidia. AMD's new chip, the Mi 300x, is designed for generative AI and offers 2.4 times the memory and 1.6 times the bandwidth of the competition. The article discusses the competition between Nvidia and AMD in creating better chips for large language models.

OpenAI's announcement that they made updates to the GPT language model to include a new 16,000-context version, which is four times larger than the previous version. This new version allows chat GPT to have about 12,000 words between input and output. Developers using the GPT 3.5 API will see cost reductions, but it's unclear whether the company will pass on those savings to end consumers.

Microsoft and OpenAI have an "amazing partnership," but tensions arise as they work together and compete on AI features. OpenAI reportedly warned Microsoft to slow down when integrating GPT 4 into the Bing search engine to avoid inaccurate and unpredictable reactions, but Microsoft moved on. People are concerned about why buy chat GPT if they can use Bing with GPT built into it for free. Microsoft and OpenAI have a unique partnership, which has led to some conflict behind the scenes while the two companies support and compete with each other. Both companies have chat models, GPT+ and Bing Chat, both with GPT built in.

Sources say Sir Paul McCartney said that artificial intelligence has enabled the last Beatles song to be completed. Google also announced a new generative AI model for virtual try-ons, allowing users to generate try-on images using AI. The model was created using Google's shopping graph and diffusion model. Google Shopping has launched a virtual try-on feature for women's tops from brands like Anthropologie, Loft, H&M and Everlane. This feature will become more precise over time and expand to more brands.

Google Shopping has partnered with Shopify, a commerce platform that allows anyone to set up their own store to sell physical products. Shopify simplifies the complexities of running a business, making it possible for anyone, anywhere to become an entrepreneur. The article discusses the ease of setting up an online store with Shopify, which is democratizing technology for entrepreneurs and helping to build tomorrow's economy.

A new trend to create amazing images using control nets and steady diffusion with QR codes (QR codes are a type of two-dimensional barcode) has recently caused a craze on Twitter. The article provides links to tutorials on how to set this trend. There's a new trend on Twitter to create 3D models that can be downloaded from Hugging Face and made at home. Rowan Chung made a tutorial with 1.3 million views that teaches you how to make these models. The images produced by this trend look cool, but generating them is difficult. Some QR codes work, but others don't.

A new technology called "re-render" that improves video quality and solves problems with flicker and artifacts. The text discusses a new video model that doesn't have the flicker effect seen in other models. This model is demonstrated in a video and compared to other models. The new model is said to be clear and has no coherence issues. The authors are excited about the model's potential use in various tools and platforms. Whether this model can be used by the public is uncertain.

The text discusses recent Mid Journey office hours calls and what to expect from the upcoming 5.2 release. This version is expected to be released any day and will bring limited Discord compatibility and new features that change the zoom and aspect ratio. The cue analyzer is also discussed, currently used by entering a "description" and uploading an image. Version 6 was originally expected to be ready in 6-8 weeks, but is now expected to be released in July. The article discusses developing a web and mobile standalone version to reduce word hints in Discord. The team is working to improve moderation and reduce rejected prompts.

In addition, 11 Labs introduced a speech classifier tool that can determine whether an audio is generated by AI. The goal of this tool is to combat deepfakes, which are becoming increasingly common in the AI world. The authors are excited to announce a new authentication tool that uses AI speech classification to identify whether an audio sample contains AI-generated audio. They discussed the importance of taking a proactive stance against the malicious use of AI, and the impact of deepfakes and disinformation.