The Model Why does biological evolution work? And, for that matter, why does machine learning work? Both are examples of adaptive processes that surprise us with what they manage to achieve. So what’s the essence of what’s going on? I’m going to concentrate here on biological evolution, though much of what I’ll discuss is also […]

Image: Nvidia Nvidia is updating its experimental ChatRTX chatbot with more AI models for RTX GPU owners. The chatbot, which runs locally on a Windows PC, can already use Mistral or Llama 2 to query personal documents that you feed into it, but now the list of supported AI models is growing to include Google’s Gemma, ChatGLM3, and even OpenAI’s CLIP model to make it easier to search your photos. Nvidia first introduced ChatRTX as “Chat with RTX” in February as a demo app, and you’ll need an RTX 30- or 40-series GPU with 8GB of VRAM or more to be able to run it. The app essentially creates a local chatbot server that you can access from a browser and feed your local documents and even YouTube videos to get a powerful search tool complete with summaries... Continue reading…

Illustration: The Verge Apple has been quiet about its plans for generative AI, but with the release of new AI models today, it appears the company’s immediate ambitions lie firmly in the “make AI run locally on Apple devices” realm. Researchers from Apple released OpenELM, a series of four very small language models on the Hugging Face model library, on Wednesday. Apple said on its Hugging Face model page that OpenELM, which stands for “Open-source Efficient Language Models,” performs very efficiently on text-related tasks like email writing. The models are open source and ready for developers to use. OpenELM is even smaller than most lightweight AI models It has four sizes: 270 million parameters; 450 million parameters; 1.1 billion parameters; and 3 billion... Continue reading…

在汽车行业难寻出路的特斯拉，似乎只剩下了 AI 这一条路。#欢迎关注爱范儿官方微信公众号：爱范儿（微信号：ifanr），更多精彩内容第一时间为您奉上。 爱范儿 | 原文链接 · 查看评论 · 新浪微博

Tesla On the cusp of what’s expected to be a very not-good earnings report, Tesla decided to shuffle its search results a bit with the reveal of the 2024 Model 3 Performance, with a specially tuned chassis and a 2.9 second 0–60 mph acceleration time. Interestingly, Tesla did not give this updated Model 3 the “Plaid” performance badge, which, so far, it has reserved for its more premium vehicles, the Model S and Model X. For the Model 3, it’s sticking to Performance. The performance trim of the company’s top-selling electric sedan will get 510 horsepower and 741 Nm of torque — a step above the 470 horsepower that was achievable in the 2019 performance trim — and can hit a top speed of 163 mph. Interestingly, Tesla is not give this updated... Continue reading…

Unlike other Firefly-powered tools that generate sections of an image, Photoshop’s new Generate Image tool will fill the entire canvas. | Image: Adobe Adobe is adding some new generative AI tools to its Photoshop creative software that aim to give users additional ways to control the designs they generate. Powered by Adobe’s new Firefly Image 3 foundation model, these new tools are available today via the Photoshop beta desktop app, and will be generally available “later this year” according to Adobe’s Press release. The most notable tool is Reference Image, which uses user-uploaded images to inspire the output generated by Adobe’s AI, matching similar elements in style and color. For example, instead of repeatedly tweaking a prompt description like “a blue vintage truck with flower decals,” users can instead provide a reference image that Photoshop will use as a guide. “Prompting is a... Continue reading…

Illustration: The Verge Microsoft launched the next version of its lightweight AI model Phi-3 Mini, the first of three small models the company plans to release. Phi-3 Mini measures 3.8 billion parameters and is trained on a data set that is smaller relative to large language models like GPT-4. It is now available on Azure, Hugging Face, and Ollama. Microsoft plans to release Phi-3 Small (7B parameters) and Phi-3 Medium (14B parameters). Parameters refer to how many complex instructions a model can understand. The company released Phi-2 in December, which performed just as well as bigger models like Llama 2. Microsoft says Phi-3 performs better than the previous version and can provide responses close to how a model 10 times bigger than it can. Eric Boyd,... Continue reading…

Introduction In this blog post we dive into inference with DBRX, the open state-of-the-art large language model (LLM) created by Databricks (see Introducing...

Innovative business models in US healthcare are creating substantial value, but much opportunity remains, even among the fastest-growing organizations.

对于自动分割医学图像的研究，本文总结了目前的微调策略，并在包括所有常见放射学模态的 17 个数据集上进行评估，研究发现微调 Segment Anything Model (SAM) 相较于以前的分割方法具有稍微更好的性能，使用参数效率高的学习在编码器和解码器中进行微调的策略优于其他策略，网络架构对最终性能影响不大，通过自监督学习进一步训练 SAM 可以提高最终模型性能。