Anthropic

• The biggest bomb in AI community this week is from Anthroic. The team has released their new Claude models, Claude 3.5 Sonnet and Claude 3.5 Haiku . Here is a benchmark result to show their excellence:



The updatedClaude 3.5 Sonnetshows significant improvements across many benchmarks, with particularly strong gains in agentic coding and tool use tasks. In coding, it improves performance on SWE-bench Verifiedfrom 33.4% to 49.0%, scoring higher than all publicly available models—including reasoning models like OpenAI's o1-preview and specialized systems designed for agentic coding. Surprisingly,Gemini 1.5 Flashalso demonstrates impressive performance, especially considering its cost. For more information, please visit:Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku

• Computer use is another innovative feature released by Anthropic. Claude can now interact with your computer to accomplish tasks you specify. Although this capability is currently in public beta, it represents a significant breakthrough in AI progress. For more information, please visit: Developing a computer use model

• A notable and practical update following Claude's release is its newfound ability to write and execute code for data processing, analysis, and real-time data visualization. This advancement enables users to leverage Claude for data analysis tasks.

• My personal experience:
• I've been using the newClaude 3.5 Sonnetfor several days, and I'm thoroughly impressed. Using Cursor with this new model, I've developed a web application—an SVG card generator. It leverages Claude's ability to create well-designed cards that display critical interpretations of user-input words. Alternatively, you can customize your system prompt to have Claude generate cards tailored to your specifications. Here are some demos:






If you are interested this, please visit here: SVG Card Generator

OpenAI

• This week doesn't belong to OpenAI, so they don't have much news. However, they introduced sCMs, the latest consistency models with a simplified formulation, improved training stability, and scalability. This approach achieves comparable sample quality to leading diffusion models while using only two sampling steps. If you're interested in more information, please visit: Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models

• There are also many rumors suggesting that OpenAI will release something significant next month, coinciding with ChatGPT's second birthday!

IDEOGRAM

• Ideogram, the image generation company, has introduced Canvas, an infinite creative board for organizing, generating, editing, and combining images. You can upload your own images or generate new ones within Canvas, then seamlessly edit, extend, or combine them using industry-leading Magic Fill and Extending tools. For more detailed information, please visit: Ideogram Canvas, Magic Fill, and Extend

Midjourney

• Midjourney has released the image editor this week. It allows users to upload images from their computers and then expand, crop, repaint, add, or modify elements in the scene. This feature is very similar to the canvas from Ideogram.

• The team has also introduced the image retexturing mode, which estimates the shape of the scene and then retextures it, changing all the lighting, materials, and surfaces.

Zhipu AI

• Zhipu AI from China has released AutoGLM and GLM-4-Voice. AutoGLM is similar to Anthropic's computer use feature, but it controls the user's phone instead of a computer. GLM-4-Voice is an end-to-end voice model that can directly understand and generate Chinese and English speech, engage in real-time voice conversations, and modify attributes such as emotion, speech rate, and dialect based on user instructions. If you want to know more detailed information, please visit: GLM-4-Voice