OpenAI

• Canvas has been updated with new features. When using Canvas, users can now see changes in their writing and code by clicking the "show changes" icons.

• gpt-4o-audio-preview is OpenAI's latest model that enables you to generate an aduio response to a prompt, allowing developers to create innovative products using its API. Here's a quick overview of this new model's capabilities:
• Generate spoken audio summaries from text input (text in, audio out)
• Perform sentiment analysis on audio recordings (audio in, text out)
• Enable asynchronous speech-to-speech interactions with the model (audio in, audio out)
OpenAI offers other models for basic speech-to-text and text-to-speech conversions. When tasks require only these conversions (without dynamic content from a model), the TTS and STT models are more efficient and cost-effective. For more information, visit: audio-generation

• ChatGPT Windows version has been released. ChatGPT Plus, Enterprise, Team, and Edu users can now test an early version of the Windows desktop app.

Anthropic

• Claude's iOS and Android apps have received a fresh new look. The app is now also available on iPad. Furthermore, they've introduced support for projects on mobile devices. Users can search chats, access project knowledge, and adjust custom instructions for each project.

• Safety has become a crucial and trending topic in today's AI landscape. Anthropic has published a blog post about their Responsible Scaling Policy, which aligns safety and security measures with an AI model's capabilities. For more information, visit: Announcing our updated Responsible Scaling Policy

• Anthropic has released new research on "Sabotage evaluations for frontier models." The team explores how effectively AI models could potentially mislead us. Their findings suggest that for current models, minimal mitigations are sufficient to address sabotage risks. For more information, visit: Sabotage evaluations for frontier models

• Anthropic has published their second Quickstart—a financial data analyst powered by Claude. You can upload spreadsheets, documents, or financial charts and instantly get actionable insights with beautiful visualizations. For the detailed code, visit: financial data analyst

• Anthropic's CEO Dario Amodei has published a blog post titled Machines of Loving Grace, which discusses how AI could transform the world for the better. For the full content, visit: Machines of Loving Grace

Google

• NotebookLM has been a standout AI product over the past few weeks. Users can now customize audio overviews and guide the AI before generating content. Moreover, Google has launched NotebookLM for business users.

• Google has introduced several new parameters for the Gemini API:
• logprobs
• candidateCount
• seed
• presencePenalty
• frequencyPenalty
• model version in response

• Compare mode in Google AI studio has now been introduced for users. It is a powerful new feature designed to help user make informed decision about which Gemini model best suits their needs. For more information, visit: Compare Mode in GoogleAI Studio

Mistral

• Mistral AI has introduced two new state-of-the-art models, Mistral 3B and Mistral 8B, designed for on-device computing and edge use cases. These models set new benchmarks in knowledge, common sense, reasoning, function-calling, and efficiency within the sub-10B category. They can be used or fine-tuned for a variety of applications. For more information, visit:Mistral 3B & Mistral 8B

DeepSeek

• The team from DeepSeek has introduced Janus, a revolutionary autoregressive framework for multimodal AI. By decoupling visual encoding and unifying it with a single transformer, Janus outperforms previous models in both understanding and generation. For more information, visit: Janus

Yi-01.AI

• The team has unveiled two latest proprietary models: Yi-Lightning and Yi-Lightning-Lite. While details about these models are scarce, you can test them yourself at: Yi-Lightning & Yi-Lightning-Lite