Aligning LLMs: DPO
Join AI Makerspace on Wednesday, Feb. 28, at 10 AM PT! For a workshop on DPO. Come learn how it differs from the techniques of RLHF and why all indications are showing it's rapidly becoming the new industry standard.
News
The open-source multilingual LLM Aya supports over 100 languages
Cohere for AI's Aya Project is working towards closing the gap of language representation in AI by providing coverage to unserved and underrepresented languages. The Aya model and its dataset are available under a fully permissive Apache 2.0 license.
This week in AI: February 18-24
Bioptimus is building the world's first universal AI foundation model for biology; Google halted Gemini's people-generation capabilities, launched Chrome AI-powered experimental features, and Gemini for Workspace plans; Dating app company Match Group announced OpenAI partnership
YOLOv9 promises to be the new state-of-the-art real-time object detector
YOLOv9 is a real-time object detector with performance competitive enough to become the newest state-of-the-art method. YOLOv9 features two improvements: Programmable Gradient Information (PGI) and Generalized Efficient Layer Aggregation Network (GELAN)
Webinar "Evaluating LLM Models for Production Systems: Methods and Practices"
Join the Data Phoenix webinar, where Dmytro Spodarets and guest Andrei Lopatenko will discuss a comprehensive understanding of the evaluation processes for LLMs, particularly in the context of preparing these models for deployment in production environments.
Google has announced the next-generation Gemini 1.5 as it expands access to Gemini 1.0 via Vertex AI
Google expanded the availability of the Gemini 1.0 models on Vertex AI, making 1.0 Pro generally available and 1.0 Ultra available via allowlist. Google also announced the private preview of the next-gen Gemini 1.5 model based on its Transformer and Mixture-of-Experts architecture.
Langchain announces Series A and the general availability of LangSmith
Langchain recently announced its successful $25 million Series A round led by Sequoia Capital. Langchain also announced the general availability launch of LangSmith, its platform supporting LLM application prototyping, beta testing, and production.
OpenAI becomes the center of attention with the Sora early announcement
OpenAI has spurred a frenzy of reactions after sharing its research advances on Sora, a SOTA diffusion model for video generation that can generate up to 60 seconds of realistic or animated video. Sora is currently available to red teams and invited creatives only.
This Week in AI: February 11-17
Otter introduces Meeting GenAI, a collection of AI-powered tools for meetings; Anthropic is testing the Prompt Shield technology for US-based users to prevent election misinformation; Slack announced the the launch of Slack AI; Chat with RTX allows select AI PC owners to customize their own chatbot
The UAE's TII launches the Falcon Foundation to support the open-source generative AI ecosystem
The UAE's Technology Innovation Institute (TII) is committing $300 million to launch the Falcon Foundation, a non-profit bringing together essential actors in AI to support and develop open-source projects, especially those based on the TII's Falcon models.
ChatGPT will soon be able to remember things
OpenAI is planning to roll out a memory feature for ChatGPT. Users can let the chatbot pick up details, or they can ask ChatGPT to remember something in the conversation. Memory will roll out to a select number of users, with plans to eventually make the feature broadly available.
Google rebrands Bard as Gemini as part of a significant AI update
Google has rebranded Bard as Gemini across all platforms and has also introduced a paid Gemini Advanced service. Powered by Google's most capable model, Gemini Advanced is currently available in English. The company also announced an updated Gemini experience in its Android and iOS apps.