NVIDIA announced its new Blackwell architecture at the GTC keynote. Blackwell is the Hopper architecture successor, featuring a 30x performance increase and a 25x cost and energy consumption decrease when training or performing real-time inference for LLMs of up to 10 trillion parameters.
Jan Koum's Newlands VC leads $10M investment into SuperAGI; Apple quietly bought DarwinAI to compete in the gen AI race; Databricks announces a partnership with Mistral AI; Zendesk confirms the acquisition of Ultimate; OpenAI partners with Le Monde and Prisa Media
Deci is debuting a series of LLMs and a Gen AI Development Platform. The first model is Deci-Nano, which features competitive benchmark scores, a fast throughput, and an affordable price. Deci-Nano and supported models can be deployed via Deci's platform via API, a VPC, or on-premises.
Join the Data Phoenix webinar, where Dmytro Spodarets and guest Andrey Cheptsov (Founder & CEO at dstack) will discuss the drawbacks of the Kubernetes stack in the context of AI and show how dstack addresses these drawbacks in the training and deployment of GenAI models.
Cohere recently launched Command-R, a RAG-focused model incorporating several features designed with scalable implementation in mind, including a large context window, business proficiency in 10 languages, and tool access via external APIs. Command-R is available for deployment starting March 11.
The dstack team has introduced dstack Sky, a managed service enabling users' access to GPUs from a wide selection of providers at competitive rates and without the hassle of managing individual provider accounts.
In this event, AI Makerspace breaks down the steps of ReFT, which consists of the warm-up stage and the reinforcement learning state. They will demonstrate how to code the core aspects of ReFT yourself in a Google Colab notebook environment, and all code will be provided directly to attendees!
DBSnapper allows teams to safely use production data in development and AI modeling. Now with database subsetting for smaller, relationally complete snapshots.
India is launching an AI advisory following the Google Gemini incident; Researchers are looking into optical computing as a foundation for AI; Amazon Bedrock announced the incorporation of the Claude 3 model family; OpenAI had another tumultuous week; Cognizant revealed its new advanced AI lab
A research team at Microsoft fine-tuned Mistral-7B to create Orca-Math, a small language model achieving state-of-the-art performance in solving grade school math problems (GSM8k). Orca-Math has a performance surpassing many LLMs, including math-specific models MetaMath-70B and WizardMa8th-70B.
Metaplane has raised $13.8 million in a Series A round led by Felicis Ventures. Metaplane incorporates metadata extraction, domain-specific ML, and integrations to deliver an end-to-end data observability solution.
Anthropic announced the next-generation Claude 3 model family. Opus, its most capable model, achieves state-of-the-art scores in many popular benchmarks, and Sonnet, the most balanced model, offers improvements in speed and performance compared to its predecessors.