Microsoft and Tsinghua University have developed a new AI architecture called “Differential Transformer” that improves the performance of large language models. Furu Wei from Microsoft Research told VentureBeat that the new method amplifies attention to relevant contexts and filters out noise. This is designed to reduce problems such as the “lost-in-the-middle” phenomenon and hallucinations in …
Nvidia has introduced a new AI model, Llama-3.1-Nemotron-70B-Instruct, which outperforms existing models from OpenAI and others, continuing a significant shift in its AI strategy. The model, available on Hugging Face, achieved impressive benchmark scores, positioning Nvidia as a competitive player in AI language understanding and generation. This development showcases Nvidia’s transition from a GPU manufacturer …
French AI company Mistral has introduced new generative AI models for laptops and smartphones. Known as “Les Ministraux,” the models are optimized for various applications such as text generation or collaboration with more powerful models. Kyle Wiggers reports for TechCrunch that two variants are available: Ministral 3B and Ministral 8B, both with a context window …
Zyphra has released Zamba2-7B, a new small language model supposedly outperforming competitors like Mistral, Google’s Gemma, and Meta’s Llama3 in quality and performance. According to the Zyphra team, Zamba2-7B is ideal for consumer devices, GPUs, and enterprise applications. It boasts 25% faster time to first token, 20% more tokens per second, and reduced memory usage …
A Reddit post titled “Which is the best version of chatgpt4” sparked a discussion about the various models of ChatGPT. The original poster inquired about the most accurate version, leading to various responses. Users highlighted the strengths of different models: “4o” is considered best for research, image generation, and creative writing, while “o1 mini” excels …
Writer has launched its new AI language model, Palmyra X 004, which seemingly excels in function calling and workflow execution for businesses. Michael Nuñez reports for VentureBeat that the model outperforms competitors like OpenAI, Anthropic, Google, and Meta by nearly 20% on Berkeley’s Tool Calling Leaderboard, achieving a score of 78.76%. Palmyra X 004 achieves …
Google has released a new version of its AI model Gemini. Gemini 1.5 Flash-8B is now available for production use, as announced on the Google Developers Blog. Compared to its predecessor, the model offers 50% lower prices, twice the request limits, and lower latency for short inputs. Developers can access Gemini 1.5 Flash-8B for free …
OpenAI introduces a new feature called “Canvas” for ChatGPT. The new interface allows users to edit text and code directly next to the chat window. Canvas is initially available for ChatGPT Plus and Teams users, later also for Enterprise and Edu customers. The function aims to simplify writing and programming with AI assistance. With Canvas, …
Nvidia has released a powerful open-source AI model that rivals proprietary systems from industry leaders like OpenAI and Google. The model, called NVLM 1.0, demonstrates exceptional performance in vision and language tasks while also enhancing text-only capabilities. Michael Nuñez reports on this development for VentureBeat. The main model, NVLM-D-72B, with 72 billion parameters, can process …
Mostly AI has introduced a new feature for generating synthetic texts. The tool allows companies to use confidential data such as emails and conversations for AI training without privacy concerns. As reported by Shubham Sharma, the platform generates a version of proprietary information free from personally identifiable data. This enables businesses to train and optimize …