Stability AI has expanded its Stable LM 2 language model to 12 billion parameters, significantly increasing its performance. The company says the update strikes a balance between power, accuracy and accessibility. It positions the model as a part of larger systems in which language models can interact with each other and with external tools.
The French startup Mistral AI has released the new AI model Mixtral 8x22B. Interested parties can download and use it free of charge. It is expected to reach and exceed the performance of models like GPT-3.5 and Llama 2 70B.
STORM is an AI system that can write Wikipedia-like articles from scratch by searching the web and using a two-step process to first create a draft with outline and references, and then the full article with citations. Source: Hacker News
Zyphra introduces Zamba, an open source 7 billion parameter model designed to bring artificial intelligence to more devices, with a decentralized approach and smaller model size to provide a more cost-effective and personalized alternative to large, centralized AI models.
Google unveiled a series of updates to its enterprise AI offerings at its annual cloud computing conference in Las Vegas. Despite recent troubles with consumer-facing tools, the company insisted the technology is safe and ready for enterprise use. Google Cloud CEO Thomas Kurian demonstrated how its most powerful AI model, Gemini, can be used for …
Opera now lets you download and run AI language models locally on your own computer – without an Internet connection, at no extra cost, and privately. There are over 150 models from more than 50 families to choose from, including Llama from Meta, Gemma from Google, and Vicuna. The feature is initially available to Opera …
KL3M is a language model that only uses content from documented, legal sources for training. It is the first such model to receive the “Fairly Trained“ seal. Source: Wired
Startup Cohere is targeting the enterprise market. Its Command-R model wants to impress with its flexibility. It includes a large context window of 128,000 tokens, and it can access external information via RAG. Source: VentureBeat
While commercial AI offerings have many guardrails and barriers to protect them from misuse, the open model Liberated-Qwen1.5-72B advertises that it has no such restrictions. Instead, it is specially trained to strictly follow the system prompt. This makes it harder to “jailbreak“. At the same time, you have to decide for yourself which answers and …
Inflection has released Inflection 2.5, an updated version of its Pi personal AI assistant that can compete with leading language models such as GPT-4. Inflection-2.5 achieves high performance on a number of benchmarks while using only 40% of the processing power of GPT-4. The basic idea behind Inflection is to make the Pi AI assistant …