In April, Meta Platforms revealed that it was working on an open-source AI model that performed as well as proprietary models from top AI companies such as OpenAI and Anthropic. Now, Meta CEO Mark Zuckerberg says that model has arrived in the form of Llama 3.1 405B, “the first frontier-level open-source AI model.” The company is also releasing “new and improved” Llama 3.1 70B and 8B models. In addition to general cost and performance benefits, the fact that the Llama 3.1 405B model is open source “will make it the best choice for fine-tuning and distilling smaller models,” according to Meta. Continue reading Meta Calls New Llama the First Open-Source Frontier Model

Apple has released a public demo of the 4M AI model it developed in collaboration with the Swiss Federal Institute of Technology Lausanne (EPFL). The technology debuts seven months after the model was first open-sourced, allowing informed observers the opportunity to interact with it and assess its capabilities. Apple says 4M was built by applying masked modeling to a single unified Transformer encoder-decoder “across a wide range of input/output modalities — including text, images, geometric and semantic modalities, as well as neural network feature maps.” Continue reading Apple Launches Public Demo of Its Multimodal 4M AI Model

Stability AI has added another audio product to its lineup, releasing the open-source text-to-audio generator Stable Audio Open 1.0 for sound design. The new model can generate up to 47 seconds of samples and sound effects, including drum beats, instrument riffs, ambient sounds, foley and production elements. It also allows for adapting variations and changing the style of audio samples. Stability AI — best known for the image generator Stable Diffusion — in September released Stable Audio, a commercial product that can generate sophisticated music tracks of up to three minutes. Continue reading Stability AI Releases Free Sound FX Tool, Stable Audio Open

Elon Musk’s xAI has secured $6 billion in Series B funding. While the company says the funds will be “used to take xAI’s first products to market, build advanced infrastructure, and accelerate the research and development,” some outlets are reporting a significant portion is earmarked to build an AI supercomputer to power the next generation of its foundation model Grok. The company publicly released the open-source Grok-1 as a chatbot on X social in November, and recently debuted Grok-1.5 and 1.5V iterations with long-context capability and image understanding. Continue reading Musk Said to Envision Supercomputer as xAI Raises $6 Billion

Meta Platforms has unveiled its first natively multimodal model, Chameleon, which observers say can make it competitive with frontier model firms. Although Chameleon is not yet released, Meta says internal research indicates it outperforms the company’s own Llama 2 in text-only tasks and “matches or exceeds the performance of much larger models” including Google’s Gemini Pro and OpenAI’s GPT-4V in a mixed-modal generation evaluation “where either the prompt or outputs contain mixed sequences of both images and text.” In addition, Meta calls Chameleon’s image generation “non-trivial,” noting that’s “all in a single model.” Continue reading Meta Advances Multimodal Model Architecture with Chameleon

Google is offering developers a toolkit for incorporating generative AI features into mobile and web applications. Firebase Genkit, an open-source framework, is available now in beta. Blending models, cloud services, agents, data sources and more in a “code-centric approach” developers are used to, the Genkit makes building and debugging for AI easier, according to Google. The first release is built for JavaScript and TypeScript developers, making building AI-powered apps available to professionals who specialize in building server-side applications using the Node.js JavaScript runtime. Continue reading Firebase Genkit: Developer Framework for AI-Powered Apps

In a move aimed at launching more accessible Android apps, Google has open-sourced code for Project Gameface, a hands-free game control feature released last year that allows users to move a computer with facial and head gestures. Developers will now have more Gameface resources with which to build Android applications for physically challenged users, “to make every Android device more accessible.” Project Gameface evolved as a collaboration with quadriplegic video game streamer Lance Carr, who has muscular dystrophy. The technology uses a smartphone’s front camera to track movement. Continue reading Google Adds Open-Source Gameface for Android Developers

IBM has released a family of its Granite AI models to the open-source community. The series of decoder-only Granite code models are purpose-built to write computer code for enterprise developers, with training in 116 programming languages. These Granite models range in size from 3 to 34 billion parameters in base model and instruction-tuned variants. They offer a range of uses, from modernizing older code with new languages to optimizing programs for on-device memory constraints, such as might be experienced when conforming for mobile gadgets. In addition to generation, the models can repair and explain code. Continue reading IBM Introduces Granite LLMs for Enterprise Code Developers

The UK AI Safety Institute announced the availability of its new Inspect platform designed for the evaluation and testing of artificial intelligence tech in order to help develop safe AI models. The Inspect toolset enables testers — including worldwide researchers, government agencies, and startups — to analyze the specific capabilities of such models and establish scores based on various criteria. According to the Institute, the “release comes at a crucial time in AI development, as more powerful models are expected to hit the market over the course of 2024, making the push for safe and responsible AI development more pressing than ever.” Continue reading UK Launches New Open-Source Platform for AI Safety Testing

The trend toward small language models that can efficiently run on a single device instead of requiring cloud connectivity has emerged as a focus for Big Tech companies involved in artificial intelligence. Apple has released the OpenELM family of open-source models as its entry in that field. OpenELM uses “a layer-wise scaling strategy” to efficiently allocate parameters within each layer of the transformer model, resulting in what Apple claims is “enhanced accuracy.” The “ELM” stands for “Efficient Language Models,” and one media outlet couches it as “the future of AI on the iPhone.” Continue reading Apple Unveils OpenELM Tech Optimized for Local Applications

Pursuant to his goal of “building the world’s leading AI,” Meta Platforms CEO Mark Zuckerberg announced Friday that Meta AI is upgrading to Llama 3 concurrent with a rollout of its open-source chatbot across the company’s social platforms, integrating it into the search boxes atop WhatsApp, Instagram, Facebook and Messenger. There is also a website, meta.ai, for those who prefer browser access. Reports of Meta upgrading its social AI capabilities began leaking out early last week, albeit on a more limited test scale than what Zuckerberg announced, which, excepting Threads, is cross-platform. Continue reading Meta AI Assistant Is Launching Across Platforms with Llama 3

Google is moving its most powerful artificial intelligence model, Gemini 1.5 Pro, into public preview for developers and Google Cloud customers. Gemini 1.5 Pro includes what Google claims is a breakthrough in long context understanding, with the ability to run 1 million tokens of information “opening up new possibilities for enterprises to create, discover and build using AI.” Gemini’s multimodal capabilities allow it to process audio, video, text, code and more, which when combined with long context, “enables enterprises to do things that just weren’t possible with AI before,” according to Google. Continue reading Google Offers Public Preview of Gemini Pro for Cloud Clients

Opera has become the first browser to add support for large language models (LLMs). At this point the feature is experimental, and available only on the Opera One Developer browser as part of the AI Feature Drops program. The update offers about 150 LLMs from more than 50 different families, including Meta’s LLaMA, Google’s Gemma, Mixtral and Vicuna. Opera had previously only offered local support for its own Aria AI, a competitor to Microsoft Copilot and OpenAI’s ChatGPT. The local LLMs are being offered for testing as a complimentary addition to Opera’s online Aria service. Continue reading Opera Browser Is Experimenting with Local Support for LLMs

Databricks, a San Francisco-based company focused on cloud data and artificial intelligence, has released a generative AI model called DBRX that it says sets new standards for performance and efficiency in the open source category. The mixture-of-experts (MoE) architecture contains 132 billion parameters and was pre-trained on 12T tokens of text and code data. Databricks says it provides the open community and enterprises who want to build their own LLMs with capabilities previously limited to closed model APIs. Compared to other open models, Databricks claims it outperforms alternatives including Llama 2-70B and Mixtral on certain benchmarks. Continue reading Databricks DBRX Model Offers High Performance at Low Cost

Apple researchers have gone public with new multimodal methods for training large language models using both text and images. The results are said to enable AI systems that are more powerful and flexible, which could have significant ramifications for future Apple products. These new models, which Apple calls MM1, support up to 30 billion parameters. The researchers identify multimodal large language models (MLLMs) as “the next frontier in foundation models,” which exceed the performance of LLMs and “excel at tasks like image captioning, visual question answering and natural language inference.” Continue reading Apple Unveils Progress in Multimodal Large Language Models