Model movement to catch up on in AI.
Meta has released the Llama 3 large language model, which powers Meta AI and offers improved reasoning, code generation, and diverse responses. Mark Zuckerberg, Meta CEO, claims that Meta AI is now the most intelligent AI assistant available for free use. Meta plans to make Llama 3 multilingual and multimodal in the future. The AI assistant can assist with tasks like recommending restaurants and planning trips, and Meta AI’s Imagine feature produces sharper images faster. Meta AI is available across various platforms and will soon be integrated into Ray-Ban Meta smart glasses.
Amazon Web Services is enhancing its Bedrock artificial intelligence service by allowing customers to import their own custom AI models alongside models from major providers. This move aims to appeal to AI application developers and compete with other cloud giants like Microsoft and Google. The new capability is available in preview and includes updates to Amazon’s Titan foundation models and the availability of third-party models on Bedrock. Additionally, Amazon is releasing its Model Evaluation capability and Guardrails feature.
Microsoft has launched Phi-3 Mini, its smallest AI model yet, with 3.8 billion parameters. It is the first of three small models the company plans to release, and it performs better than the previous version. Small AI models like Phi-3 are cheaper to run and perform well on personal devices. Phi-3 was trained with a “curriculum” inspired by how children learn from simpler books and sentence structures.
Adobe has introduced new generative AI tools in Photoshop, powered by the Firefly Image 3 model. The Reference Image tool allows users to upload images that serve as a guide for the AI-generated output. Other tools include Generate Background, Enhance Detail, Generate Similar, and Generate Image. Adobe claims that Firefly is safe for commercial use and offers higher-quality image generation.
According to Absolute Security, the majority of organizations need to update or replace their systems to be AI-ready. A recent analysis found that 92% of Windows machines in organizations with 500 or more active devices running Windows 10 or 11 don’t meet the basic requirements to handle modern AI applications efficiently. Absolute Security warns that considerable investments in AI-capable endpoint fleets can divert resources from critical IT and security priorities, potentially leaving gaps in security and risk policies.
Right tool for the right job continues to be the theme for AI models. They are distinctly not created equal.
Let’s dwell on that Absolute Security statement. Warning of diversion from IT and security spend by a security provider has an interesting inverse… perhaps organizations will get a higher return on that investment, and thus why they will pick it? For security teams, you may need to make a better argument.