TLDR AI 2023-10-23

Universal Music is suing AI start-up Anthropic for copyright infringement, alleging the chatbot "Claude" reproduces copyrighted lyrics without permission. The music industry is contending with AI's capability to create "deepfake" songs.

Nvidia Research has developed Eureka, an AI agent powered by OpenAI's GPT-4 that autonomously trains robots in complex tasks.

OpenAI is reportedly nearing a deal led by Thrive Capital that would push its valuation to $80 billion, tripling its value in under six months. This surge in its valuation started with OpenAI's ChatGPT launch, igniting significant investments in leading AI companies. AI entities remain investor favorites despite waning start-up funding due to their transformative potential in tech.

IBM has developed a brain-inspired computer chip (NorthPole) that significantly enhances AI's speed and efficiency by reducing the need to access external memory.

Most organizations don't have the resources to serve large models on GPUs, but there's a strong open-source push to run these models locally. This post explores that push and LangChain + Ollama integrations.

This project introduces a new method called 'saliency unlearning' (Salun) that improves how AI systems forget data.

Chat with and learn from your code repository.

With language models becoming multimodal, we now have more ways than ever to prompt them. State of Mark prompting runs images through Segment Anything first and numbers the detected classes. It then uses that to prompt visual models.

One of the main drivers of quality for the new DALLE-3 image generation model is improved data quality. This was obtained via caption upsampling with internal models. Here is an implementation with strong open-source models.

One of the main enterprise use cases for AI models is extracting structured data like JSON or CSV from unstructured text. This is mostly done today with function calls to gpt-3.5-turbo. GoLLIE performs slightly better in general and can be run more confidently with sensitive data.

Despite claims to the contrary, MathGLM still can not do math. This highlights the limitations of relying solely on large language models for certain tasks.

Nvidia's Deepu Talla highlights the advancements in generative AI for robotics, with the company unveiling tools in the Jetson Generative AI Lab to boost productivity, enable natural language interfaces, and enhance AI deployments at the edge.

Powerful model trained for math and reasoning.

Solve most customer questions on autopilot.

Ad creation made simple by using your existing assets.