AI News

June 12, 2024

RAG Systems

Fine-tuning

  1. Llama 3 and Qwen 2 Performance Comparisons Llama 3 8B Instruct has shown impressive performance on consumer-grade hardware, achieving a 0.8 correlation with GPT-4 scores. Meanwhile, Qwen 2 has surpassed Llama 3 on MMLU, leading to discussions on its potential as a superior model. Read more

Language Model

  1. Mixture of Agents (MoA) Framework The new MoA framework leverages multiple LLMs to refine responses, achieving a high performance on AlpacaEval 2.0. This approach highlights the potential of collaborative AI models in improving accuracy. Read more

  2. Google Gemini and LlamaGen Google’s RecurrentGemma 9B and the LlamaGen project are making waves in the AI community for their impressive performance and innovative approaches to language modeling and image generation. Read more

Security

  1. OpenAI Uses Oracle Cloud OpenAI has selected Oracle Cloud Infrastructure to extend its AI platform, enhancing its capabilities alongside Microsoft Azure. Read more

Others

  1. ARC Prize Challenge A $1M competition launched by François Chollet aims to create AI that adapts to novelty and solves reasoning problems, pushing the boundaries towards AGI. Read more

  2. MLX 0.2 for Apple Silicon Macs The latest MLX update offers a revamped user experience, faster retrieval-augmented generation, and full-featured chat capabilities on Apple Silicon Macs. Read more

  3. Stable Artisan for Discord Stability AI introduces Stable Artisan, a Discord bot integrating various models for media generation and editing, sparking conversations about its open-source status and paid API service. Read more

  4. Tesla Optimus Deployment Tesla has deployed two Optimus robots performing autonomous tasks in their factory, marking a significant step in industrial automation. Read more

  5. Stable Diffusion 3.0 Launch Stability AI has released Stable Diffusion 3.0 Medium, promising enhanced detail, color, and prompt understanding. Despite the excitement, users report mixed reactions about its human anatomy accuracy and restrictive licensing terms. The community is actively discussing finetuning challenges and integration issues with existing frameworks. Read more

  6. Autoregressive Image Models vs. Diffusion Models New research shows autoregressive models like Llama outperforming diffusion models for scalable image generation, challenging previous assumptions. Read more