All tags
Model: "openelm"
Apple's OpenELM beats OLMo with 50% of its dataset, using DeLighT
openelm llama-3 llama-3-8b-instruct llama-3-70b apple meta-ai-fair google layer-wise-scaling context-length quantization ai-alignment open-source ai-regulation eric-schmidt sebastian-raschka
Apple advances its AI presence with the release of OpenELM, its first relatively open large language model available in sizes from 270M to 3B parameters, featuring a novel layer-wise scaling architecture inspired by the DeLight paper. Meanwhile, Meta's LLaMA 3 family pushes context length boundaries with models supporting over 160K tokens and an 8B-Instruct model with 262K context length released on Hugging Face, alongside performance improvements in quantized versions. A new paper on AI alignment highlights KTO as the best-performing method, with sensitivity to training data volume noted. In AI ethics and regulation, former Google CEO Eric Schmidt warns about the risks of open-source AI empowering bad actors and geopolitical rivals, while a U.S. proposal aims to enforce "Know Your Customer" rules to end anonymous cloud usage.
OpenAI's Instruction Hierarchy for the LLM OS
phi-3-mini openelm claude-3-opus gpt-4-turbo gpt-3.5-turbo llama-3-70b rho-1 mistral-7b llama-3-8b llama-3 openai microsoft apple deepseek mistral-ai llamaindex wendys prompt-injection alignment benchmarking instruction-following context-windows model-training model-deployment inference performance-optimization ai-application career-advice drive-thru-ai
OpenAI published a paper introducing the concept of privilege levels for LLMs to address prompt injection vulnerabilities, improving defenses by 20-30%. Microsoft released the lightweight Phi-3-mini model with 4K and 128K context lengths. Apple open-sourced the OpenELM language model family with an open training and inference framework. An instruction accuracy benchmark compared 12 models, with Claude 3 Opus, GPT-4 Turbo, and Llama 3 70B performing best. The Rho-1 method enables training state-of-the-art models using only 3% of tokens, boosting models like Mistral. Wendy's deployed AI-powered drive-thru ordering, and a study found Gen Z workers prefer generative AI for career advice. Tutorials on deploying Llama 3 models on AWS EC2 highlight hardware requirements and inference server use.