@Meta's newest #LlaMa release is just 🚀. The team chose to prioritize model capability over efficiency by training their LLaMa-3 model on 15 trillion tokens, diverging from the established Chinchilla scaling laws. This has propelled LLaMa-3 into the top 5 of the LLM Arena.
We summarize major news from AI research, industry, and investment (cutting through hype and sharing only what matters). Read through full newsletter and subscribe today!
genai360.beehiiv.com/subscribe
@Microsoft unveiled three new AI model families:
- Compact Phi to compete with Gemini and GPT-3.5
- VASA-1 for realistic talking videos
- WizardLM-2, temporarily halted for toxicity testing
Initial tests suggest Phi's smallest version rivals Llama-3 8B despite a smaller dataset
@BostonDynamics unveiled a new version of the Atlas robot with enhanced mobility and human-like capabilities (did the Robotics investment FOMO already start?)
@GoogleAI introduced mobile device processing advancements and a new transformer architecture called TransformerFAM. TransformerFAM outperforms existing models on long-context tasks, highlighting its ability to manage longer sequences and maintain context over extended periods.
Major investment wins from UAE-based AI firm G42 ($1.5 billion investment from Microsoft), and @Perplexity's latest round makes it a unicorn.
Read 20 more news items that mattered this week in our round up of news across industry, academia, and investments. Subscribe to stay informed with weekly insights into the rapidly evolving field of AI!
genai360.beehiiv.com/p/of-llamas-and-phis