Weekly AI and NLP News — May 22th 2023

StableStudio, ChatGPT for iOS, and Meta's in-house chips for AI

Here are your weekly articles, guides, and news about NLP and AI chosen for you by NLPlanet!

😎 News From The Web

📚 Guides From The Web

  • Auto-Evaluation of Anthropic 100k Context Window. Anthropic's 100k-token Claude model sparks debate on whether document retrieval is necessary for Q&A. The retriever-less architecture performs well but has higher latency than retriever-based approaches, and is most suitable for small corpora and non-critical latency apps.

  • The "Three-Hills" Model for evaluating GPT startups. "Three-Hills" Model assesses success of GPT startups with "Productivity Enhancements," "Non zero-sum-game Value," and "Moat = Value from Context" in 3 levels. It identifies 3 moat categories to avoid generic GPT displacement.

  • ChatGPT vs. Bard: A realistic comparison. The article compares ChatGPT and Bard's performance through an unscientific field experiment.

🔬 Interesting Papers and Repositories

  • ray-project/llm-numbers: Numbers every LLM developer should know, like: GPT-3.5 Turbo vs OpenAI cost, 13B model training on 1.4T tokens, average tokens/word, GPU memory, LLM requests batching.

  • microsoft/guidance: Microsoft's Guidance Language streamlines the creation of precise and parseable outputs with features such as smart caching, role-based models, and token healing.

  • Tree of Thoughts: Deliberate Problem Solving. Introducing "Tree of Thoughts" framework to language models improves problem-solving by exploring text units, optimizing tasks like planning or search.

  • CodeT5+: Open Code Large Language Models for. Introducing CodeT5+: flexible Code LLMs for various code-related tasks with a mix of pretraining objectives, achieving state-of-the-art performance. Also explores instruction-tuning.

Thank you for reading! If you want to learn more about NLP, remember to follow NLPlanet. You can find us on LinkedIn, Twitter, Medium, and our Discord server!