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

😎 News From The Web

• Google launches PaLM 2, its next-gen large language model. PaLM 2 will power Google’s updated Bard chat tool, the company’s competitor to OpenAI’s ChatGPT, and function as the foundation model for most of the new AI features the company is announcing today. PaLM 2 is now available to developers through Google’s PaLM API, Firebase, and on Colab.

• Meta open-sources multisensory AI model that combines six types of data. Meta has unveiled ImageBind, an open-source AI model indexing six data types (visual, audio, text, thermal, depth, movement) for multisensory AI.

• Introducing 100K Context Windows: Claude, by Anthropic. AI language model, Claude, now analyzes and synthesizes vast text in seconds with a 100K token context window. Summarize docs, assess risks, and more.

• Language models can explain neurons in language models, by OpenAI. GPT-4 automates understanding of language models by producing natural language explanations of neuron behavior.

• The AI takeover of Google Search starts now. Google's new "AI snapshots" use language models to generate summaries for richer searches. Users can opt-in to the experiment called Search Generative Experience.

• ‘Godfather of AI’ says AI threat is ‘more urgent’ to humanity than climate change. AI expert Geoffrey Hinton, "Godfather of AI," considers risks of unbridled AI more pressing for humanity than climate change. Hinton warns of threats like job loss and misinformation, but does not support a stoppage of AI development.

• AMP Robotics attracts investment from Microsoft’s Climate Innovation Fund. AMP Robotics, a Denver, Colorado-based startup creating robotic systems that can automatically sort recyclable material, announced that it extended its Series C round to $99 million, thanks to an investment from Microsoft’s Climate Innovation Fund.

• EU lawmakers' committees agree tougher draft AI rules. EU to classify AI tools by risk level; proposed draft legislation bans facial recognition in public spaces, with grace period of 2 years before law.

📚 Guides From The Web

• Plan-and-Execute Agents. Introducing Plan-and-Execute: a new agent executor that separates higher-level planning from short-term execution for complex long-term planning.

• AI Will Create More Developers, Not Less. AI will accelerate global developer growth by minimizing entry barriers with open- and closed-source products for different audiences, building smaller firms.

• Learnings exploring the GPT/ LLM space. LLM has huge potential for businesses but requires software engineering skills, GPT-4 is the best-performing, but costly. Training costs are expected to decrease.

• GPT-4’s Maze Navigation: A Deep Dive into ReAct Agent and LLM’s Thoughts. The article explores GPT-4 for maze navigation. It uses memorization-based navigation, labeling, and A* search but lacks planning skills.

• Building ML infrastructure: An interview with Aditya Nambiar, Founding engineer at Fennel, about building ML infrastructure.

🔬 Interesting Papers and Repositories

• InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning. The authors conduct a systematic and comprehensive study on vision-language instruction tuning based on the pre-trained BLIP-2 models The resulting InstructBLIP models achieve state-of-the-art zero-shot performance across all 13 held-out datasets, substantially outperforming BLIP-2 and the larger Flamingo.

• Augmented Large Language Models with Parametric Knowledge Guiding. The authors propose a novel Parametric Knowledge Guiding (PKG) framework, which equips LLMs with a knowledge-guiding module to access relevant knowledge at runtime without altering the LLMs' parameters.

• Active Retrieval Augmented Generation. The authors propose Forward-Looking Active REtrieval augmented generation (FLARE), a generic retrieval-augmented generation method which generates articles/overviews by iteratively using a prediction of the upcoming sentence to anticipate future content, which is then utilized as a query to retrieve relevant documents to regenerate the sentence if it contains low-confidence tokens.

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!