![](https://crypto4nerd.com/wp-content/uploads/2023/07/1V7qoc3hYtiO-iQYwrrjsIQ-1024x683.jpeg)
Here are your weekly articles, guides, and news about NLP and AI chosen for you by NLPlanet!
- Anthropic releases Claude 2. Anthropic has released Claude 2, an advanced AI model that outperforms Claude 1.3 in various evaluations, achieving impressive scores on Codex HumanEval and GSM8k. Claude 2 excels in coding, math, and even scored higher on the Bar exam. Additionally, it offers improved effectiveness in harmless responses and can handle inputs up to 100K tokens, making it suitable for processing larger texts.
- Elon Musk launches AI firm xAI as he looks to take on OpenAI. Elon Musk’s new AI startup, xAI, is recruiting top engineers from tech giants like Google and Microsoft to develop a “maximally curious” AI. Although separate from X Corp, xAI will closely collaborate with companies like Twitter and Tesla, aiming to bring about breakthroughs and innovation in the field of AI through synergistic efforts.
- Google’s medical AI chatbot is already being tested in hospitals. Google’s Med-PaLM 2, an AI system designed to provide medical information, has been undergoing testing in a hospital setting. While the model shows promise in areas with limited access to doctors, it does exhibit some inaccuracies and irrelevant responses. In terms of reasoning, consensus-supported answers, and comprehension, Med-PaLM 2 performs comparably to human doctors.
- Bard’s latest update: more features, languages and countries. Bard, a language model, has expanded its availability worldwide and now supports multiple languages. New features include the ability to listen to Bard’s responses, customize the tone and style of its output, pin and rename past conversations, export Python code to Replit and Google Colab, share responses with others, and utilize images in prompts with the help of Google Lens integration.
- Shutterstock expands deal with OpenAI to build generative AI tools. OpenAI and Shutterstock have announced a partnership where OpenAI will use Shutterstock’s media library to train its AI models. In return, Shutterstock gains priority access to OpenAI’s advanced image transformation tools. Shutterstock is also working towards becoming a leader in generative AI by collaborating with top AI companies and compensating artists for their contributions to training the AI.
- Nvidia deepens bets on AI in drug discovery with Recursion investment. Nvidia invests $50M in Recursion, a biotech company using AI to revolutionize drug discovery. This partnership allows Recursion to utilize Nvidia’s platform and access their advanced AI technology. Recursion’s share price surged by 83% post-announcement, highlighting the market’s recognition of AI’s significance in drug discovery.
- NotebookLM: How to try Google’s experimental AI-first notebook. Google has introduced NotebookLM, a tool that combines Google Drive documents with LLMs. It automatically generates summaries, identifies key topics, and suggests questions to enhance understanding. Users can also ask questions about uploaded documents and request ideas or scripts related to specific topics. This tool can be beneficial for AI professionals seeking efficient document management and idea generation.
- Programs to detect AI discriminate against non-native English speakers, shows study. AI text detectors are facing bias in mistakenly labeling non-native English speakers’ articles as AI-generated. Stanford researchers found that over 50% of essays written by non-native speakers were flagged as AI-generated, emphasizing the need to address discrimination faced by non-native writers using AI detectors. This has implications for college/job applications and search engine algorithms, potentially harming academic careers and psychological well-being.
- Train LLMs using QLoRA on Amazon SageMaker. This guide explains how to use QLoRA on Amazon SageMaker for finetuning large language models. It highlights the usage of tools like Hugging Face Transformers, Accelerate, and PEFT library for adapting pre-trained language models to different applications without fine-tuning all parameters. Additionally, it emphasizes the advantages of QLoRA, such as efficient fine-tuning on a single GPU with up to 65 billion parameters and state-of-the-art results in language tasks. PEFT is also mentioned as a game-changing tool for efficient adaptation of pre-trained language models.
- A new Hugging Face Audio course. Hugging Face is offering a free and open-source Deep Learning course on audio transformers for AI professionals. This comprehensive course covers speech recognition, audio classification, and generating speech from text, providing theoretical components, quizzes, and practical exercises.
- What AI can do with a toolbox… Getting started with Code Interpreter. OpenAI has released the GPT-4 Code Interpreter plugin for ChatGPT Plus users, enabling non-coders in the artificial intelligence field to utilize AI capabilities. This tool reduces errors and improves accuracy by interacting with Python code instead of manipulating data, making it user-friendly for data analysis and AI interaction.
- Instruction Mining: High-Quality Instruction Data Selection for Large Language Models. “InstructMining” is a method that enhances the performance of large language models (LLMs) by automatically selecting high-quality instruction data. With a 42.5% improvement over models using unfiltered data, this approach focuses on the significance of data quality in fine-tuning LLMs for effective interpretation of instructions. The selection process employs natural language indicators such as naturalness, coherence, and understandability.
- Generative Pretraining in Multimodality. Emu is a powerful foundational model that excels in generating images and texts in multimodal contexts. It can handle various types of data input, such as images, text, and videos, and outperforms other large multimodal models in tasks like image captioning, visual question answering, and text-to-image generation. Emu’s strength lies in its ability to explore pretraining data sources at scale.
- Becoming self-instruct: introducing early stopping criteria for minimal instruct tuning. The Instruction Following Score (IFS) is a new metric that evaluates language models’ ability to follow instructions. It helps differentiate between base and instruct models and can prevent unnecessary finetuning that may alter a model’s semantics. Additionally, researchers found that when the IFS plateaus, significant semantic shifts occur, highlighting the relationship between instruction following and model semantics.
- GPT4RoI: Instruction Tuning Large Language Model on Region-of-Interest. GPT4RoI is an innovative model that enhances vision-language tasks by incorporating regions of interest. This allows for precise alignment between visual features and language embeddings, enabling users to interact with the model through language and spatial instructions. Moreover, GPT4RoI supports multi-region spatial instructions, expanding its multimodal capabilities. Its versatility lies in the ability to use any object detector as a spatial instruction provider, enhancing its understanding abilities with object details.
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!