Weekly AI and NLP News — February 6th 2023

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

😎 News and Guides From The Web

• ChatGPT Plus: a subscription plan for ChatGPT. Subscribers will receive general access to ChatGPT, even during peak times, faster response times, and priority access to new features and improvements. ChatGPT Plus will be available for $20/month.

• OpenAI is launching a classifier trained to distinguish between AI-written and human-written text. The classifier correctly identifies 26% of AI-written text (true positives) as "likely AI-written," while incorrectly labeling human-written text as AI-written 9% of the time (false positives).

• Chatbots with GPT3. Recent advancements in LLMs, such as GPT-3, can be used for chatbot development. Instead of having many very specific intents, each intent can be broader and leverage a Knowledge Base document.

• OpenAI has ramped up its hiring around the world, bringing on roughly 1,000 remote contractors over the past six months to create massive sets of images and audio clips to train AI tools. 60% of the contractors do data labeling and 40% are computer programmers.

• Scale AI hosted a Generative AI hackathon.

• Querying NBA stats with GPT-3 + Statmuse + Langchain. Using Langchain, the author composed an AI program that combines GPT-3 with Statmuse, a sports stats search engine, to answer multi-part questions about NBA stats.

• Noise2Music: a series of diffusion models trained to generate high-quality 30-second music clips from text prompts. The generated audio is able to faithfully reflect key elements of the text prompt such as genre, tempo, instruments, mood and era.

• Chinese Search Giant Baidu to Launch ChatGPT-Style Bot. Baidu, known as China’s Google, will embed it in search engine.

• Microsoft, GitHub, and OpenAI ask court to throw out AI copyright lawsuit. The three companies want to dismiss a complaint that alleges that the AI-powered Copilot relies on ‘software piracy on an unprecedented scale.

• Trends in AI  in 2023. Predictions on language models, reinforcement learning, robotics, computer vision...

• A Short Chronology Of Deep Learning For Tabular Data.

• The Flan Collection: a newer & more extensive publicly available collection of tasks, templates & methods for instruction tuning. It yields improved performance on all tested evaluation benchmarks & enables models to reason more competently over arbitrary tasks.

• The Annotated Transformer: an annotated version of the paper "Attention is all you need" in the form of a line-by-line implementation.

• The Transformer Family: a summary of Transformer architecture improvements.

• Google invested $300 million in AI firm Anthropic, founded by former OpenAI researcher.

• GPT-3 integrated in Microsoft Teams Premium, with AI-generated notes and tasks.

🫶 NLPlanet News

• The new lessons to be published on the website will be about implementing Transformers from scratch in PyTorch. They'll be published in a few weeks.

🔬 Interesting Papers

• Watermarks for large language models can be embedded with minimal impact on text quality and detected using an efficient algorithm.

• What are the cognitive capabilities of today's large language models? A cognitive perspective on dissociating language and thought in large language models.

• SparseGPT: a new pruning method that can reduce GPT-family models to at least 50% sparsity in one-shot, without any retraining and with minimal loss of accuracy. With SparseGPT, more than 100 billion weights from the largest available open-source models can be ignored at inference time.

• Low-Rank Adaptation (LoRA): efficient fine-tuning of large language models using fewer parameters and less GPU memory.

• Faithful Chain-of-Thought Reasoning, a faithful-by-construction framework that boosts Language Models' performance on complex reasoning tasks.

🙃 Memes

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!