Grammarly announced today that it's (unsurprisingly) diving into the generative AI fray. GrammarlyGo is an upcoming set of auto-composition features to help the AI proofreading software keep up with the many companies adding the ChatGPT API (or different generative AI backends) to their products. GrammarlyGo can use context like voice, style, purpose and where you're writing to determine its approach. So, for example, it can spit out email replies, shorten passages, rewrite them for tone and clarity, brainstorm or choose from one-click prompts -- all while adhering to your company's voice or other provided context. In addition, since Grammarly's desktop service can pop up in any text field on your computer, its generative writing could be slightly more convenient than competitors (like Notion or Gmail's Smart Compose) that require you to visit an app or website.

Prompts may well be the new oil. Writing the text strings that instruct AI systems like ChatGPT and DALL-E 2 to generate essays, articles, images and more has become a veritable profession, commanding salaries well into the six-figure range. Anyone can come up with prompts, of course. But only certain prompts (e.g. "Create a watercolor of a solider standing in the middle of a field, in the style of John Singer Sargent) accomplish very specific, desirable (or undesirable) things. Prompt writing requires skill and dedication, owing to the black box and unpredictable nature of today's bleeding-edge AI systems. Complicating matters further, the systems are frequently changing and responding to malicious prompts, bypassing the guardrails that their makers put in place. But not every company or developer has the budget to hire a so-called prompt engineer. Prompt marketplaces, or e-commerce portals where users can buy, sell or give away prompts "designed" for various AI systems, are a growing industry. When we first profiled prompt marketplaces last July, there was only one major player. But since then, the landscape has expanded dramatically. Even a cursory Google search turns up a dozen or more prompt marketplaces, with new ones added on a monthly basis. ChatX, for instance, offers prompts tuned to ChatGPT as well as popular image-generating systems like DALL-E 2, Midjourney and Stable Diffusion. NeutronField's prompts for sale cover a slightly wider range of AI systems, including Disco Diffusion and Craiyon. Many of the marketplace operators, like NeutronField's Miroslav Kostic, have no background in AI or even data science. They were hobbyists to start, experimenting with systems like Stable Diffusion but running into hurdles unlocking their full potential. "I've been playing with AI text-to-image models since Disco Diffusion first appeared in September 2021," Kostic told TechCrunch in an email interview. "I spent countless hours trying to bring to life the ideas that had been in my head for years -- dystopian sci-fi landscapes and otherworldly spacescapes.

OpenAI 推出的 ChatGPT 到底是不是 1750 亿参数的等价大模型呢？这篇文章或许能带给你答案。

Talk of the town, an unsaid threat in the minds of many professionals, a competitor's catalyst for sleepless nights, and just a source of uncertainty and amazement for many, ChatGPT has crawled its way into our existence with an ear-crashing drumroll! With 13 million active individual users, a valuation of $29 billion, and many such unbelievable numbers, Chat GPT is impossible to ignore! The Chatbot that has taken the world by storm of late, you do know what it is, right? If not, which is shocking, let's get acquainted! Developed by OpenAI, ChatGPT is the fine-tuned version of OpenAI's GPT-3 family of large language models, trained using both supervised and reinforcement learning techniques.

We did it all for free! I hope you found this article informative and hopefully the code is not too intimidating. If you face any hangups, feel free to leave a comment and I will do my best to help. If you would like to create an AI Text Editor in Google Docs, I wrote a piece on that too. As I mentioned, I plan on covering more practical use cases of AI in the future. I would love to hear about any interesting applications you would like to see covered. I would also love to hear about your creative and quirky ways to improve on the Apps Script AI code that was mentioned.

Artificial-intelligence systems that can churn out fluent text, such as OpenAI's ChatGPT, are the newest darlings of the technology industry. But when faced with mathematical queries that require reasoning to answer, these large language models (LLMs) often stumble. A line parallel to y 4x 6 passes through (5, 10). What is the y-coordinate of the point where this line crosses the y-axis? Although LLMs can sometimes answer these types of question correctly, they more often get them wrong. In one early test of its reasoning abilities, ChatGPT scored just 26% when faced with a sample of questions from the'MATH' data set of secondary-school-level mathematical problems1. This is to be expected: given input text, an LLM simply generates new text in accordance with statistical regularities in the words, symbols and sentences that make up the model's training data.

Generative AI has hit the public imagination in full force during 2022. Perhaps the biggest splash was made by OpenAI's launch of the text-to-image generator DALL-E 2 with its stunning illustrations. Under the guise of generative AI art, code-completion programs are boosting developer productivity by automating repetitive and mundane programming tasks. The world's largest source code host GitHub released its code-completion tool called Copilot in June 2022. It is trained on 45 terabytes of coding data from the GitHub code repository and runs on OpenAI's Codex model.

Even if you haven't tried artificial intelligence tools that can writing essays and poems or conjure new images on command, chances are the companies that make your household products are already starting to do so. Mattel has put the AI image generator DALL-E to work by having it come up with ideas for new Hot Wheels toy cars. Used vehicle seller CarMax is summarizing thousands of customer reviews with the same "generative" AI technology that powers the popular chatbot, ChatGPT. Meanwhile, Snapchat is bringing a chatbot to its messaging service. And the grocery delivery company Instacart is integrating ChatGPT to answer customers' food questions.

Following the hype around OpenAI's ChatGPT conversational agent, the last straw in the recent development of Large Language Models (LLMs) that demonstrate emergent unprecedented zero-shot capabilities, we audit the latest OpenAI's GPT-3.5 model, `gpt-3.5-turbo', the first available ChatGPT model, in the LexGLUE benchmark in a zero-shot fashion providing examples in a templated instruction-following format. The results indicate that ChatGPT achieves an average micro-F1 score of 47.6% across LexGLUE tasks, surpassing the baseline guessing rates. Notably, the model performs exceptionally well in some datasets, achieving micro-F1 scores of 62.8% and 70.2% in the ECtHR B and LEDGAR datasets, respectively. The code base and model predictions are available for review on https://github.com/coastalcph/zeroshot_lexglue.

We introduce ViLPAct, a novel vision-language benchmark for human activity planning. It is designed for a task where embodied AI agents can reason and forecast future actions of humans based on video clips about their initial activities and intents in text. The dataset consists of 2.9k videos from \charades extended with intents via crowdsourcing, a multi-choice question test set, and four strong baselines. One of the baselines implements a neurosymbolic approach based on a multi-modal knowledge base (MKB), while the other ones are deep generative models adapted from recent state-of-the-art (SOTA) methods. According to our extensive experiments, the key challenges are compositional generalization and effective use of information from both modalities.