Discover more from Towards AI Newsletter
This AI newsletter is all you need #42
What happened this week in AI by Louie
AI agents were in the spotlight in the AI and LLM community this week because of projects like Auto-GPT. The concept is that language models such as GPT-4 can “self-prompt” or “auto-prompt” — that is, they can generate and execute their own prompts based on initial input. Through this approach, multiple instances of these models can be assigned different roles and work collaboratively towards achieving a common goal, such as executing a series of tasks, reviewing and fixing mistakes, and even developing and debugging their own code. While current AI Agent projects are still in the early stages and require further refinement, we see tremendous potential for AI Agents. Furthermore, by granting language models access to other tools, such as web search engines, mathematical engines, and information retrieval systems, we believe that many new and emergent capabilities can be unlocked.
After witnessing the rapid pace of progress and continuous newsflow of new models in NLP, we were delighted to see some of these concepts being applied to advance Computer Vision, as demonstrated by Meta’s new Segment Anything Model (SAM). The open-source release of this project is particularly noteworthy, and we hope it will help to reduce the costs and barriers to entry for many computer vision tasks. The Segment Anything Model serves as a promptable foundation model for image segmentation, allowing it to learn a general notion of objects and generate masks for any object in images or videos. Moreover, it can be linked to other tasks and other generative models. For example, it can be connected to stable diffusion to build amazing image-editing end-to-end models like Grounded Segment Anything recently shared on GitHub. We believe this is just the first of many to integrate SAM into a larger and “real-world applicable” project.
- Louie Peters — Towards AI Co-founder and CEO
According to documents obtained by TechCrunch, Anthropic aims to raise up to $5 billion in the next two years to challenge competitor OpenAI using its new AI model, Claude-Next. The company also plans to enter more than a dozen major industries.
Meta has introduced Segment Anything, which aims to democratize segmentation. The Segment Anything Model is a versatile, promptable model trained on a diverse dataset. It is available under a permissive open-source license (Apache 2.0).
The AI program ChatGPT was found to have generated a false claim of sexual harassment without a factual basis. This incident highlights the biases and dangers of using AI to tackle disinformation. ChatGPT relied on a cited article that was never written and quoted a statement that was never made by the newspaper.
Poe, created by the Q&A site Quora, enables users to experiment with cutting-edge AI technologies through a mobile interface. With Poe, users can build their own chatbots using prompts and an existing bot as the foundation. Additionally, users have the option to pay for access to more advanced bots powered by new language models, such as GPT-4 and Claude+.
Last week’s topic was the concerns surrounding recent advancements in systems such as ChatGPT and GPT-4. With significant AI news, including the letter and the Times Magazine post, there is a growing need for greater transparency on the inner workings of these large AI systems, including their training and evaluation processes.
Three 5-minute reads/videos to keep you learning
This blog post demonstrates the various steps involved in training a LlaMa model to answer questions on Stack Exchange using RLHF. The approach combines Supervised Fine-tuning (SFT), Reward/preference modeling (RM), and Reinforcement Learning from Human Feedback (RLHF). By employing these techniques, HuggingFace has released the StackLLaMA model.
This report offers a valuable compilation of practical tips to consider when training a larger machine learning model. It discusses crucial topics such as initialization and experimentation strategies and offers various examples of useful tools that may come in handy.
AI has the potential to impact around 300 million jobs worldwide. The legal industry is among the fields where AI shows promise in automating language-based jobs, but human guidance remains essential. This article presents an unbiased analysis of how AI is likely to impact various industries.
This is a thoughtfully curated and high-quality list of machine learning (ML) papers that one can regularly refer to. With so much hype around ML in online discussions, it is important to go back to the source and understand the current trends.
Mito shared the concrete lessons they learned from launching their AI feature. The biggest lesson was that adding a chatbot that generates code is easy, but actually figuring out how that code interacts with your existing product is much harder.
Papers & Repositories
This is an experimental open-source project aimed at making GPT-4 fully autonomous. Auto-GPT is an app showcasing the capabilities of GPT-4 in autonomously managing businesses to increase their net worth.
This is a chatbot powered by GPT-4 and LangChain designed for handling large PDF documents. With this technology, users can create a question-answering chatbot that can process several large PDF files. The tech stack used includes LangChain, Pinecone, Typescript, Openai, and Next.js.
HuggingGPT is a framework that utilizes LLMs to connect various AI models in machine learning communities to solve AI tasks. It employs ChatGPT to plan tasks upon receiving a user request, select models based on their available function descriptions in Hugging Face, execute each subtask using the chosen AI model and summarize the response according to the execution results.
This paper offers a comprehensive survey of ChatGPT and GPT-4 and their potential applications across various domains. The study aims to provide insights into ChatGPT’s capabilities, potential implications, and ethical concerns while also offering direction for future advancements in this field.
This is a survey of 141 arXiv papers that compared ChatGPT to other NLP models. It revealed that ChatGPT’s performance was lower than anticipated. The reasons for this included suboptimal utilization, biased results, and incomplete multilingual evaluation.
Enjoy these papers and news summaries? Get a daily recap in your inbox!
The Learn AI Together Community section!
Announcing Our Discord Job Board
We have launched a new job board on the Learn AI Discord community that allows any company to share their job openings directly with our community, and you can apply within Discord. Simply scroll up the channel to view all the job offers. This job board is powered by our friends at @Freeflow, and once you share a job, you can also share it with other partnered servers to reach more potential candidates. We hope you find this resource helpful.
If you have a job opening to share and are looking for candidates, please take a look at the job board here and consider sharing it with our community.
Weekly AI Podcast
In this week’s What’s AI podcast, Louis Bouchard interviewed Jérémy Cohen, the founder, and CEO of Think Autonomous. This interview delves into the self-driving car industry and the different machine learning-related roles within it, including a discussion on how to become a self-driving car engineer. The conversation also covers topics such as LiDAR technology and its functionality, the comparison between using LiDAR versus cameras on self-driving cars, and demystifying industry myths like the $200,000 salary for a computer vision engineer. Listen on YouTube, Spotify, or Apple Podcasts!
Meme of the week!
Meme shared by DrDub#0108
Featured Community post from the Discord
rmarquet#2048 has shared a GitHub repository containing a collection of artificial intelligence projects, that features detailed documentation and code. It covers various topics and techniques in the field of AI, including computer vision and reinforcement learning. Check it out here and support a fellow community member. Share your feedback in the thread here!
AI poll of the week!
TAI Curated section
Article of the week
This article examines the disparities between machine learning in research environments and production environments. The implementation of machine learning differs significantly depending on the context in which it is used. Furthermore, it delves into the components of production machine learning and the challenges it poses, to provide a better understanding of the complexities involved.
Our must-read articles
If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.
Interested in sharing a job opportunity here? Contact email@example.com.
If you are preparing your next machine learning interview, don’t hesitate to check out our leading interview preparation website, confetti!