🤖 Last Week's AI Highlights from HN #18

Discover the latest breakthroughs and trends in artificial intelligence

Author

Vasco
September 29, 2023

Welcome back to HN AI Highlights, your weekly digest of all things AI!

Greetings, all!

We are beginning to roll out new voice and image capabilities in ChatGPT
🤩 ChatGPT, an AI assistant, is now able to see, hear, and speak. It offers a new, more intuitive type of interface which allows users to have a voice conversation or show ChatGPT what they are talking about. This is available for Plus and Enterprise users, and is powered by a new text-to-speech model, capable of generating human-like audio from just text and a few seconds of sample speech. Additionally, users can now show ChatGPT images and use the drawing tool for more specific guidance.
https://openai.com/blog/chatgpt-can-now-see-hear-and-speak
https://news.ycombinator.com/item?id=37642335

First Impressions with GPT-4V(ision)
🤔 A review of the new OpenAI GPT-4V(ision) which is a multimodal model that accepts text and images as input. It is tested with input such as a computer vision meme and currency and the results are discussed.
https://blog.roboflow.com/gpt-4-vision/
https://news.ycombinator.com/item?id=37673409

Be My Eyes’ AI assistant starts rolling out
🤩 Be My Eyes is excited to announce the launch of their AI assistant, Be My AI, which is rolling out to hundreds of thousands of iOS users over the coming weeks. Be My AI can assist users with tasks such as detailed descriptions of photos, as well as answering questions related to the photos. The AI can also be used for tasks such as social media posting and messaging. Be My AI is not intended to replace a white cane, guide dog, or other mobility aids.
https://www.bemyeyes.com/blog/announcing-be-my-ai
https://news.ycombinator.com/item?id=37673300

AI is fundamentally ‘a surveillance technology’
🤔 Signal's Meredith Whittaker believes that AI is a fundamentally a surveillance technology and is often used to entrench and expand the surveillance business model. She also noted that the data that underlies these systems is frequently organized and annotated by the same people who it can be used against. However, there are still some positive uses of AI, like Signal's face blur feature which helps prevent intimate biometric data from being revealed.
https://techcrunch.com/2023/09/25/signals-meredith-whittaker-ai-is-fundamentally-a-surveillance-technology/
https://news.ycombinator.com/item?id=37656091

Workers AI: Serverless GPU-powered inference
🤩 Workers AI is a serverless GPU-powered inference service from Cloudflare, designed to make AI more accessible to developers. It features a curated set of popular, open source models for a wide range of inference tasks (text generation, ASR, translation, text classification, image classification, and more).
https://blog.cloudflare.com/workers-ai/
https://news.ycombinator.com/item?id=37674097

Show HN: Generative Fill with AI and 3D Open with GitHub Desktop
🤖 Generative fill in 3D. fill3d.ai is an open source project that is licensed under MIT and allows users to create 3D objects. It provides tools for automation and devOps as well as code review and issue tracking.
https://github.com/fill3d/fill
https://news.ycombinator.com/item?id=37695530

Show HN: Get your entire ChatGPT history in Markdown files
😃A Python script that allows you to quickly and easily extract and format ChatGPT conversations data from JSON files to well-structured markdown files with YAML metadata headers, all happening locally.
https://github.com/mohamed-chs/chatgpt-history-export-to-md
https://news.ycombinator.com/item?id=37636701

TinyML and Efficient Deep Learning Computing
🤩 This MIT HAN Lab course focuses on AI computing, transforming the future with TinyML and efficient deep learning computing. It will be offered in Fall 2023 and includes useful links to the Prof. Song Han and MIT HAN Lab Homepages, as well as MIT accessibility information.
https://efficientml.ai/
https://news.ycombinator.com/item?id=37620507

Show HN: Rapidpages – OSS alternative to vercel's v0
🤩This project is used to generate React and Tailwind components using AI. It is an open source project hosted on GitHub and it offers a license of AGPL-3.0. It has 633 stars and 26 forks.
https://github.com/rapidpages/rapidpages
https://news.ycombinator.com/item?id=37614177

Show HN: Carton – Run any ML model from any programming language
🤔Summary: Carton is an open-source API that allows users to run ML models from any programming language without the need to modify the original model, avoiding error-prone conversion steps. It is implemented in Rust and supports x86_64 Linux and macOS, aarch64 Linux and macOS, and WebAssembly.
https://carton.run
https://news.ycombinator.com/item?id=37682286

Ollama for Linux – Run LLMs on Linux with GPU Acceleration
🤩 Summary: Ollama for Linux is now available with GPU acceleration enabled, and supports CPU-only, as well as small to powerful GPUs. New features and bug fixes have been added.
https://github.com/jmorganca/ollama/releases/tag/v0.1.0
https://news.ycombinator.com/item?id=37661755

The Cambridge Law Corpus: A corpus for legal AI research
🤓 This paper introduces the Cambridge Law Corpus (CLC), a corpus for legal AI research containing over 250 000 court cases from the UK. Annotations on case outcomes for 638 cases were done by legal experts and used to train and evaluate case outcome extraction with GPT-3, GPT-4 and RoBERTa models. The paper also includes legal and ethical discussion of the potentially sensitive nature of the material.
https://arxiv.org/abs/2309.12269
https://news.ycombinator.com/item?id=37627129

Getty made an AI generator that only trained on its licensed image
😀Getty Images is introducing a new AI-generated image tool called Generative AI by Getty Images that is trained only on their vast library of licensed images. The tool creates realistic-feeling images and comes with copyright indemnification, meaning users can publish the images commercially without legal issues. The tool has some limitations, such as not being able to generate images of real people, and users can access the tool through the Getty Images website.
https://www.theverge.com/2023/9/25/23884679/getty-ai-generative-image-platform-launch
https://news.ycombinator.com/item?id=37643456

OpenAI and Jony Ive in talks to raise $1B from SoftBank for AI device venture
🤔 OpenAI and Jony Ive are in talks to raise $1bn from SoftBank to create an AI device venture. The Financial Times provides readers with informed decisions and global reporting to help them stay up-to-date with significant corporate, financial and political developments around the world. New customers can try unlimited access for 1€ for 4 weeks.
https://www.ft.com/content/4c64ffc1-f57b-4e22-a4a5-f9f90a7419b7
https://news.ycombinator.com/item?id=37690080

Causality for Machine Learning (2020)
🤔 This report covers the intersection of causality and machine learning, exploring how causal inference can improve prediction, enable interventions, and facilitate robustness. It also covers the prototype created to demonstrate the application of causality to machine learning.
https://ff13.fastforwardlabs.com/
https://news.ycombinator.com/item?id=37663523

ChatGPT can now search the web in real time
😃 OpenAI announced that the ChatGPT feature "Browse with Bing" can now search the web in real time, with direct links to sources and up-to-date information. Currently only available for Plus and Enterprise subscribers, the feature will be rolled out to all users soon. Microsoft and Google already offer similar services for their AI chatbots. Instructions are provided to use the feature, but it's a bit slow. OpenAI also added the ability to browse the internet within its ChatGPT iOS app in late June but quickly pulled it.
https://www.theverge.com/2023/9/27/23892781/openai-chatgpt-live-web-results-browse-with-bing
https://news.ycombinator.com/item?id=37681047

The Llama Ecosystem: Past, Present, and Future
🤩A research project with the potential to benefit billions of people, the Llama Ecosystem has seen massive success over the past seven months. With over 30 million downloads of Llama-based models, cloud usage on major platforms, innovation from startups, crowd sourced optimization, and hardware support, the community has embraced the project and created incredible momentum and development.
https://ai.meta.com/blog/llama-2-updates-connect-2023/
https://news.ycombinator.com/item?id=37680851

Searchable Database of the 183,000 Pirated Books Meta, et al., Used to Train AI
🤔 Summary: A data set of 191,000 books, mostly published in the past 20 years, was acquired and used without permission to train generative-AI systems by Meta, Bloomberg, and others. This has sparked copyright infringement lawsuits from authors such as Sarah Silverman, Michael Chabon, and Paul Tremblay. A search tool has been created to allow authors to see if their work is included in the data set.
https://www.theatlantic.com/technology/archive/2023/09/books3-database-generative-ai-training-copyright-infringement/675363/
https://news.ycombinator.com/item?id=37685313

Project Gutenberg has implemented one of the worst AI fears of striking actors
🤔 Project Gutenberg's use of AI to produce thousands of audiobooks voiced by AI has sparked fear among actors who are on strike in the US, as the technology could replace their work. The AI reader is not as versatile as human actors, but its free availability, scalability, and quick process could be attractive to some. SAG-AFTRA is demanding protections from AI-generated digital replicas of actors.
|https://qz.com/project-gutenberg-ai-to-ebooks-audiobooks-1850856297
https://news.ycombinator.com/item?id=37626326

Jony Ive and OpenAI’s Altman reportedly collaborating on mysterious AI device
🤔 Jony Ive and OpenAI’s Altman are reportedly collaborating on a mysterious AI device, however there are no details on what it might be. It has caused speculation as to whether it will be a reimagined smartphone or something else entirely.
https://arstechnica.com/information-technology/2023/09/jony-ive-and-openais-altman-reportedly-collaborating-on-mysterious-ai-device/
https://news.ycombinator.com/item?id=37681663

Show HN: ChatGPT for Med-School and Healthcare
😃This is a summary of an AI-driven chatbot called Radiant Chat that helps provide medical reference information.
https://chat.radiantai.health/
https://news.ycombinator.com/item?id=37620043

AI language models can exceed PNG and FLAC in lossless compression, says study
🤯A DeepMind study shows that AI language models can exceed PNG and FLAC in lossless compression, suggesting that AI models may be able to effectively understand and represent data. The implications of this could lead to a better understanding of general intelligence.
https://arstechnica.com/information-technology/2023/09/ai-language-models-can-exceed-png-and-flac-in-lossless-compression-says-study/
https://news.ycombinator.com/item?id=37691535

Llama 2 Long
🤔 This paper presents a series of long-context LLMs that support effective context windows of up to 32,768 tokens and offers an in-depth analysis of the individual components of the method. It reports consistent improvements on most regular tasks and significant improvements on long-context tasks over Llama 2.
https://arxiv.org/abs/2309.16039
https://news.ycombinator.com/item?id=37698604

Show HN: Use ChatGPT with Apple Shortcuts
🤩 COPILOT is an AI assistant integrated into Apple Shortcuts, designed for all Apple devices, allowing you to access ChatGPT directly in your favorite apps. It can be activated from anywhere, including Siri, and offers features such as deep ecosystem integration, advanced web capabilities, detailed progress updates, follow-up talk, dynamic personas, and natural language math calculations.
https://meetcopilot.app/
https://news.ycombinator.com/item?id=37671821

Cloudflare and Meta Collaborate to Make Llama 2 Available Globally
🎉 Cloudflare and Meta have collaborated to make Llama 2 available globally, providing privacy-first, local inference to all developers. Through Cloudflare's platform and Data Localization Suite, developers are able to run and deploy their own LLMs with data localization built in. This will help companies earn trust with their customers and ensure powerful AI is accessible to all developers around the world.
https://www.cloudflare.com/press-releases/2023/cloudflare-and-meta-collaborate-to-make-llama-2-available-globally/
https://news.ycombinator.com/item?id=37692679

GPT Excel
🤩 AI-Powered Excel formula Generator with features to generate powerful spreadsheet formulas, explain spreadsheet formulas, generate SQL queries, and automate tasks with generated scripts. Ideal solution for individuals and businesses to streamline their spreadsheet processes. Free and Pro plans available with various features.
https://gptexcel.uk/
https://news.ycombinator.com/item?id=37630630

Knuth's 20 Questions for ChatGPT
🤔 A curious Don Knuth asked 20 questions to chatGPT, testing orthogonal skills. He got interesting responses which he decided to post online after lightly editing them.
https://www-cs-faculty.stanford.edu/~knuth/chatGPT20.txt
https://news.ycombinator.com/item?id=37634792

Show HN: Summary Cat, a YouTube Video Summary Generator
🤩 This summarizer tool created by Bing Dai in Vancouver, Canada, quickly and accurately summarizes videos in English. It provides a short and concise summary of the videos for easier understanding.
https://www.summarycat.com
https://news.ycombinator.com/item?id=37617288

AI-generated naked child images shock Spanish town of Almendralejo
😱 AI-generated images of naked young girls have been circulating on social media, shocking the small Spanish town of Almendralejo. Local police are investigating the case and have identified 11 boys aged 12-14 as suspects. A support group of 28 victims and their families has been formed and one mother is using her social media profile to bring the issue into public debate.
https://www.bbc.co.uk/news/world-europe-66877718
https://news.ycombinator.com/item?id=37632061

A poor man's guide to fine-tuning Llama 2
🐦 A concise summary of Duarte O. Carmo's guide on fine-tuning Llama 2: Learn how to quickly train an LLM with just a few dollars and one hour of your time using the Axolotl toolkit to streamline the process.
https://duarteocarmo.com/blog/fine-tune-llama-2-telegram
https://news.ycombinator.com/item?id=37657237

GPT-4V(ision) system card [pdf]
🤔 A long string of numbers and letters that appears to be some type of computer code with no relation to AI. Output: N/AB.
https://cdn.openai.com/papers/GPTV_System_Card.pdf
https://news.ycombinator.com/item?id=37642712

New AI experiences across our family of apps and devices
🤩 Meta is introducing new AI experiences across its family of apps and devices, including AI stickers, an advanced conversational assistant, and AI with unique personalities played by cultural icons and influencers. Over time, Meta will make AIs for businesses and creators available, as well as its AI studio for people and developers to build their own AI.
https://about.fb.com/news/2023/09/introducing-ai-powered-assistants-characters-and-creative-tools/
https://news.ycombinator.com/item?id=37678431

Oops Google Search caught publicly indexing users’ conversations with Bard AI
😣 Google recently caught indexing conversations with AI startup, Bard AI, prompting the firm to issue an apology. This article investigates the incident and discusses Google's response.
https://venturebeat.com/ai/oops-google-search-caught-publicly-indexing-users-conversations-with-bard-ai/
https://news.ycombinator.com/item?id=37668735

The AI revolution is rotten to the core [video]
🤔 The text discusses the implications of the AI revolution, suggesting that it has a potentially negative effect on society. It also provides information about YouTube and its associated companies.
https://www.youtube.com/watch?v=-MUEXGaxFDA
https://news.ycombinator.com/item?id=37629753

Game of Thrones creator and other authors sue ChatGPT-maker for ‘theft’
🤬 Writers and artists, including George R.R. Martin, John Grisham and Jodi Picoult, are suing OpenAI, a ChatGPT-maker, for ‘theft’ of their copyrighted works. The Authors Guild is leading the class-action lawsuit and OpenAI has responded by saying it respects authors’ rights and is in conversation to find mutually beneficial ways to work together. OpenAI has also previously asked a court to dismiss two similar lawsuits.
https://www.aljazeera.com/news/2023/9/21/openai-sued
https://news.ycombinator.com/item?id=37638817

The Pentagon’s Budget Is So Bloated That It Needs an AI Program to Navigate It
🤯 The Pentagon developed an AI program, codenamed GAMECHANGER, to help make sense of its own "byzantine" and "tedious" bureaucracy and manage its $816.7 billion budget - a stark wake-up call for lawmakers who throw more money at the Department of Defense than it even asks for.
https://theintercept.com/2023/09/20/pentagon-ai-budget-gamechanger/
https://news.ycombinator.com/item?id=37628962

OpenAI is reportedly raising funds at a valuation of $80B to $90B
🤑OpenAI's potential secondary-market valuation, which may be boosted to $90 billion, is being discussed. This follows the company's recent funding of $300 million at a valuation of $29 billion. ChatGPT is the AI company's popular generative assistant, and is set to become more interactive. OpenAI expects to reach $1 billion in revenue in 2023.
https://techcrunch.com/2023/09/26/openai-is-reportedly-raising-funds-at-a-valuation-of-80-billion-to-90-billion/
https://news.ycombinator.com/item?id=37668211

Yann LeCun on comparing the number of parameters of PaLM 2 vs. GPT 4
https://twitter.com/ylecun/status/1706545305762582580
https://news.ycombinator.com/item?id=37663866

Thanks, and see you next week!