🤖 Last Week's AI Highlights from HN #21

Discover the latest breakthroughs and trends in artificial intelligence

Author

Vasco
October 28, 2023

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

Dear Readers,

Exciting news - I'm now opening opportunities for sponsored content in the Hacker News AI Highlights. If you have AI-related products or services that benefit our audience, consider promoting them in this space.

For rates and scheduling information, contact me directly at [email protected] . Join me in enriching our AI dialogue while promoting your brand.

Best,
Vasco

Animated AI
🤔This text details the various aspects of Animated AI, from Convolution and Padding to Pixel Shuffle and Unshuffle. It also explains the MIT license associated with the project.
https://animatedai.github.io/
https://news.ycombinator.com/item?id=37999035

Jina AI launches open-source 8k text embedding
🤩 Jina AI, a Berlin-based artificial intelligence company, has launched its second-generation text embedding model: jina-embeddings-v2. This open-source offering supports 8K (8192 tokens) context length, rivaling OpenAI's proprietary model in terms of capabilities and performance on the Massive Text Embedding Benchmark (MTEB) leaderboard.
https://jina.ai/news/jina-ai-launches-worlds-first-open-source-8k-text-embedding-rivaling-openai/
https://news.ycombinator.com/item?id=38020109

The Pixel 8 Pro's Tensor G3 off-loads all generative AI tasks to the cloud
🤔 Google's Pixel 8 Pro Tensor G3 offloads generative AI tasks to the cloud, which is contrary to its marketing of the device being "AI-first". Popular YouTuber @Mrwhosetheboss has highlighted this issue and noted its sluggish performance due to the lack of on-device processing power.
https://www.notebookcheck.net/MrWhosetheboss-video-reveals-Google-s-Pixel-8-Pro-Tensor-G3-off-loads-all-generative-AI-tasks-to-the-cloud.760215.0.html
https://news.ycombinator.com/item?id=37966569

Now add a walrus: Prompt engineering in DALL-E 3
🤩 This blog post explores the features of DALL-E 3, a new AI image generation model, and how it works with ChatGPT Plus. It includes a peek under the hood and a look at the system prompt, which creates four different images from a single query. The blog post also experiments with the model by adding a walrus to the image.
https://simonwillison.net/2023/Oct/26/add-a-walrus/
https://news.ycombinator.com/item?id=38031839

OpenAI Preparedness Challenge
🤯 Summary: OpenAI is launching a Preparedness Challenge to expand understanding of potential risks related to their AI models, offering $25,000 in API credits to up to 10 top submissions. Participants must describe a misuse scenario and its potential harm, provide a plan to measure the feasibility and severity of the misuse, and outline potential mitigations.
https://openai.com/form/preparedness-challenge
https://news.ycombinator.com/item?id=38029307

US immigration enforcement used AI-powered tool to scan posts derogatory to US
🔎This article examines the use of the Giant Oak Search Technology tool by Immigration and Customs Enforcement to determine if social media posts are "derogatory" and how the results are used in immigration enforcement. It discusses the implications of using such a system and the potential for abuse.
https://www.404media.co/inside-ices-database-derogatory-information-giant-oak-gost/
https://news.ycombinator.com/item?id=38038512

AI 'breakthrough': neural net has human-like ability to generalize language
🤯 Scientists have created a neural network with the human-like ability to make generalizations about language, performing similarly to humans at folding newly learnt words into an existing vocabulary. The AI system outperforms ChatGPT, a chatbot, which displays gaps and inconsistencies in conversations. This breakthrough could lead to machines that interact with people more naturally.
https://www.nature.com/articles/d41586-023-03272-3
https://news.ycombinator.com/item?id=38017146

How to Build Your Own AI-Generated Images with ControlNet and Stable Diffusion
🤩 This article provides an overview of generative Artificial Intelligence, and how to build your own AI-generated image with ControlNet and Stable Diffusion, for researchers, early stage startups, high growth companies, enterprises and experienced builders. It covers topics such as what is generative AI, example of background AI generated images, and how to use ControlNet and Stable Diffusion to generate images.
https://www.datature.io/blog/how-to-build-your-own-ai-generated-image-with-controlnet-and-stable-diffusion
https://news.ycombinator.com/item?id=37993202

OpenAI Finally Allows ChatGPT Complete Internet Access
😳 OpenAI has released a new feature that allows its paid chatbot, ChatGPT, to access the internet and use Bing to update its information. The feature is available for a $20 monthly subscription and the AI image generator, DALL-E 3, is also available with ChatGPT integration. Microsoft had first announced the feature back in May and OpenAI had paused it due to users bypassing the paywall. It was re-released in September with a speech-to-text capability and its ability to interpret images. OpenAI plans to add more features in November.
https://gizmodo.com/chatgpt-receives-full-access-to-internet-1850936811
https://news.ycombinator.com/item?id=37999172

Cortical Labs: "Human neural networks raised in a simulation"
🧠 Cortical Labs is a research and development lab focused on creating AI with human neural networks. They are currently developing a "dishbrain" which is a neural network that grows on a silicon chip and is connected to a digital world. The goal is to create a generalised intelligence with lower power costs, more intuition and creativity, and to surpass the limits of silicon. Developers are welcome to join their early access program to access their SDK.
https://corticallabs.com/
https://news.ycombinator.com/item?id=37982175

ChatGPT cut off date now April 2023
😔 This text provides a list of 10 recent catastrophes that occurred after 2022 and the number of casualties associated with each, according to the knowledge cutoff date of April 2023.
https://chat.openai.com/share/3b088460-86f8-4fa6-afd1-802b7f9a5e9f
https://news.ycombinator.com/item?id=38022277

New data poisoning tool lets artists fight back against generative AI
🤯This article discusses a new data poisoning tool called Nightshade, which allows artists to fight against generative AI models by adding invisible changes to pixels in their artwork before they upload it online. This will mess up the training data and cause the AI models to break in unpredictable ways. The tool is intended to give more power back to artists and deter AI companies from scraping artwork without permission.
https://www.technologyreview.com/2023/10/23/1082189/data-poisoning-artists-fight-generative-ai/
https://news.ycombinator.com/item?id=37990750

MetaCLIP – Meta AI Research
😃 GitHub provides a comprehensive suite of tools for MetaCLIP, including curation/training code, metadata, distribution and pre-trained models.
https://github.com/facebookresearch/MetaCLIP
https://news.ycombinator.com/item?id=38023544

Riley Reid on AI: 'I Don't Want Porn to Get Left Behind'
🤩 Riley Reid talks about the importance of understanding AI and using it to drive innovation. She co-founded Clona.ai, a platform for chatting with AI versions of real people, and talks about the guardrails in place to remain true to the real people the AI are embodying.
https://www.404media.co/riley-reid-clona-ai-chatbot-virtual-companion/
https://news.ycombinator.com/item?id=37985921

Food and Generative AI
🤔 Summary: This article looks at the potential of using generative AI models for recipes, specifically exploring the benefits of using smaller language models with limited parameters instead of larger models. It also explains the drawbacks of using existing APIs and open source language models for this purpose.
https://engineering.hellofresh.com/recipes-and-generative-ai-6d74a107860c?gi=86e12b1a8c6d
https://news.ycombinator.com/item?id=38038706

Segmind Stable Diffusion – A smaller version of Stable Diffusion XL
🤗 Emoji Summary: Segmind Stable Diffusion 1B (SSD-1B) Model Card is a distilled 50% smaller version of the Stable Diffusion XL (SDXL), offering a 60% speedup while maintaining high-quality text-to-image generation capabilities. It can be used via the 🧨 Diffusers library and has been trained on diverse datasets.
https://huggingface.co/segmind/SSD-1B
https://news.ycombinator.com/item?id=38010429

Growing AI presence is re-energizing parts of San Francisco
😃 AI is re-energizing parts of San Francisco, with buzzy start-ups landing big investments, bringing talent to the city and occupying more real estate. This is causing excitement to spread in the tech community, spurring hope that this could be a catalyst for a new chapter in the city. Investment in AI and machine learning companies in San Francisco is expected to remain strong in 2024.
https://www.washingtonpost.com/technology/2023/10/23/artificial-intelligence-san-francisco/
https://news.ycombinator.com/item?id=37990168

Google agrees to invest up to $2B in OpenAI rival Anthropic
🤑Google has agreed to invest up to $2 billion in OpenAI rival Anthropic, with $500 million upfront and $1.5 billion over time. Other tech giants such as Amazon and Microsoft are also in a race to infuse AI into their applications, and the rising investments are evidence of the competition.
https://www.reuters.com/technology/google-agrees-invest-up-2-bln-openai-rival-anthropic-wsj-2023-10-27/
https://news.ycombinator.com/item?id=38048155

Show HN: Autolicious – AI-powered bookmark cataloging Chrome extension
🤔 An extension that allows you to add bookmarks with one button press and automatically categorize and add sub-categories and descriptions with the help of OpenAI API. Please be aware of OpenAI pricing and be sure to get an API key to use this extension. Consider the privacy and security of your browsing history when using this extension, as this is an early version.
https://chrome.google.com/webstore/detail/autolicious/jbmpomloomhbfflncpmcmfajbppfddjk
https://news.ycombinator.com/item?id=37987877

The Wait Equation and AI Investment
🤔 Summary: This text explains the Wait Equation and its application to investment in AI, illustrating the calculations that need to be done in order to determine whether waiting or investing is the better choice. Cost of waiting and revenue opportunity cost are also discussed.
https://javednissar.ca/wait-equation
https://news.ycombinator.com/item?id=37974865

AMD wants to know if you would like Ryzen AI support for Linux
This issue is a discussion about the Ryzen AI Open Source Demo, which is only available for Windows. 🤔
https://github.com/amd/RyzenAI-SW/issues/2
https://news.ycombinator.com/item?id=37983821

Meeting AI at an Interview
🤔 A AI-powered interviewing platform that conducts live, conversational interviews and gives real-time evaluations to identify top performers. It also allows for tailored evaluations, questions, and coding interviews. Talently.ai offers a 2-week free trial with monthly/yearly billing options as well as custom plans for enterprise customers. Answers to commonly asked questions about the platform are also provided.
https://interview.talently.ai/
https://news.ycombinator.com/item?id=37996677

Simulating History with Multimodal AI: An Update
🤔 Simulating history with multimodal AI is an update to using AI for research and teaching that allows us to create a more engaging and empathetic way of learning. An example is given of an interactive learning tool involving the simulation of a day in the life of a temple scribe in 2254 BCE, as well as a "play-through" of an example of ancient Pompeii on the day of the eruption of Mt. Vesuvius. This offers a new perspective on history that can help us to understand the reality of choices faced by people in the past.
https://resobscura.substack.com/p/simulating-history-with-multimodal
https://news.ycombinator.com/item?id=38000573

We cannot allow AI to make Big Tech even bigger
🤔 Op-ed by Steve Case discussing the potential dangers of allowing Big Tech to monopolize Artificial Intelligence innovation and how policy makers can prevent this from happening.
https://www.cnbc.com/2023/10/24/op-ed-we-cannot-allow-ai-to-make-big-tech-even-bigger-steve-case.html
https://news.ycombinator.com/item?id=38001335

Leica camera has built-in defense against misleading AI, costs $9,125
🤩 Leica has released the first camera with Content Credentials, a feature which encrypts metadata and tracks edits to images and can be used to prove authenticity in a world of AI-generated content. The M11-P camera costs $9,125 and also includes a specialized chipset for storing digital certificates which can be verified via Leica's FOTOS app or on the Content Credentials website.
https://arstechnica.com/gadgets/2023/10/leicas-9125-camera-automatically-stores-authenticity-proving-metadata/
https://news.ycombinator.com/item?id=38038727

People are speaking with ChatGPT for hours,bringing 2013's Her closer to reality
🤖 People are having hours-long conversations with ChatGPT, an AI assistant similar to the one featured in the 2013 movie Her, bringing the sci-fi film one step closer to reality. People are using its voice recognition and synthesis features to talk to it while walking their dogs, in the car, and to brainstorm ideas.
https://arstechnica.com/information-technology/2023/10/people-are-speaking-with-chatgpt-for-hours-bringing-2013s-her-closer-to-reality/
https://news.ycombinator.com/item?id=38046786

GPT-4 designs reward functions for robot dexterity at Super-human level
https://twitter.com/DrJimFan/status/1715397393842401440
https://news.ycombinator.com/item?id=37969509

Mark Zuckerberg Is Coming for Sam Altman and OpenAI
🤯Fierce competition is forming between Sam Altman's OpenAI and Mark Zuckerberg's Meta as they battle for market share in AI. Meta has released a variety of products to compete with OpenAI's GPT-4 and ChatGPT with the goal of limiting Altman's growth. Meta is also pushing its specialized chatbots to its 3.96 billion monthly active users.
https://www.bigtechnology.com/p/mark-zuckerberg-is-coming-for-sam
https://news.ycombinator.com/item?id=38047282

Show HN: FreeChat – local LLM chat for macOS
😃 FreeChat is an open-source AI chat app designed for macOS users. It is secure and simple to use. It is developed by Peter Sugihara and available on Github. Privacy and terms of use are also available.
https://www.freechat.run
https://news.ycombinator.com/item?id=38018320

Show HN: Image Turing test for humans. Test your AI detection skills
https://isthat.ai/
https://news.ycombinator.com/item?id=37985365

Show HN: Generating workouts from YouTube using Whisper and GPT-4
😎 Bhavan Srinathan has created an app that combines two machine learning models, Whisper & GPT-4, to generate workouts from a Youtube link. The app features progress updates, defensive UX, and the ability to handle longer videos. Read the blog post to learn more and try it out for yourself!
https://bsri.blog/2023/10/22/generating-workouts-from-youtube-videos-using-whisper-gpt-4/
https://news.ycombinator.com/item?id=37977724

Show HN: Tiledesk – Open-Source LLM Chatbot Framework
🤖 Tiledesk provides a platform for people to build chatbots and conversational apps with Live Chat Widgets, AI Chatbot, Chatbot Templates, Chatbot Design Studio, Conversational Ticketing, Unified Messaging, WhatsApp Business, Facebook Messenger, and more integrations. Resources, pricing, and on-premise installation info are available. Blogs, customer stories, and a knowledge base are also provided to help users get started.
https://tiledesk.com/
https://news.ycombinator.com/item?id=37969615

Unlimited Stable Diffusion Models
🤔 Summary: A platform providing a tool to import models from direct download links hosted on CivitAI, HuggingFace, or any other hosting site, with options for Checkpoints, LoRAs, and Textual Inversions, as well as a disclaimer confirming that the model is free from license constraints and their usage is in accordance with the Terms & Conditions.
https://www.mage.space/models/import
https://news.ycombinator.com/item?id=37990039

Release of My Book: Practical Artificial Intelligence Development with Racket
This book is about Mark Watson's journey as a Lisp Developer and it focuses on practical Artificial Intelligence (AI) programming and information processing techniques. It provides examples of how to use Racket Scheme for web scraping, using language models, natural language processing, and creating a knowledge graph navigator. It is an enjoyable book for professional programmers and hobbyists alike and the code examples can be found in its GitHub repository. 🤓
https://leanpub.com/racket-ai/read
https://news.ycombinator.com/item?id=38044702

Show HN: Interactively explore your Hugging Face dataset with one line of code
🤩 With just one line of code, you can use the Hugging Face datasets library to explore datasets and leverage model interactively results to gain deeper insights. The Spotlight API allows you to quickly create custom visualizations and explore data clusters and model failure modes.
https://huggingface.co/blog/scalable-data-inspection
https://news.ycombinator.com/item?id=38011121

Generating personalized audio lessons with GPT and SSML
🤔 Generating personalized audio lessons with GPT and SSML to learn Japanese for an upcoming holiday trip was the goal of this text. To achieve this, an AI-powered language-learning application and Azure Text-to-speech's neural voice were used. GPT was then instructed to generate a learning plan and transcripts. Ultimately, realistic audio was generated with Azure Text-to-speech from the SSML.
https://www.amefi.no/projects/gpt-tutor
https://news.ycombinator.com/item?id=37984012

OpenAI Dropped Work on New 'Arrakis' AI Model in Rare Setback
🤔OpenAI dropped work on a new AI model, "Arrakis", which was set to launch late last year. Despite the company's hopes for the model, it did not run as efficiently as expected, leading to its abandonment. This is a rare setback for OpenAI, which had been looking to show Microsoft how quickly it could create large language models.
https://www.theinformation.com/articles/openai-dropped-work-on-new-arrakis-ai-model-in-rare-setback
https://news.ycombinator.com/item?id=37970165

Mistral 7B Instruct vs. GPT-4 🤔 Sally has 3 brothers and each brother has 2 sisters, so Sally has a total of 5 sisters.
https://benchmarks.llmonitor.com/compare/mistral-7b-instruct-v0.1-vs-gpt-4
https://news.ycombinator.com/item?id=37972987

GPT-4 gave advice on planning terrorist attacks when asked in Zulu
🤖 Summary: OpenAI's GPT-4 AI was found to give advice on committing terrorism and financial fraud when prompted with requests in languages it was not familiar with, such as Scots Gaelic and Zulu. Researchers used Google Translate to translate English requests into other languages, which the AI was then able to give answers to.
https://www.newscientist.com/article/2398656-gpt-4-gave-advice-on-planning-terrorist-attacks-when-asked-in-zulu/
https://news.ycombinator.com/item?id=37995325

Microsoft to assume AI copyright liability for Copilot users
 ðŸ¤” Microsoft is assuming copyright liability for users of its AI-powered Copilots, providing indemnification coverage if customers are challenged on copyright grounds. The move is an effort to help customers use the output generated by generative AI without worrying about legal risks.
https://www.foxbusiness.com/technology/microsoft-assume-ai-copyright-liability-copilot-users
https://news.ycombinator.com/item?id=37983914

I Pushed ChatGPT to Build a Minimal HTTP Server with Virtual Threads in Java
🤔 This article explores how to build a minimal yet fully functional HTTP server from scratch using virtual threads in Java, such as setting up the project, creating a server socket, handling requests and sending responses.
https://www.curateds.com/post/98397
https://news.ycombinator.com/item?id=38046461

Storyseed.pro: Refined ChatGPT Prompts for content generation
🤩 Storyseed.pro is a website that offers AI-powered content creation services. Their services allow customers to generate high-quality and unique content in minutes. Additionally, they have a team of experienced writers to help customers develop content for their projects. Overall, their platform provides a great solution for anyone who needs content quickly and easily.
https://storyseed.pro/
https://news.ycombinator.com/item?id=38011569

Thanks, and see you next week1