🤖 Last Week's AI Highlights from HN #12

Discover the latest breakthroughs and trends in artificial intelligence

Author

Vasco
August 13, 2023

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

Greetings, all!

As we navigate through another week, here's your fresh rundown on the developments in artificial intelligence. AI continues to make monumental strides, and it's buzzing more than ever. This week, we're bringing a robust collection of thought-provoking and insightful content. We'll be digging into an array of riveting reports, articles, and debates that shed light on the latest happenings in the AI space - from the unveiling of AI tools for diagnosing diseases, remarkable breakthroughs in AI for environmental sustainability, discussions stirring around AI ethics, revolutionary releases, and so much more.


What happened in this GPT-3 conversation?
🤔 It outlines the calculations needed to produce one liter of sunflower oil, and provides information on how the yield of oil per plant can vary with different varieties of sunflower and growing conditions. It also highlights the energy and resources needed to extract and refine the oil, and suggests that it may not be feasible or sustainable to produce 1 liter of oil from a single sunflower plant.
https://chat.openai.com/share/f5341665-7f08-4fca-9639-04201363506e
https://news.ycombinator.com/item?id=37054241

Do Machine Learning Models Memorize or Generalize?
🤔This article explores the training dynamics of a tiny model to understand if Machine Learning models memorize or generalize. It examines the modular addition model and reverse engineers the solution it finds to prove that models can indeed generalize after being trained longer.
https://pair.withgoogle.com/explorables/grokking/
https://news.ycombinator.com/item?id=37076210

GPT-4 can't reason
🤔This paper discusses the nature of reasoning, critiques the current formulation of reasoning problems in the NLP community, and introduces a collection of 21 diverse reasoning problems. It analyzes GPT-4's performance on these problems and argues that the model is currently incapable of reasoning.
https://www.preprints.org/manuscript/202308.0148/v1
https://news.ycombinator.com/item?id=37050257

Fear of AI just killed a useful tool
🤯 This article discusses how fear of AI has caused an incredibly useful tool to be shut down, highlighting how fear of new technology can lead to missed opportunities. It highlights the need for a more realistic and open approach to embracing and understanding AI, and the potential it has for creative people.
https://www.techdirt.com/2023/08/08/the-fear-of-ai-just-killed-a-very-useful-tool/
https://news.ycombinator.com/item?id=37060974

Supermarket AI meal planner app suggests recipe that would create chlorine gas
😐A New Zealand supermarket launched an AI-generated meal planner app to help customers use leftovers. However, the app has generated some unappealing recipes, such as "bleach-infused rice surprise" and "poison bread sandwiches". The supermarket has acknowledged the issue and is working on fine-tuning the bot's controls.
https://www.theguardian.com/world/2023/aug/10/pak-n-save-savey-meal-bot-ai-app-malfunction-recipes
https://news.ycombinator.com/item?id=37072530

Llama from scratch, or how to implement a paper without crying
🤯 Brian Kitano dives into how to implement a paper without tears, walking through his experience of creating a scaled-down version of Llama for TinyShakespeare.
https://blog.briankitano.com/llama-from-scratch/
https://news.ycombinator.com/item?id=37059479

How Zoom’s terms of service and practices apply to AI features
https://blog.zoom.us/zooms-term-service-ai/
https://news.ycombinator.com/item?id=37037196

GPT Unicorn has drawn a unicorn
😃 Tracking changes in GPT-4 by asking it to draw a unicorn every day and measuring the results.
https://gpt-unicorn.adamkdean.co.uk/
https://news.ycombinator.com/item?id=37073701

Fine-Tuning Llama-2: A Comprehensive Case Study for Tailoring Custom Models
🤔 Analyzing the performance of Llama-2 models for various use cases, this blog post provides a comprehensive guide to fine-tuning to achieve significant accuracy improvements compared to GPT-4.
https://www.anyscale.com/blog/fine-tuning-llama-2-a-comprehensive-case-study-for-tailoring-models-to-unique-applications
https://news.ycombinator.com/item?id=37090632

Show HN: Phind V2 – A GPT-4 agent that’s connected to the internet and your code
https://www.phind.com/hn
https://news.ycombinator.com/item?id=37034951

Zoom's TOS Permit Training AI on User Content Without Opt-Out
🤔 Zoom's new terms of service allow them to use user content without opt-out, raising questions about user privacy.
https://stackdiary.com/zoom-terms-now-allow-training-ai-on-user-content-with-no-opt-out/
https://news.ycombinator.com/item?id=37038494

Testing Intel’s Arc A770 GPU for Deep Learning
🤔 Summarizing my experience with Intel's PyTorch extension for Arc GPUs on Ubuntu and WSL, initially there were some headaches but eventually I was able to get it working and test training performance.
https://christianjmills.com/posts/arc-a770-testing/part-2/
https://news.ycombinator.com/item?id=37057332

Llama 2 on ONNX runs locally
😃Summary: Microsoft's open source repository for Llama-2-Onnx, featuring 547 stars, 36 forks, and 3 issues.
https://github.com/microsoft/Llama-2-Onnx
https://news.ycombinator.com/item?id=37082117

Deep Learning Systems
📚 This page provides an overview of the lectures, assignments, staff, and forums for 10-414/714: Deep Learning Systems course at CMU, with lecture videos and slides linked.
https://dlsyscourse.org/lectures/
https://news.ycombinator.com/item?id=37101515

The company behind Stable Diffusion appears to be crumbling into chaos
😱 Summary: The company behind Stable Diffusion, Stability AI, appears to be in turmoil with accusations of a CEO not knowing how to run a business, unpaid bills, and a talent exodus causing issues for the AI firm.
https://futurism.com/the-byte/stability-ai-stable-diffusion-chaos
https://news.ycombinator.com/item?id=37070014

What Is Happening with ChatGPT?
🙃 Exchanging humorous messages between two people about Transhumanism and Christianity, while repeating "YW" (You're Welcome) throughout the conversation.
https://chat.openai.com/share/ca5e0c33-7a19-41a1-8f8a-b146b68e2d1a
https://news.ycombinator.com/item?id=37070838

Show HN: Axilla – Open-source TypeScript framework for LLM apps
🔆 A comprehensive AI framework for TypeScript, built with axilla-io for creating and deploying AI models.
https://github.com/axilla-io/ax
https://news.ycombinator.com/item?id=37034575

To opt out of Grammarly AI training requires business account with 500+ users
😃 Front-End Social is excited to announce AI features in any product they are using.
https://front-end.social/@fox/110846484782705013
https://news.ycombinator.com/item?id=37042779

ML is useful for many things, but not for predicting scientific replicability
🤔 Summary: ML can be useful for many things, but not for predicting scientific replicability due to drawbacks such as small training sample, potential for data distribution changes, and easy gaming of ML models.
https://www.aisnakeoil.com/p/machine-learning-is-useful-for-many
https://news.ycombinator.com/item?id=37089573

Doctor GPT: A Large Language Model That Can Pass the US Medical Licensing Exam
🤖 DoctorGPT is an LLM that can pass the US Medical Licensing Exam, working offline, cross-platform, and keeping your health data private.
https://github.com/llSourcell/DoctorGPT
https://news.ycombinator.com/item?id=37105352

Midjourney API and Diffusion and Dalle2 on Picogen
https://picogen.io/
https://news.ycombinator.com/item?id=37030884

Show HN: PC Builder AI
🤔PC Builder AI provides options to build your PC according to your budget.
https://www.pcbuilderai.com/
https://news.ycombinator.com/item?id=37064044

Show HN: Pykoi – a Python library for LLM data collection and fine tuning
https://www.cambioml.com/pykoi/
https://news.ycombinator.com/item?id=37091122

Show HN: Travel site made with Midjourney, GPT4 and Svelte
😺Discover the world with Meoweler's travel community and explore ~5000 cities across the globe! 🌍
https://meoweler.com
https://news.ycombinator.com/item?id=37045417

Sites scramble to block ChatGPT web crawler after instructions emerge
🤯 Sites scramble to block OpenAI's ChatGPT web crawler after instructions emerge, with restrictions not applying to current models but potentially future versions.
https://arstechnica.com/information-technology/2023/08/openai-details-how-to-keep-chatgpt-from-gobbling-up-website-data/
https://news.ycombinator.com/item?id=37094463

Is ChatGPT a Bit Unsporting?
🤔 Novelist and essayist Walter Kirn is unimpressed with ChatGPT, criticising its bland, homogenised writing and unsporting nature of showboating on demand.
https://www.kmjn.org/notes/chatgpt_unsporting.html
https://news.ycombinator.com/item?id=37069387

Using GPT as an HVAC control system
🤔 Summary of paper exploring the potential of pre-trained large language models to be used as controllers to control HVAC buildings.
https://arxiv.org/abs/2308.03028
https://news.ycombinator.com/item?id=37058364

Author discovers AI-generated counterfeit books written in her name on Amazon
🤯Author discovers AI-generated counterfeit books using her name on Amazon, Amazon requests "trademark registration numbers" and doesn't take them down until the story goes viral.
https://arstechnica.com/information-technology/2023/08/author-discovers-ai-generated-counterfeit-books-written-in-her-name-on-amazon/
https://news.ycombinator.com/item?id=37055909

The Little Book of Deep Learning
 📖 Summary: François Fleuret's Homepage is offering a free phone-formatted pdf and a $8.50 paperback pocket book of a short introduction to deep learning for readers with a STEM background.
https://fleuret.org/francois/lbdl.html
https://news.ycombinator.com/item?id=37085365

PlayHT2.0: State-of-the-Art Generative Voice AI Model for Conversational Speech
🤩 Introducing PlayHT2.0, the state-of-the-art Generative Voice AI Model for Conversational Speech with the addition of emotions, available in closed beta and with realtime speech generation capability.
https://news.play.ht/post/introducing-playht2-0-the-state-of-the-art-generative-voice-ai-model-for-conversational-speech
https://news.ycombinator.com/item?id=37091221

ChatGPT now seems to be embedding ads in its responses
🤔 A peculiar behavior of GPT-4 appending ads to responses is discussed, and the possible implications of this technique are explored.
https://jacobw.xyz/projects/gpt_ads/
https://news.ycombinator.com/item?id=37053601

ChatGPT's odds of getting code questions correct are worse than a coin flip
https://www.theregister.com/2023/08/07/chatgpt_stack_overflow_ai/
https://news.ycombinator.com/item?id=37042223

Generative AI – A curated list of Generative AI projects and services
💡 A curated list of modern Generative Artificial Intelligence projects and services.
https://github.com/steven2358/awesome-generative-ai
https://news.ycombinator.com/item?id=37107101

Medium's Partner Program Changes to Incentivize Human Writing over AI Articles
🤩 Medium introduces new partner program incentives to reward quality writing, expanding to 12 new countries in August 2023.
https://blog.medium.com/new-partner-program-incentives-focus-on-high-quality-human-writing-7335f8557f6e
https://news.ycombinator.com/item?id=37024680

How to Get ChatGPT to Stop Apologizing?
🤔 Summarizing a prompt design question on how to get ChatGPT to stop apologizing excessively.
https://genai.stackexchange.com/questions/177/how-to-get-chatgpt-to-stop-apologizing#1
https://news.ycombinator.com/item?id=37090081

Google launches Project IDX, an AI-enabled browser-based development environment
🤖Google launches Project IDX, a new AI-enabled browser-based development environment, powered by Codey & PaLM 2 models to help developers write code faster and with higher-quality.
https://techcrunch.com/2023/08/08/google-launches-project-idx-a-new-ai-enabled-browser-based-development-environment/
https://news.ycombinator.com/item?id=37052378

News Corp using AI to produce 3k Australian local news stories a week
😃News Corp using AI to produce 3,000 Australian local news stories a week, providing service information such as fuel prices, traffic conditions, and weather.
https://www.theguardian.com/media/2023/aug/01/news-corp-ai-chat-gpt-stories
https://news.ycombinator.com/item?id=37085329

DoD Announces Establishment of Generative AI Task Force
🤖The Department of Defense has established a taskforce to analyze and integrate generative Artificial Intelligence tools across the DoD.
https://www.defense.gov/News/Releases/Release/Article/3489803/dod-announces-establishment-of-generative-ai-task-force/
https://news.ycombinator.com/item?id=37088695

Building search for the post-ChatGPT world
🤔 Building search for the post-ChatGPT world to make Internet search feel like you're being personally guided through the grand total of human knowledge.
https://platform.metaphor.systems/blog/building-search-for-the-post-chatgpt-world
https://news.ycombinator.com/item?id=37037083

Thanks, and see you next week!