Update on MSFT + OpenAI

PLUS: the simplest, fastest repository for training

Roll up, roll up! Welcome one and all to the most spectacular email on earth! The AI circus is in town and we've got a lineup of acts that will leave you in awe.

Me getting ready to write the email:

Let's get to it.

ps. a bunch of meetups going on very soon - see if there's one in a town near you.

🤌 Ben's Picks

1/

An update on the AI love story of the year, Microsoft & OpenAI; “Once $92 billion in profit plus $13 billion in initial investment are repaid (to Microsoft) and once the other venture investors earn $150 billion, all of the equity reverts back to OpenAI.” This was from the article in Fortune: "Inside the structure of OpenAI’s looming new investment from Microsoft and VCs". Gary Marcus commented about it here.

Then The Information published its piece, "Microsoft + OpenAI: Inside Tech’s Hottest Romance", which revealed a few more interesting points.

  • proposing nonprofit status again after profit is paid back

  • potentially increasing investor cap. Instead of hard cap, 20% increase per year. So investors can still get VC-like returns

  • Altman hasn’t got equity in the for profit arm. Unclear if this deal changes that

I wonder what today will bring for this story...

2/

Andrej Karparthy, previously Director of AI at Tesla, and OpenAI, launched (not on purpose) his project: NanoGPT - The simplest, fastest repository for training/fine tuning medium-sized GPTs. In 2 weeks, Andrej is releasing a video where he 'will build nanoGPT from scratch and aspire to spell everything out'. It'll be the last video from his series 'Neural Networks - Zero to hero'. (link)

3/

For those of you who don't know, I work at Zapier. I'm now focusing on people who want to build with AI + Zapier for anything! My teammate, Reid, wrote this awesome blog post: How to write an effective GPT-3 prompt.

p.s. I'm running a build program for using AI and Zapier. If you want access, pop your email here. I'll be letting users into the Slack regularly. We'll show what we've built and how, give feedback for products and feature/integration requests.

🛠️ Cool Tools

  • Mason's AI Assistant - Revolutionising data analytics by generating answers optimised for your team. (link)

  • Manna - AI autocomplete across your MacOS apps. (link)

  • Metavoice Studio - AI voice editing platform for creators. (link)

  • Penelope AI - Markdown editor powered by AI writing assistant. (link)

  • Vivid - AI wingman to build your web app. (link)

  • ChatGPT Detector - Bilingual ChatGPT detector to detect AI generated text and answers. (link)

  • Godly AI - A super simple way to add context to GPT3 for relevant and accurate completions. (link)

  • Campbell - Write better performance reviews for your team members. I genuinely had to write performance reviews on Tuesday and was close to tears looking at the blank page. It's so difficult! Not any longer... this was sent to me via our Discord and the founder, , even did a review of this newsletter using the tool 😂 . I like it! (link)

🤓 Miscellaneous

  • ChatGPT takes my midterm and gets a D. (link)

  • Unofficial implementation of Tune-A-Video: One shot tuning of image generation models for text to video generation. (link)

  • Buildspace Notes - Stories, tips, tricks, tutorials and learnings from top builders. (link)

  • How language models could be misused for disinformation and how to mitigate those risks. (link)

  • Generative Voice AI - This voice doesn't exist. (link)

  • Seek AI secures $7.5m to build generative AI platform for data. (link)

  • Giggles - GPT-4 generates superhuman catfishing output to steal this man’s wife over WeChat. (link)

  • Lay of the land and what might happen next in AI in 2023. (link)

  • Elicit deploys a new model for abstract summaries trained using RLAIF. (link)

  • This take on AI image tech by an artist is the most lucidly honest and hilarious one yet. (link)

  • CNET is quietly publishing entire articles generated by AI. (link)

🎓 Learn

  • Mickey, CEO of FlairAI, showing how to do AI-powered product photoshoots. (link)

  • Train custom models on your own art style with Runway Academy. (link)

  • An introduction to Unity ML agents. (link)

  • Generative question answering with OpenAI's GPT-3.5 and Davinci. (link)

  • Simple ML for Sheets - Build machine learning models directly within Google sheets. (link)

👋 Too many links?! I created a database for all links mentioned in these emails. Refer 1 friend using this link and I'll send over the link database.

🔬 Research

  • Hold onto your diamond pickaxes 'cause Google just announced DreamerV3 - the first general and scalable RL algorithm for playing games. It outperforms previous approaches and it's the first to collect diamonds in Minecraft from scratch without any help from human data or curricula. Mastering Diverse Domains through World Models. (link)

  • This paper takes a deep dive into OpenAI's text-davinci-003 and the previous versions of GPT to see if they're ready for the big leagues. And let me tell you, the results are iiiiinteresting. Text-davinci-003 scored a 14.4% accuracy on a sample regulation exam, but the latest and greatest generation of GPT-3, it hit an accuracy of 57%."GPT as knowledge worker - A zero-shot evaluation of its capabilities with the CPA examination. (link)

  • Latent Blending: Incredibly smooth transition videos between two prompts within seconds. (link)

  • Model pruning approaches can be used to further optimise ConvNets trained through neural architecture search. (link)

  • ORBIT: A unified simulation framework for interactive robot learning environments. (link)

  • Geometry-biassed transformers for novel view synthesis of an object. (link)

  • Learning without forgetting from multiple tasks arriving sequentially using HyperTransformers. (link)

  • Learning to exploit temporal structure for biomedical vision–language processing. (link)

  • Transformer-based LLMs are computationally universal when augmented with an external memory. (link)

  • Does localization inform editing? Surprising differences in causality-based localization vs. knowledge editing in language models. (link)

  • ReVISE - First universal audio-visual speech enhancement model powered by SSL. (link)

  • The Bananarama Conjecture - Ethics and use capabilities of ChatGPT for academic finance research. (link)

🖼 AI images of the day

🤗 Share Ben's Bites

Send this to 1 AI-curious friend and receive my AI project tracker database!

or copy/paste this link: https://bensbites.beehiiv.com/subscribe?ref=PLACEHOLDER

⭐️ How did we do?

How was today's email?

Login or Subscribe to participate in polls.

⭐️ REAL REVIEWS

Reply

or to participate.