OpenAI's roadmap

PLUS: More research and asking if AI is really bad.

Hey folks, today we have a bunch of stuff from Open AI; they are shipping like a beast with a new method for improving the mathematical performance of LLMs, future plans of the company, and minor updates.

Then we talk a little about the AI risk and what’s going on there. For the links, there was too much good stuff today that curation took a long time. Check the newsfeed if you want more.

Let’s get to it.

From our sponsor:

Nyckel: The end-to-end custom classification API

Train and deploy image and text classification functions without writing code or dealing with infrastructure. Nyckel automatically captures important data points in production – enabling you to iterate on your data to improve your model.

Try Nyckel: Lightning-fast machine learning for developers and product managers

🤌 Our Picks

Highlights if you've only got 2 minutes

1/

So Open AI dropped another technical blog, and they have trained a model to achieve a new state-of-the-art in mathematical problem-solving. They are doing it by introducing supervision into the step-by-step process of solving complex mathematical problems.

This is big for two reasons: 1) the models get better at maths. Duh. and 2) The step-by-step process with human supervision creates hope for solving alignment problems in LLMs.

2/

Sam Altman talked about Open AI’s plans in a talk and revealed a bunch of stuff:

  • OpenAI is heavily GPU limited at present

  • Cheaper GPT-4 is a near-term priority. One million token context length is also possible

  • Plugins don’t have a product-market fit yet

  • A stateful API that remembers conversation history

Also, they have a self-serve security portal (link) and plugins are searchable now. (link)

3/

Dangers of AI - reading list.

Following the AI statement yesterday, there was a long reading list of pros and cons of both sides. Replit’s founder tweeted that the experts kinda jump from “this is smart” to “existential threat” very quickly. I made a small reading list on this topic, collecting different views:

  • Is avoiding extinction from AI really an urgent priority? (link)

  • AI Is not an arms race. (link)

  • If we’re going to label AI an ‘extinction risk,’ we need to clarify how it could happen. (link)

Personally, I believe that there are some clear dangers, but instead of collecting signatures, focusing on research is more important.

4/

Anna-Sofia from Contrary has been writing about hard tech topics like climate change, chips, fusion, and more for a while, and now it is a standalone publication. I think many of you might like it outside of the AI stuff. Check it out, Anna does a great job at it.

🛠️ Cool Tools

Product launches, updates and demos
  • Caption It - Write witty photo captions with AI that are great for social media. (link)

  • InkyMM - First commercially available image-to-text multimodal model. (link)

  • Recraft - Generate vector art, 3D images and more. (link)

  • Siit - AI assistant for employee support. (link)

  • Play I spy with ChatGPT and vision AI. (link) Writeup here.

  • Studyflow - Learn by quizzing yourself with AI. (link)

  • Paige - Rewrite your web presence in a page generated in 30 seconds. (link)

  • Hyper Online - Build AIs for companionship, motivation, and fun. (link)

  • ChatOCR - ChatGPT plugin that reads text from PDFs, including scans and handwriting. (link)

  • WaifuChat - Your AI anime girlfriend. (link)

  • Gamma - Let AI finish your presentation for you, shape your content, add visuals, and even rewrite sections for you. (link) - disclaimer, Ben is an investor.

  • Embedding Dev - A playground to compare different text embedding models from companies like OpenAI, CohereAI and GoogleAI. (link)

  • The open-source vector toolkit for Postgres from Supabase. (link) - disclaimer, Ben is an investor.

🤓 Miscellaneous

News, podcasts, videos, blogs etc
  • Macaw-LLM - Multi-modal language modeling with image, video, audio, and text integration. (link)

  • SelFee - Iterative self-revising LLM empowered by self-feedback generation. (link)

  • Large sequence models for software development activities. (link)

  • GILL - Generating images with multimodal language models. (link)

  • Teaching fewer words to large language models might help them sound more human. (link)

  • UAE's Falcon 40B has waived royalties on its use for commercial and research purposes. (link)

  • Open sourcing the Scale Venture partners chatbot. (link)

  • Behind the Scenes - How LangChain calculates OpenAI pricing. (link)

  • New AI features from Instacart (link) and Snapchat. (link)

  • Harnessing AI's potential to improve equity. (link)

  • AI intensifies the fight against paper mills that churn out fake research. (link)

  • In pursuit of a better book - How AI fixes e-books' sins. (link)

  • AI anthology - We are an information revolution species. (link)

  • Japan goes all in - Copyright doesn’t apply to AI training. (link)

    Check out BB News for funding news, research and more →

🎓 Learn

How-to’s and resources
  • Andrew Ng launches 3 new generative AI courses. (link)

  • Train your own private ChatGPT model for the cost of a Starbucks coffee. (link)

  • Using the Hugging Face LLM inference container with Amazon SageMaker. (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.

📰 Unclassifieds

Short, sponsored links 
  • If Mark Zuckerberg and Tim Cook could agree on anything, it would be this business newsletter. Check it out subscribe for free here.

  • AI tools suffer from Last Mile gap. Lychee offers stunning short-form videos which are AI-powered, human-polished. Used by 20VC & Zendesk

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 100k subscribers? 
Advertise

🖼 AI Images of the day

Funny memes and pics from around the web

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

How was today's email?

Login or Subscribe to participate in polls.

Reply

or to participate.