GPT-4 is here

Advertise | Job board | Community | Investor + Founders connect

Hey folks, yesterday was the biggest AI launch day; GPT-4, Google, Anthropic, Adept, Intercom, Duolingo, Poe, Notion and so many more companies launched or announced something big.

AND we hit over 50k subscribers!!!! in 5 months - what a ride. Thank you all for your support. I’ve got some big news soon 👀

Let’s get to it.

p.s. we’ve got some sponsor slots available in April here.

p.p.s. the email is a little longer today for obvious reasons, don’t freak out, skim, click, and archive, it’s an email, not a to-do list. You do not need to complete every single one!

🤌 Our Picks

Highlights if you've only got 2 minutes

1/

GPT-4 is here

Laaadiies, Gentlemen, and robots, it’s the moment you’ve all been waiting for… Enter GPT-4, the long-awaited update to the model underlying ChatGPT! It has come with lots of ‘mind-blown’ emojis over its additional performance quality, disappointment about OpenAI’s lack of transparency, and the ability to involve both images and text - known as ‘multimodality’. It even came with a developer live stream on YouTube.

According to OpenAI, GPT-4 has hit its best results yet on factuality, steerability, and keeping within guidelines. It’s even taken ChatGPT from being in the 10th percentile to the 90th percentile on the bar, as well as drastic improvements on many other human exams. Beyond the much-expected language improvements, it can generate responses to prompts and documents that include images - all whilst maintaining similar skills.

For those who have played around with prompting, you will probably have found that ‘steering’ the AI towards behaving in a specified style has ended up producing the best results. This is the ‘steerability’ side of the model, and according to the team, GPT-4 is far more attuned to being steered. Wicked.

However, beyond all the excitement is a slight air of disappointment floating around the research community on Twitter, and it’s to do with transparency. True - OpenAI hasn’t expanded on much in their formal publication, mostly using competition and AI safety as their reasoning for keeping almost every detail of the model behind closed doors. You could write a book on this discussion, although if AI safety was really a big concern, wouldn’t you just slow things down a bit? But hey - I’m not complaining!

2/

Google drop their own generative AI tools, with inclusions in Google Docs

Google is like that dog that gets jealous when it sees you stroking another one. As soon as OpenAI dropped the GPT-4 bomb, they dropped their own slick new AI developer tool! Get a load of all of this. The PaLM API is their upcoming tool that provides a way to build apps on top of their best language models; and they’ve coupled it with MakerSuite, the sister tool to help with prototyping ideas and more. They’ve also announced a bunch of new companies and tech partners’ training models with them, including Midjourney!

But the coolest part of all of this is the inclusion of generative AI within their workspace tools a.k.a Google Docs and Gmail. Type out a topic and it’ll generate a draft, allowing you to adjust for tone and style too. Hold your horses though - it’s only available to ‘trusted’ developers, so you’re gonna have to wait!

For the nitty-gritty developers out there, you’ll probably be most interested in the Generative AI App Builder, which you can use to build AI-powered chat interfaces and digital assistants. Things are gonna get a whole lot easier!

3/

Adept get a $350m funding round for their laptop-assistant AI vision

Wow-eeeee that’s a monster funding round! Despite all the SVB mania, the tech industry is clearly still on a roll. At Adept, their vision is centred around allowing AI to work with you on your computer, letting it develop with you and execute operations for you at your request. This could allow your creativity and productivity to run wild. I’m imagining another mouse running around my computer with me. Maybe even multiple all at once working on two things at the same time… The possibilities might be endless with a wad of cash that big.

🛠️ Cool Tools

Product launches, updates and demos

• Collie - Create a delightful search experience in a snap. (link)

• Anthropic AI has made Claude open for all with a waitlist. (link)

• Poe subscriptions - Access to GPT-4 from OpenAI and Claude+ from Anthropic. (link)

• AskNotion - an AI chatbot, powered by your Notion pages. (link)

• Diagram, the generative design company, is using GPT-4 for its products Magician and Genius. (link)

• Intercom’s Fin - Customer service AI chatbot, built on GPT-4. (link)

• CSV analyzer - Copy/paste CSV data, and receive an email with summary, key risks, and follow-up questions. (link)

• Khanmigo - Khan Academy’s GPT4 powered guide for learners and teachers. (link)

• Virtual Volunteer - Answer questions or provide real-time help related to an image. (link)

• Neptyne - Supercharge your spreadsheets with Python and AI. (link)

• Elicit, the AI research assistant, announces their waitlist for their GPT-4 version. (link)

• Autoblocks - Building blocks for deploying LLMs. (link)

• Milo - AI to take care of all the details to keep your family running, so you can focus on love, not logistics. (link)

• Access the OpenAI GPT-4 chat AI model using Node.js with this repl. (link)

• LangFlow — An open source UI for langchain. (link)

• Examples of GPT-4 generating code. (link)

• GPT-4 Chat UI - The easiest way to ship a chatbot built with next.js. (link)

• Transformers chat - 100% open source knowledge-grounded chatbot for the Hugging Face transformers docs. (link)

• Chat, create and compose with the new Bing in the Edge sidebar. (link)

• GPT best - Create a link to share your best ChatGPT conversations. Great for longer conversations that don't fit in screenshots. (link)

• Kissan GPT - A ChatGPT and Whisper-based assistant for underserved agriculture domain of India. (link)

• Moving goalposts - A satirical website about critics always moving their goalposts when a new AI model comes out. (link)

🤓 Miscellaneous

News, podcasts, videos, blogs etc

• Diagram on how AI helps designers be more creative and build ambitious projects. (link)

• Everyone, everywhere, all at once. (link)

• How to use AI to unstick yourself. (link)

• Nabla, a digital health startup, launches Copilot, using GPT-3 to turn patient conversations into action. (link)

• Meet the team developing an open-source ChatGPT alternative. (link)

• Overcoming the challenges in building a CS Bot. (link)

• Adobe’s subtitling AI is good, but this free Resolve add-on has far more potential. (link)

• An AI start-up boomed, but now it faces a slowing economy and new rules. (link, without paywall here)

• Lifesaver or job killer? Why AI tools like ChatGPT are so polarising. (link)

• “Duolingo Max” shows the future of AI education. (link)

• How did I get GPT-4 to design new drugs? (link)

• The implications of today's huge AI announcements. (link)

• Microsoft confirms the new Bing runs on OpenAI’s GPT-4. (link)

• The chat stack, GPT-4, and the near-term future of software. (link)

🎓 Learn

How-to’s and resources

• Building an image recognition app in javascript. (link)

• Use chat in LangChain. (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

Published research papers

• GPT-4 technical report. It’s 98 pages long. (link)

• WHOOPS - Benchmark for common sense evaluation of synthetic images. (link)

• 3DFuse - A novel framework that incorporates 3D awareness into pretrained 2D diffusion models. (link)

• Edit-A-Video - Single video editing with object-aware consistency. (link)

• Editing implicit assumptions in text-to-image diffusion models. (link)

• MeshDiffusion - Score-based generative 3D mesh modelling. (link)

• Eliciting latent predictions from transformers with the tuned lens. (link)

• Simfluence - Modeling the influence of individual training examples by simulating training runs. (link)

• ViperGPT - Visual inference via python execution for reasoning. (link)

• OVRL-V2: A simple state-of-art baseline for IMAGENAV and OBJECTNAV. (link)

• FreeNeRF - Improving few-shot neural rendering with free frequency regularisation. (link)

• MELON - NeRF with unposed images using equivalence class estimation. (link)

📰 Unclassifieds

Short, sponsored links 

• Ben’s Bites is hiring for sales. All contract-based and commissions only (25%). To apply, get someone to pay for an ad slot on sponsor.bensbites.co and say they were referred by you, then email [email protected]

• Inari is hiring a head of AI. Learn more »

• Curai Health is hiring a machine learning engineer. Learn more »

• Foundation Capital is hosting an AI Hackathon in SF on March 22nd. (link)

Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 45k subscribers? 
Advertise | Job board | Community | Investor + Founders connect

🖼 AI Images of the day

Funny memes and pics from around the web

This one has to be here, right?

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