Conversational answer engine - huh?!
PLUS: The AI application layer and moats.
Hey folks, it's Friiiiyay! 🤢 - don't be the person to ever say that.
We had our first official meetup yesterday in London and I loved it. Balderton Capital hosted ~200 of us in their offices and provided fab food and drinks. A big TY to them!
And a massive thank you to those who came, I spoke to as many of you as possible and it was so awesome having that many people come together all because of a bloody email I write sat in my pyjamas at home 😳. You're real, which is a relief! And I feel incredibly lucky that you give up space in your inboxes each day for me to pop in, give you a load of tabs, ruin your productivity then leave to do it again the next day. It's a joy. Honestly.
We've had some folks talk about hosting the next one and sponsoring another hackathon so stay tuned for that.
Okay ben you're babbling, let's get to it.
p.s. those gyoza were awesome, wish I'd thrown a load in my bag for the train home!
🤌 Ben's Picks
Queue "The Boys Are Back In Town" on Spotify...Perplexity have come out swinging on their third (?) release. Perplexity Ask: World’s first conversational search engine.
Basically, like ChatGPT where you chat back and forth, now you can search (like you would on Google) and Perplexity will provide answers plus sources for those answers and any related links. You don't have to do the work of searching through those search results that came from your search. Phew.
It prompts you to ask follow up questions to dig a little deeper. Further down the rabbit hole we go! Perplexity say "We’re working towards a conversational answer engine: i.e. ChatGPT + latest references pulled from a search engine." You can read their announcement thread here.
Artificial General Intelligence is so yesterday - it’s all about Artificial Domain Intelligence these days. In fact, Scenario, the generative gaming company who just received $6m, are taking it one step further. They’ve built a platform on which users can tune existing generative models on their particular art styles with a focus on gaming.
This seems to be the way AI is progressing. With the realisation that tools such as ChatGPT require extensive context to be most effective, it seems that domain-level branches will grow from its root to produce many highly-effective, domain-specific AIs.
A great piece by Rex Woodbury, AI in 2023: The Application Layer Has Arrived, talking about the world of AI and what that future might look like. It's interesting because he touches on business models we already know (SaaS, Marketplaces, etc) having a sort of second-coming with AI at the helm - what will that look like? He also touches on moats and where companies have them (or not).
🛠️ Cool Tools
Check for AI - Detect AI written text in essays, emails, and more! (link)
Image Mixer - Mix up concepts in multiple images and words to generate novel pictures! (link)
Promptable - The ultimate workspace for prompt engineering. (link)
Luma Shares and Embeds - Luma's interactive NeRFs, panoramas, video renders in your own websites. (link)
Hippocratic AI - Free search tool for peer-reviewed medical knowledge. (link)
Olli AI - Your team’s AI data analyst. Answer questions and create charts 10x faster. (link)
Clever voice assistant demo hacked together with GPT-3 and Siri. (link)
A template from Langchain quickstart guide to take user input, construct a prompt, and send it to the LLM. (link)
ChatGPT + Google calendar integration using Langchain pull request. (link)
AI is taking on ever larger puzzles. Real-world applications of machine learning are just beginning. (link)
Who owns the generative AI platform? (link)
Deep Papers episode 1 - InstructGPT - Aligning language models to human intention. (link)
Linus Lee: At the boundary of machine and mind. (link)
The AI hype cycle with Gary Marcus - The Prof G Pod. (link)
The next generation of content creation and AI. (link)
Is AI the new crypto? (link)
I left my job as a Goldman Sachs MD to work in AI. (link)
Bring still images to life with Runway Academy. (link)
Generative AI and long-term memory for LLMs. (link)
Transformer models: an introduction and catalogue — 2023 edition. (link)
Universal image segmentation with Mask2Former and OneFormer. (link)
Deep learning tuning playbook - Actually getting good results with deep learning. (link)
Self supervision does not help natural language supervision at scale. (link)
From English to more languages - Parameter efficient model reprogramming for cross lingual speech recognition. (link)
Multiview compressive coding for 3D reconstruction. (link)
Learning-rate free learning by distance adaptation. (link)
RecolorNeRF: Layer decomposed radiance field for efficient colour editing of 3D scenes. (link)
Self supervised learning from images with a joint embedding predictive architecture. (link)
AtMan: Understanding transformer predictions through memory efficient attention manipulation. (link)
A few Ben's Bites community meetups going on in the next week:
Have a product, service, job, event, newsletter, app, book, movie, tool, or anything you'd like to share with over 20k subscribers? Get a short unclassified ad included here.
🖼 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?