TLDR AI 2023-09-21

OpenAI launches DALL路E 3 3锔忊儯, GitHub Copilot Chat now widely available 馃, LLMs in medicine 馃┖

Headlines & Launches

OpenAI launches DALL路E 3 (4 minute read)

DALL路E 3 understands significantly more nuance and detail than previous systems, allowing users to easily translate your ideas into exceptionally accurate images. It will soon be integrated natively into ChatGPT.

GitHub鈥檚 AI-Powered Coding Copilot Is Now Widely Available (1 minute read)

GitHub is expanding the availability of its AI-powered coding chatbot, Copilot Chat. Previously in public beta for business users, Copilot Chat is now available to individual users in Visual Studio and Visual Studio Code.

TRI Is Developing A New Method To Teach Robots Overnight (5 minute read)

The Toyota Research Institute (TRI) is showcasing advancements in research that can teach a robot a new skill overnight. TRI utilizes traditional robot learning techniques and diffusion models similar to generative AI models. Its system has been trained on 60 skills, focusing on versatility in different environments, aiming to make robots that can operate effectively in diverse settings and adapt to new tasks.
Research & Innovation

A Better Way to Understand Hand Movements (22 minute read)

This study introduces RenderIH, a massive collection of ultra-realistic images showing hands in various positions that tackles issues with older datasets that were too simple. It also presents TransHand, a new tool that reads these hand positions very accurately.

Boosting Recommendations Using Large Language Models (13 minute read)

This study looks at how LLMs, which are revolutionizing many tech areas, can make sequential recommendations (like suggesting what song or movie comes next) even better.

Making LLMs Useful in Medicine (11 minute read)

This research focuses on making large language models, which are already great at understanding and generating text, even better at answering medical questions.
Engineering & Resources

Turn expensive prompts into cheap models (GitHub Repo)

OpenPipe is a tool for model distillation based on few shot described behavior via prompting.

A New Approach to Better Understand RGB-D Images (GitHub Repo)

DFormer is a unique way to study both color and depth in images. Unlike older methods, DFormer is designed especially for RGB-D images.

AI companies aren鈥檛 the only ones scraping your data (Sponsor)

Worried your PII might end up in a chatbot? Bad news: data brokers are already selling your details to spammers, scammers, and insurers.

It would take you weeks to chase every broker down鈥

So use Incogni to send hundreds of deletion requests on your behalf, continuously.

50% off with this link. 30-day money back guarantee. Cancel anytime.

Best and safest $6.49 a month you鈥檒l spend.

Retrieval is more than just embedding search (8 minute read)

Language models regularly hallucinate details to factual queries. By retrieving the facts, we can reduce these inaccuracies. However, the naive form of Retrieval Augmented Generation (RAG) of embedding a query and searching against a vector store often performs poorly. This post discusses how to dramatically improve performance with query objects and information retrieval techniques.

How AI can spark both technological and business model innovation (2 minute read)

Truly great innovation comes in pairs. Technology innovation and business model innovation work together to deliver outlier results. AI is the next technology innovation, but have we found its counterpart on the business model side yet?

Generative AI鈥檚 Act Two (15 minute read)

Generative AI鈥檚 first act of developing foundational models and novelty apps is coming to an end. In its place comes Act Two, where generative AI solves human problems end-to-end.

How Google Taught AI To Doubt Itself (6 minute read)

Google's Bard chatbot now offers a "double-check" feature, which evaluates its responses by checking the web for corroborating or contradictory information when prompted by users. This addition aims to enhance accuracy and accountability but may still require users to verify search results and interpretations themselves.
Quick Links

Microsoft Will Lay Out It鈥檚 AI Vision For Windows At Special Event (3 minute read)

Microsoft is planning a special event to unveil its vision for integrating AI into Windows, Microsoft 365, Surface, and more.

Telling AI To 鈥淭ake A Deep Breath鈥 Causes Math Scores To Soar (4 minute read)

Google has developed Optimization by PROmpting (OPRO), a technique to improve the math skills of LLMs using human-style encouragement, such as 鈥渓et鈥檚 think step by step鈥.

OneAudio (Product)

Transform voice recordings into clean summarized notes.
TLDR is a daily newsletter with links and TLDRs of the most interesting stories in startups 馃殌, tech 馃摫, and programming 馃捇!
Join 500,000 readers for one daily email