TLDR AI 2023-09-27

OpenAI raising more funding 💰, AI’s $200B question ❓, semantic image editing 🖼️

Headlines & Launches

Getty launches commercially viable image model (6 minute read)

Powered by Nvidia, Getty launches a commercial image generation service with strong commercial guarantees.

OpenAI Is Raising Funds At A Valuation Of $80B to $90B (1 minute read)

OpenAI is in discussions to possibly sell shares in a move that would boost the company’s valuation from $29 billion to somewhere between $80 billion and $90 billion.

YouTube is going all in on AI (3 minute read)

YouTube unveiled new AI tools for creators, including AI-generated photo and video backgrounds, topic suggestions, and music recommendations. A highlight is Dream Screen for AI-generated backgrounds in YouTube Shorts. This shift may reshape how creators design content, with AI-driven insights influencing content focus and an increased prevalence of AI-generated material on the platform.
Research & Innovation

Making Better Virtual Views of Images (24 minute read)

Neural radiance fields (NeRF) are great for creating virtual views but struggle with shiny or see-through objects. This new method, called refractive-reflective field, fixes this by better handling light changes and producing clearer images of such tricky objects.

A More Consistent Image Editing (3 minute read)

Most current tools for semantic image editing, where images are changed based on their meaning, struggle with blending new objects smoothly into scenes. This new method improves this by smartly understanding both fully seen and partially hidden objects.

A New Tool to Fix Broken Videos (19 minute read)

Over the last ten years, there have been big improvements in fixing damaged videos using special technologies. However, many of these tools aren't perfect at fixing common video issues seen in things like live streaming. This study created BSCV, a new collection of tools and videos to help fix these problems even better.
Engineering & Resources

Useful Annotated Papers (GitHub Repo)

A great resource for annotated papers you may find useful.

LAION OpenLM for medium sized models (11 minute read)

OpenLM is a library and framework for training medium-sized language models up to 7B parameters. It is designed with GPU speed and hackability in mind. Models trained on 1.25T tokens are available as validation of convergence.

Category theory and Machine Learning (GitHub Repo)

Category theory is the abstract study of relationships. It is typically used as an additional layer of abstraction in mathematics. That gives it powerful tools to study machine learning as well.

What’s next for LLMs in production? (Sponsor)

The MLOps Community is running its third LLMs in Production virtual conference. This is the place to be if you want to go deep on the latest developments in gen AI. Join now to hear from real practitioners who have hands on experience incorporating LLMs into real-life products, rather than Twitter demos.

See the full list of speakers and register (FREE)


OpenAI’s GPT-4 With Vision Still Has Flaws (3 minute read)

Although OpenAI has made strides in preventing GPT-4 with vision from breaking CAPTCHAS and estimating people’s age based on age and race, it still sometimes struggles to make inferences and is still prone to hallucinating.

Why Silicon Valley’s Biggest AI Developers Are Hiring Poets (7 minute read)

AI giants are now hiring poets and writers in a bid to boost the literary quality of generative writing tools.

AI’s $200B Question (4 minute read)

The surge in Generative AI, exemplified by Nvidia's success, underscores a vast demand for GPUs and AI training, leading to significant investments in the sector. As major tech companies drive this growth, the challenge for startups is to shift focus from infrastructure to creating tangible end-customer value.
Quick Links

How To Use AI to Brainstorm A Billion-Dollar Business Idea (3 minute read)

Guest author Dan Kraemer for Crunchbase News goes into his methodology around leveraging AI to brainstorm the next big idea.

Cardinal (Product)

An AI-powered backlog enriching your features with customer feedback and revenue data.

The CIA Is Developing An AI Chatbot (3 minute read)

The CIA and other US intelligence agencies are developing an AI chatbot to assist agents in sifting through vast amounts of information. The chatbot, which will be trained on publicly available data, aims to provide sources with its answers, but concerns arise over what counts as public data and potential privacy issues.
TLDR is a daily newsletter with links and TLDRs of the most interesting stories in startups 🚀, tech 📱, and programming 💻!
Join 500,000 readers for one daily email