TLDR AI 2023-11-01

AlphaFold is an amazing model for science. The new work by Isomorphic Labs (a spin off from Google) extends it beyond just proteins and improves overall performance.

A federal judge dismissed most claims in a landmark lawsuit where artists accused generative AI art generators of unauthorized image use, pinpointing deficiencies in copyright infringement accusations against Midjourney and DeviantArt.

OpenAI's co-founder and chief scientist Ilya Sutskever has shifted his focus from building the next model like GPT to figuring out how to avoid future artificial superintelligence from behaving undesirably. He believes the reality of AI surpassing human intelligence is imminent and anticipates technologies that will allow humans and AI to merge. Sutskever and his team at OpenAI are actively working on "superalignment", a set of procedures to control future AI technology.

The study introduces MPVSS, a method for segmenting video content that reduces computational load by focusing on key frames and then predicting masks for other frames based on these key ones.

The study assesses how well GPT-4 with Vision (GPT-4V) answers questions related to medical images from various sources, such as X-rays and CT scans.

Researchers have found ways to use diffusion models to make long-term plans, but these plans aren't always realistic or safe. This study introduces a method to fix those unreliable plans, using a tool called the "restoration gap".

Researchers dived deep into the visual encoders used in Multi-modal Large Language Models (MLLMs) and discovered that certain features from CLIP and DINO models are especially effective for detailed visual tasks. They then introduced COMM, a strategy that combines the strengths of both models.

The study introduces PUCA, a new self-supervised denoising approach that maintains J-invariance, a critical aspect for effective denoising.

This project introduces TESTA, a method designed to speed up the process of understanding long videos by combining similar frames and patches. Using TESTA, researchers managed to greatly reduce the computational load and improve performance in matching paragraphs to videos and answering questions about long videos.

AI is poised to dramatically improve the success of phishing attacks, which is already used in 90% of cyberattacks.

This article discusses the influence of Gen Z on various industries and how their behavior is causing seismic waves in the market. The author emphasizes the importance of understanding Gen Z's habits, preferences, and values, as this generation drives significant changes in consumer behavior. By examining the social, technological, and economic aspects of Gen Z's lives, businesses can better anticipate their needs and adapt their offerings to cater to this influential demographic.

Death of a Flywheel discusses the decline of business flywheels, which represent self-sustaining momentum that drives company growth, due to factors including market saturation, evolving customer preferences, competitive threats, regulatory changes, and misaligned incentives. The article emphasizes the importance of recognizing and addressing these threats to ensure the continued health of a company's flywheel.

Alibaba has unveiled an enhanced AI model, Tongyi Qianwen 2.0, that aims to compete with global tech behemoths like Amazon and Microsoft, particularly in generative AI applications similar to ChatGPT.

Generative AI tools for photo editing and design.

Labelbox is a data-centric AI platform for building intelligent applications. As the Foundation Models Lead, you will lead research and development for applying foundation models to real world problems for customers like Walmart and Adobe. Learn more.

AI expert and Google Brain co-founder Andrew Ng suggests that Big Tech companies are amplifying fears about AI risks to stifle competition and prompt strict regulations that could hinder the open-source community.