Last Week in AI #216: Video-generating AI based on Stable Diffusion, Stack Overflow to charge for LLM training data, Google combines Google Brain and DeepMind, and more!

Nvidia showcases text-to-video based on Stable Diffusion, Stack Overflow to charge companies that use its data for LLM training, DeepMind becomes Google DeepMind once more to compete with Msft/OpenAI

Apr 24, 2023

Align your Latents: High-Resolution Video Synthesis with Latent Diffusion Models

Researchers have developed a new method for generating high-resolution videos. The method, called Latent Diffusion Models (LDMs), works by compressing high-quality data into a lower-dimensional space and then creating videos in that space, before “decompressing” this “video” of latent representations back into videos. This allows the creation of high-quality videos without the need for excessive computing power. In the study, the scientists applied LDMs to video generation and focused on two real-world applications: in-the-wild driving simulations and creative content creation through text-to-video modeling. The results, showing state-of-the-art performance in real driving videos and efficient text-to-video modeling, could open new opportunities for personalized content creation in the future.

Our take: Many research teams have now released prototypes for text-to-video AIs, and what’s so special about this Nvidia work is that it is based on an open-source text-to-image model, Stable Diffusion, pointing out a way that folks can build video models without the expensive training required for text-to-image models. Unfortunately, Nvidia has yet to release the code and model of LDMs, but if the last few months are any indication, an open-source clone will probably be posted on Github sooner or later.

Stack Overflow Will Charge AI Giants for Training Data

Stack Overflow, a forum for computer programming help, plans to charge large AI developers such as OpenAI and Google for access to its 50 million questions and answers later this year. This move to seek compensation for companies using its data follows a similar decision by Reddit. Stack Overflow's CEO, Prashanth Chandrasekar, described the additional revenue as vital for attracting users and maintaining high-quality information for future chatbots. Large-scale AI projects have traditionally used training data scraped from the web without cost, but this could increase AI development expenses and extend timelines for profitability.

Our take: This is probably a sign of things to come. With AI chatbots directly delivering answers to users, the websites that generated the training data for these chatbots will see less and less traffic, and as a consequence less and less revenue. Paying sources of high-quality training data seems like a reasonable move, but it remains to be seen how this affects the dynamics of LLM data collection (e.g. do non-commercial or academic entities have to pay? how does this affect the license/copyright of the text/code produced?).

Google to combine AI research units Google Brain, DeepMind

DeepMind has once again become Google DeepMind. In response to competitive pressures in the AI world, Google has merged its AI research division Google Brain with DeepMind. According to Alphabet CEO Sundar Pichai, "combining all this talent into one focused team, backed by the computational resources of Google, will significantly accelerate our progress in AI." Demis Hassabis will lead the unit as CEO of Google DeepMind, while Jeff Dean's role at Google will be elevated to Chief Scientist.

Our Take: It's telling that Sundar Pichai's April update that announces Google DeepMind mostly focuses on the pace of progress and competitive pressures and how the merge will accelerate progress. The desire to develop AI systems "safely and responsibly" is stated in the letter, though it's clearly far from a primary motivation here. I think an interesting facet of this dynamic is that the companies locked in this AI race are truly locked: we're in a game theory situation where no one can afford to slow down unless everybody else does, and that doesn't seem likely. Sam Altman, for his part, is very open to regulators stepping in at this stage, and I suspect that at least a few tech leaders wouldn't categorically object to a coordinated slowdown. That being said, I think a slowdown is incredibly unlikely. In an ideal world, we'd speed up safety research as much as possible—this is also not a simple order, but would require massive investment in directing talent towards these problems and paying them competitively, funding different safety research directions, and plenty more. I hope we see substantial increases in funding for safety research, but the current race we're in seems like it will be the status quo for some time.

Other News

Research

Meet MiniGPT-4: An Open-Source AI Model That Performs Complex Vision-Language Tasks Like GPT-4 - "MiniGPT-4 uses an advanced LLM called Vicuna as the language decoder, which is built upon LLaMA and is reported to achieve 90% of ChatGPT’s quality as evaluated by GPT-4. MiniGPT-4 has used the pretrained vision component of BLIP-2 (Bootstrapping Language-Image Pre-training) "

Machine Learning an Efficient Method for Reducing Cardiovascular Disease Risk - "New UCLA research suggests that a novel machine-learning technique known as "causal forest" was about five times more efficient than the current clinical practice of treating patients with high blood pressure."

DINOv2: State-of-the-art computer vision models with self-supervised learning - "DINOv2 is able to take a video and generate a higher-quality segmentation than the original DINO method. DINOv2 allows remarkable properties to emerge, such as a robust understanding of object parts, and robust semantic and low-level understanding of images."

Researchers developed an innovative robotic link with multimodal perception to help make human-robot interactions safer and more convenient - "In recent years, robots have become incredibly sophisticated machines capable of performing or assisting humans in all tasks. The days of robots functioning behind a security barrier are long gone, and today we may anticipate robots working alongside people in close contact."

Spacecraft-inspired system enhances quadruped agility and balance - "Researchers in Carnegie Mellon University's Robotics Institute (RI) have designed a system that makes an off-the-shelf quadruped robot nimble enough to walk a narrow balance beam -- a feat that is likely the first of its kind."

Model that uses machine learning methods and patient data at hospital arrival predicts strokes more accurately than current system - "A recent investigation utilized machine learning techniques and information accessible at the time of admission to the hospital to fashion a prototype that foretells strokes more precisely than existing models."

Machine learning to examine the genetic underpinnings of cardiac fibrosis at scale - "We developed a machine learning model to quantify cardiac fibrosis (which is associated with cardiovascular disease) using cardiac MRI data from 41,505 UK Biobank participants."

ChatGPT is still no match for humans when it comes to accounting - "Massive crowd-sourced study comes from 327 co-authors at 186 institutions from 14 countries"

RedPajama, a project to create leading open-source models, starts by reproducing LLaMA training dataset of over 1.2 trillion tokens - "You can find instructions on how to install and use the dashboard on Github. Having reproduced the pre-training data, the next step is to train a strong base model."

Machine Learning Detects Social Grooming Behavior in Baboons - "Scientists from Swansea University and the University of Cape Town have tracked social grooming behaviour in wild baboons using collar-mounted accelerometers."

A.I. takes center stage at HIMSS health conference - "Debates over artificial intelligence and its role in health care took center stage at the HIMSS Global Health Conference in Chicago this week, where more than 35,000 physicians, executives, engineers and health-care workers convened to discuss the latest advancements in health and technology."

Hydrogen’s Hidden Phase: Machine Learning Unlocks the Secrets of the Universe’s Most Abundant Element - "Putting hydrogen on solid ground: simulations with a machine learning model predict a new phase of solid hydrogen. Hydrogen, the most abundant element in the universe, is found everywhere from the dust filling most of outer space to the cores of stars to many substances here on Earth."

Researchers at Stanford Introduce Gisting: A Novel Technique for Efficient Prompt Compression in Language Models - "This paper, presented by researchers from Stanford University, proposes a novel technique for prompt compression called gisting, which trains an LM to compress prompts into smaller sets of “gist” tokens. "

Applications

Machine Learning Can Help to Flag Risky Messages on Instagram While Preserving Users’ Privacy - "As regulators and providers grapple with the dual challenges of protecting younger social media users from harassment and bullying, while also taking steps to safeguard their privacy, a team of researchers from four leading universities has proposed a way to use machine learning technology to flag r"

Google’s Bard AI chatbot can now generate and debug code - "Google’s conversation AI tool Bard can now help software developers with programming, including generating code, debugging and code explanation— a new set of skills that were added in response to user demand."

Cyborgs and Cherries: The Unlikely Connection- This AI Approach Uses Reinforcement Learning to Teach Robots How to Use Chopsticks - "Robots are becoming increasingly prevalent in our daily lives, from automated vacuum cleaners to drones delivering packages. We are witnessing growth in their ability to handle complex tasks as technology advances. They are starting to do the tasks that were once limited to human capabilities only."

Bill Gates says A.I. chatbots will teach kids to read within 18 months: You'll be 'stunned by how it helps' - "Soon, artificial intelligence could help teach your kids and improve their grades. That's according to billionaire Microsoft co-founder Bill Gates, who says AI chatbots are on track to help children learn to read and hone their writing skills in 18 months time."

How machine learning can make agriculture more sustainable - "In an era of a rapidly changing climate, achieving agricultural sustainability is critical to ensure the health and well-being of our planet. With limited resources and an ever-increasing population, traditional farming practices can no longer support a sustainable food system."

Business

You can now change the look of Snap’s My AI.

Snapchat is releasing its AI chatbot to everyone for free - "Snap is releasing its “My AI” chatbot to all of Snapchat’s 750 million monthly users for free, a move that comes less than two months after the OpenAI-powered bot was first made available to the app’s more than 3 million paid subscribers."

Carbon Robotics Raises $30 Million in Funding to Scale AI-Powered LaserWeeder™ Platform - "Carbon Robotics, a leader in AI-powered robotics, today announced it closed $30 million in Series C financing from new lead investor Sozo Ventures along with existing investors Anthos Capital, Fuse Venture Capital, Ignition Partners, Liquid2 and Voyager Capital."

SoundHound AI Secures $100 Million in Strategic Financing From Atlas Credit Partners - "SoundHound AI, Inc. a global leader in voice artificial intelligence, today announced that it closed a new $125 million loan facility, with $100 million fully-funded at closing, from Atlas Credit Partners."

Broadcom Launches AI Chip Capable Of Connecting Up To 32K GPUs - "Broadcom Inc AVGO launched the artificial intelligence chip Jericho3-AI Tuesday at a time when the use of AI is skyrocketing. Jericho3-AI can connect up to 32,000 GPU chips together."

Amazon offers free access to its AI coding assistant to undercut Microsoft - "Amazon is making its AI-powered coding assistant free for individual developers, undercutting the $10 per month pricing of its Microsoft-made rival. In a post shared on Thursday, Amazon announced that its CodeWhisperer tool is now available to everyone who signs up to use it."

ChatGPT could cost over $700,000 per day to operate. Microsoft is reportedly trying to make it cheaper. - "Using ChatGPT to write cover letters, generate lesson plans, and redo your dating profile could cost OpenAI up to $700,000 a day because of the pricey tech infrastructure the AI runs on, Dylan Patel, chief analyst at semiconductor research firm SemiAnalysis, told The Information."

Adobe Firefly's new AI tools aim to cut down drudge work for editors - "Less than a month after debuting its new suite of Firefly generative AI editing tools, Adobe announced Monday that it is already working on a host of upgrades to further empower its users through Creative Cloud video and audio applications."

How A.I.-powered robots are changing retail - "Eager to boost sales, relieve workers from mundane tasks and respond to the ongoing labor shortage, retailers and supermarkets are adding robots to their store aisles."

How ChatGPT and generative A.I. are already changing how we book trips and travel - "The ability of ChatGPT to seize upon a specific, verifiable application in a focused business sector could go a long way to underscoring it as a monetizable asset."

Google employees label AI chatbot Bard ‘worse than useless’ and ‘a pathological liar’: report - "Google employees repeatedly criticized the company’s chatbot Bard in internal messages, labeling the system “a pathological liar” and beseeching the company not to launch it."

Concerns

Google CEO Sundar Pichai warns society to brace for impact of A.I. acceleration, says ‘it’s not for a company to decide' - "Google and Alphabet CEO Sundar Pichai said "every product of every company" will be impacted by the quick development of AI, warning that society needs to prepare for technologies like the ones it's already launched."

ChatGPT and its ilk are making it easier for remote workers to secretly hold two or more full-time jobs - "Now add to the mix ChatGPT and its ilk, which can make many jobs much easier to perform."

FTC chair Lina Khan warns AI could ‘turbocharge’ fraud and scams - "Artificial intelligence tools such as ChatGPT could lead to a “turbocharging” of consumer harms including fraud and scams, and the US government has substantial authority to crack down on AI-driven consumer harms under existing law, members of the Federal Trade Commission said Tuesday."

AI anxiety: The workers who fear losing their jobs to artificial intelligence - "Claire has worked as a PR at a major consulting firm, based in London, for six years. The 34-year-old enjoys her job and earns a comfortable salary, but in the past six months, she’s started to feel apprehensive about the future of her career. The reason: artificial intelligence."

How to worry wisely about artificial intelligence - "Rapid progress in AI is arousing fear as well as excitement. How worried should you be?"

Artificial intelligence is infiltrating health care. We shouldn’t let it make all the decisions. - "AI paternalism could put patient autonomy at risk—if we let it."

Michael Schumacher: Seven-time F1 champion's family plan legal action after AI-generated interview - "Michael Schumacher's family are planning legal action against a magazine which published an artificial intelligence-generated interview with the former Formula 1 driver."

Analysis

‘AI isn’t a threat’ – Boris Eldagsen, whose fake photo duped the Sony judges, hits back - "Since 52-year-old German artist Boris Eldagsen went public with the fact that he won a Sony world photography award with an AI-generated image, relations between him and the award body have soured."

What an octopus’s mind can teach us about AI’s ultimate mystery - "Such machines don't exist, of course, and maybe never will. Indeed, the concept of a machine with a subjective experience of the world and a first-person view of itself goes against the grain of mainstream AI research."

Breaking Down AutoGPT: What It Is, Its Features, Limitations, Artificial General Intelligence (AGI) And Impact of Autonomous Agents on Generative AI - "Generative AI is evolving and getting popular. Since its introduction, new models and research papers are getting released almost every other day. The major reason for the exponentially increasing popularity is the development of Large Language Models."

Inside the secret list of websites that make AI like ChatGPT sound smart - "An analysis of a chatbot data set by The Washington Post reveals the proprietary, personal, and often offensive websites that go into an AI’s training data."

There Is No A.I. - "As a computer scientist, I don’t like the term “A.I.” In fact, I think it’s misleading—maybe even a little dangerous. Everybody’s already using the term, and it might seem a little late in the day to be arguing about it."

Policy

EU lawmakers call for summit to control 'very powerful' AI - "EU lawmakers urged world leaders on Monday to hold a summit to find ways to control the development of advanced artificial intelligence (AI) systems such as ChatGPT, saying they were developing faster than expected."

Tesla wins lawsuit that blamed Autopilot for crash - "Tesla has won a lawsuit that tried to blame the company’s Autopilot driver assist software for a 2019 crash, Reuters reports. The jurors in the case found that the software wasn’t at fault in a crash where the car turned into a median on a city street while Autopilot was engaged."

Elon Musk calls for regulation of A.I., plans 'TruthGPT' to rival OpenAI, DeepMind - "Tesla, SpaceX and Twitter CEO Elon Musk called for federal regulation of AI technology during a taped interview on Fox News Channel's "Tucker Carlson Tonight" that aired on Monday."

OpenAI’s hunger for data is coming back to bite it - "OpenAI has just over a week to comply with European data protection laws following a temporary ban in Italy and a slew of investigations in other EU countries. If it fails, it could face hefty fines, be forced to delete data, or even be banned. "

United States: FDA Issues Guidance on Predetermined Change Control Plans for Artificial Intelligence/Machine Learning - "In brief On 3 April 2023, FDA issued a draft guidance for industry, Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence/Machine Learning (AI/ML)-Enabled Device Software Functions, for manufacturers whose medical devices use machine learning (ML) to improve care."

DHS Announces First-Ever AI Task Force - "On Friday, Department of Homeland Security Secretary Alejandro Mayorkas announced the formation of a new resource group focused solely on combating negative repercussions of the widespread advent of artificial intelligence technologies."

As AI eliminates jobs, a way to keep people afloat financially (that’s not UBI) - "In Silicon Valley, some of the brightest minds believe a universal basic income (UBI) that guarantees people unrestricted cash payments will help them to survive and thrive as advanced technologies eliminate more careers as we know them, from white collar and creative roles — lawyers, journalists,"

Expert Opinions

OpenAI chief says age of giant AI models is ending; a GPU crisis could be one reason why - "Speaking at an MIT event last week, Altman suggested that further progress would not come from “giant, giant models.”"

Art

Photographer admits prize-winning image was AI-generated - "A photographer is refusing a prestigious award after admitting to being a “cheeky monkey” and generating the prize-winning image using artificial intelligence."

Drake and The Weeknd AI song pulled from Spotify and Apple - "Image source, Getty ImagesBBC NewsA song that uses artificial intelligence to clone the voices of Drake and The Weeknd is being removed from streaming services. Heart On My Sleeve is no longer available on Apple Music, Spotify, Deezer and Tidal."

AI-Oasis are musics' Deepfake-moment - "Oasis have a new album out and it’s bonkers. Except it’s not Oasis, but a sonic deepfake."

AI Art Sites Censor Prompts About Abortion - "“Why are they censoring something that is clearly under attack?”"

Last Week in AI