Last Week in AI #204: Google's GPT3 for music, Buzzfeed embraces AI, a new video from Boston Dynamics, and more!
Discover more from Last Week in AI
Last Week in AI #204: Google's GPT3 for music, Buzzfeed embraces AI, a new video from Boston Dynamics, and more!
Google created an AI that can generate music from text descriptions, BuzzFeed says it will use AI tools from OpenAI to personalize its content, Stanford students used ChatGPT on final exams
Top News
Google created an AI that can generate music from text descriptions, but won’t release it
Summary (written by ChatGPT): Google’s new AI system, MusicLM, can create music in any genre by taking a text description as input and generating a coherent song. It was trained on a dataset of 280,000 hours of music and generates songs with significant complexity and a sound that resembles that of a human artist. The system can also be instructed via a description of an image, or generate audio for a specific type of instrument. However, the system is not without flaws, including the use of copyrighted material from its training data, distorted quality in some samples, and poor vocals that range from barely English to gibberish. Google is aware of the ethical challenges posed by MusicLM and has no immediate plans to release it.
Our take: although there have been previous impressive music-generation tools (notably Jukebox by OpenAI), this is a huge advance. The clarity and quality of the sound, minutes-long consistency in the composition, and versatility of this is truly impressive. As the article notes, it is also clearly a sign that musicians will soon face the same dilemmas as visual artists already do due to text-to-image tools being released publicly. Copyright may delay that somewhat, but it’s ony a matter of time.
BuzzFeed says it will use AI tools from OpenAI to personalize its content
Summary: According to a January 26 memo from BuzzFeed CEO Jonah Peretti, the media company plans to use tools from OpenAI to “enhance” and “personalize” its content. BuzzFeed’s “A-inspired content” will launch this year: this content could include personalized rom-com pitches for readers. Peretti’s memo clearly demonstrates his faith that AI tools are an important part of BuzzFeed’s future and his aspiration for BuzzFeed to lead the future of AI-powered content. This news was followed by a significant increase in BuzzFeed’s share price.
Our take: First, it’s interesting to note the share price reaction to this memo: the news that “we’re going to use AI-powered content” alone was apparently enough to send BuzzFeed’s share price soaring after a rocky start in the public markets. This does seem an artifact of the clear investor hype we’ve seen since the release of ChatGPT. It will also be interesting to see how BuzzFeed implements these tools in production. CNET found errors in more than half of a number of articles offering financial advice where it had used AI systems to assist. That being said, BuzzFeed already has extensive experience with AI-assisted curation. Will that expertise translate to creation? We’ll have to wait and see.
The best use for AI eye contact tech is making movie stars look straight at the camera
Summary: Daniel Hashimoto, a VFX specialist, shared a tweet showing edited movie clips using Nvidia's real-time Eye-Contact feature. The feature is a part of Nvidia's Broadcast software, a video conferencing application that allows for real-time editing of video calls to make the caller appear to be making eye contact with the camera. Hashimoto used the software to make characters in the movie clips directly look at the camera. He is known for his Youtube channel, "Action Movie Kid" and his weekly podcast, "VFX and Chill". He praised the technology as being truly incredible but also commented that the characters staring directly at the camera made the clips look funny.
Our Take: The use of Nvidia's real-time Eye-Contact feature by VFX specialist Daniel Hashimoto highlights the potential of AI in the entertainment industry. The ability to edit the eye gaze of characters during post-production could ease the workload for actors and production teams, as well as reducing the number of takes required. However, as the article and comments on Hashimoto's tweet thread suggest, having characters stare directly into the camera can come across as strange and unnatural. It's not a common thing that humans maintain such prolonged eye contact. Nevertheless, such advancements in technology have the potential to revolutionize the entertainment industry and push the boundaries of what is possible.
Other News
Research
Atlas Gets a Grip | Boston Dynamics - “It’s time for Atlas to pick up a new set of skills and get hands on. In this video, the humanoid robot manipulates the world around it: Atlas interacts with objects and modifies the course to reach its goal—pushing the limits of locomotion, sensing, and athleticism.”
Machine learning identifies T cell receptor repertoire signatures associated with COVID-19 severity - “T cell receptor (TCR) repertoires are critical for antiviral immunity. Determining the TCR repertoire composition, diversity, and dynamics and how they change during viral infection can inform the molecular specificity of host responses to viruses such as SARS-CoV-2.”
Uncovering contaminants via machine learning - “Detecting and identifying contaminants in biological and environmental samples in a laboratory setting can be quite difficult, as the contaminants often occur in mixtures of similar molecules.”
Machine learning identifies drugs that could potentially help smokers quit - “HERSHEY, Pa. — Medications like dextromethorphan, used to treat coughs caused by cold and flu, could potentially be repurposed to help people quit smoking cigarettes, according to a study by Penn State College of Medicine and University of Minnesota researchers.”
Natural language model jumpstarts protein design with creation of active enzymes - “Scientists have created an AI system capable of generating artificial enzymes from scratch. In laboratory tests, some of these enzymes worked as well as those found in nature, even when their artificially generated amino acid sequences diverged significantly from any known natural protein.”
Versatile robo-dog runs through the sandy beach at 3 mile/sec - “KAIST (President Kwang Hyung Lee) announced on the 25th that a research team led by Professor Jemin Hwangbo of the Department of Mechanical Engineering developed a quadrupedal robot control technology that can walk robustly with agility even in deformable terrain such as sandy beach.”
AI has designed bacteria-killing proteins from scratch – and they work - “An AI was tasked with creating proteins with anti-microbial properties. Researchers then created a subset of the proteins and found some did the job An AI has designed anti-microbial proteins that were then tested in real life and shown to work.”
Google created an AI that can generate music from text descriptions, but won’t release it - “An impressive new AI system from Google can generate music in any genre given a text description. But the company, fearing the risks, has no immediate plans to release it. Called MusicML, Google’s certainly isn’t the first generative AI system for song.”
Applications
Can an AI ‘angel’ help find thousands in Mexico who were forcibly disappeared? - “In a country with around 110,000 people reported as missing, presumably dead, locating them could be just a few clicks away. That is the hope behind Angelus 2.0, a computer program developed by the Mexican government in an effort started four years ago.”
People are already using ChatGPT to create workout plans - “When I opened the email telling me I’d been accepted to run the London Marathon, I felt elated. And then terrified.”
Learn to Write: AI Advancements Mean Code Will Be Written in Natural English - “AI-powered chatbot ChatGPT is, according to its boosters, poised to disrupt or even make obsolete everything writerly—from the college essay to legal arguments. But large language models are already transforming another human pursuit: writing code. ”
Zillow’s new AI feature can talk to home shoppers like it’s a friend - “Zillow launched a new AI-powered feature called “natural language search” that can now interpret home shoppers’ colloquial lingo, much like how they would talk to friends and family. Zillow is the first company to implement such a search engine.”
Big Tech was moving cautiously on AI. Then came ChatGPT. - “Google, Facebook and Microsoft have used AI in their products for years. But OpenAI’s ChatGPT is stealing the limelight, forcing them to move faster.”
ChatGPT: Grading artificial intelligence’s writing - “OpenAI’s artificial intelligence writing program ChatGPT will, with a few prompts, compose poetry, prose, song lyrics, essays, even news articles. And that has ethicists and educators worried about the program’s ease at replacing human ideas with chatbot-generated words.”
Business
University of Texas Will Offer Large-Scale Online Master’s Degree in A.I. - “The University of Texas at Austin, one of the nation’s leading computer science schools, said on Thursday that it was starting a large-scale, low-cost online Master of Science degree program in artificial intelligence.”
Microsoft to Invest $10 Billion in OpenAI, the Creator of ChatGPT - “Microsoft said on Monday that it was making a “multiyear, multibillion-dollar” investment in OpenAI, the San Francisco artificial intelligence lab behind the experimental online chatbot ChatGPT.”
Generative AI funding exploded over the past 2 years - “US venture capital funding of generative AI was up 27% year over year in 2022, hitting $1.4 billion, according to PitchBook. That money, however, was spread among fewer deals—78 last year compared with 85 in 2021. Nonetheless, it represents a massive increase from 2020, when funding totaled $200 million.”
Alphabet-owned AI firm DeepMind shutters office in Canada’s Edmonton - “Jan 24 (Reuters) - DeepMind Technologies, owned by Alphabet Inc (GOOGL.O), is closing its office in the Canadian city of Edmonton, a spokesperson for the AI firm said, just days after Google’s parent announced it would lay off 12,000 employees.”
Amazon robotics leader Siddhartha Srinivasa leaves to join self-driving vehicle venture Cruise - “— Siddhartha Srinivasa, a leading robotics expert and professor at the University of Washington, has left his position as director of Robotics AI at Amazon to join Cruise, General Motors’ autonomous-vehicle subsidiary.”
After inking its OpenAI deal, Shutterstock rolls out a generative AI toolkit to create images based on text prompts - “When Shutterstock and OpenAI announced a partnership to help develop OpenAI’s Dall-E 2 artificial intelligence image-generating platform with Shutterstock libraries to train and feed the algorithm, the stock photo and media giant also hinted that it would soon be bringing its own generative AI too”
This Startup Is Using AI to Unearth New Smells - “Alex Wiltschko opens a black plastic suitcase and pulls out about 60 glass vials. Each contains a different scent. One smells starchy with soft floral notes, like jasmine rice cooking. Another brings to mind ocean air and the white rind of a watermelon.”
Artificial intelligence could save healthcare industry $360B a year - “Dive Brief: Artificial intelligence could save the U.S. up to $360 billion annually if adopted more widely in healthcare, according to a new report from McKinsey and Harvard researchers.”
Anthropic, an A.I. Start-Up, Is Said to Be Close to Adding $300 Million - “Anthropic, a San Francisco artificial intelligence start-up, is close to raising roughly $300 million in new funding, two people with knowledge of the situation said, in the latest sign of feverish excitement for a new class of A.I. start-ups.”
Concerns
Scores of Stanford students used ChatGPT on final exams, survey suggests - “Stanford students and professors alike are grappling with the rise of ChatGPT, a chatbot powered by artificial intelligence, and the technology’s implications for education.”
AI-generated art a concern for local artists who say there’s still a need for human creativity - “Two local artists have shared their opinion of recent technological advancements in artificial intelligence (AI) generated art, arguing that while there are concerns over its usage, human creativity cannot be taken over by technology.”
I’m a Congressman Who Codes. A.I. Freaks Me Out. - “Imagine a world where autonomous weapons roam the streets, decisions about your life are made by AI systems that perpetuate societal biases and hackers use AI to launch devastating cyberattacks.”
ChatGPT can’t be credited as an author, says world’s largest academic publisher - “Springer Nature, the world’s largest academic publisher, has clarified its policies on the use of AI writing tools in scientific papers. The company announced this week that software like ChatGPT can’t be credited as an author in papers published in its thousands of journals.”
Don’t be sucked in by AI’s head-spinning hype cycles - “The last year was a roller coaster ride in the AI world, and no doubt many people are dizzied by the number of advances and reversals, the constant hype and equally constant fearmongering.”
CNET found errors in more than half of its AI-written stories - “CNET issued corrections on 41 of the 77 stories the outlet published that were written using an AI tool.”
Robot Cars Are Causing 911 False Alarms in San Francisco - “For some residents of San Francisco, the robotic future of driving is just a tap away. Ride-hailing services from GM subsidiary Cruise and Alphabet company Waymo allow them to summon a driverless ride with an app. But some riders have become perhaps too comfortable with the technology.”
A watermark for chatbots can expose text written by an AI - “Hidden patterns purposely buried in AI-generated texts could help identify them as such, allowing us to tell whether the words we’re reading are written by a human or not.”
Analysis
How to Spot AI-Generated Art, According to Artists - “How long will the naked eye be able to spot the difference between images made by generative artificial intelligence and art created by humans? Ari Melenciano, an artist who works at Google’s Creative Lab, squints at her computer screen during our Zoom chat and scans artwork created with generative”
AI Is Improving Faster Than Most Humans Realize - “Technologies such as ChatGPT will profoundly change the economy, but they are getting better so quickly that it’s impossible to say how.”
Policy
Sweden to upgrade Berzelius supercomputer with Nvidia AI systems - “STOCKHOLM, Jan 24 (Reuters) - Sweden will upgrade its Berzelius supercomputer using Nvidia’s (NVDA.O) latest artificial intelligence (AI) systems to aid in understanding various diseases such as cancer and make it among the world’s fastest AI supercomputers.”
When May a Robot Kill? New DOD Policy Tries to Clarify - “Did you think the Pentagon had a hard rule against using lethal autonomous weapons? It doesn’t. But it does have hoops to jump through before such a weapon might be deployed—and, as of Wednesday, a revised policy intended to clear up confusion.”
California Might Allow Self-Driving Semi-Trucks on Roads - “(TNS) — Following years of pleadings from the autonomous vehicle industry, California officials are revisiting a statewide policy that prohibits self-driving semi-trucks and big rigs on its roads.”
Expert Opinions
Animal testing can be replaced with AI, Machine Learning and Deep Learning, says expert - “Animal testing can be replaced with Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning, said P.”
ChatGPT is ‘not particularly innovative,’ and ‘nothing revolutionary’, says Meta’s chief AI scientist - “Why hasn’t the public seen programs like ChatGPT from Meta or from Google? “The answer is, Google and Meta both have a lot to lose by putting out systems that make stuff up,” says Meta’s chief AI scientist, Yann LeCun.”
Explainers
The Transformer Family Version 2.0 - “Many new Transformer architecture improvements have been proposed since my last post on “The Transformer Family” about three years ago. Here I did a big refactoring and enrichment of that 2020 post — restructure the hierarchy of sections and improve many sections with more recent papers.”
Fun
New AI Tech Gives Twitch Streamers Constant, Creepy Eye Contact - “Nvidia’s come up with new AI-powered tech that swaps out your real eyes for digital ones that will never stop looking at your webcam. It’s very creepy, and one more sign of how unnatural you have to act online to become a popular streamer in 2023.”