Last Week in AI #200: A Review of AI in 2022

From text-to-everything to ChatGPT, here's a look at the most important AI news in 2022

Happy New Year!

We began publishing our weekly newsletter in 2018 to provide accessible and informed coverage of the latest AI news and trends. As grad students in AI, it was clear to us that the field was rapidly advancing and would significantly impact the rest of the world, but media coverage often focused on hype rather than facts. We wanted to make a newsletter that made it possible to find the signal amid the noise and understand what is happening in the world of AI, even if you are not a technical person. Much has changed since 2018, and the pace of AI progress and deployment has grown enormously, making us more confident than ever this sort of newsletter needs to exist.

Here's to the next 200!

New Year and 200th Issue Special Offer!

Help us celebrate the new years and crossing the 200th week! Subscribe now to get 40% off forever! Subscribers get early access to exclusive editorials and help us offset expenses from running Last Week in AI. Thank you!

2022 A Year in Review

January

The world of AI began the year relatively quietly. We saw continued deployments of AI-powered solutions in real applications, like self-driving tractors, nurse-assistant robots, and early Covid detection systems. AI-powered fighter jets saw new tests and inch closer to becoming a reality.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• The Rise of A.I. Fighter Pilots

• We’re one step closer to self-farming farms

• BioNTech and London A.I. company create "early warning system" for COVID-19 variants

• Machine-Learning Program Connects to Human Brain and Commands Robots

• Researchers Propose Mitigation Strategies to Tackle Overinterpretation of Deep Learning Methods

• Robo-dogs and therapy bots: Artificial intelligence goes cuddly

• Feds' spending on facial recognition tech expands, despite privacy concerns

• ‘We Were Blown Away’: How New A.I. Research Is Changing the Way Conservators and Collectors Think About Attribution

• In Texas, driverless trucks are set to take over roads

• Delivery Care Robots Are Being Used to Alleviate Nursing Staff Shortage

Editorials:

• The Messy History of Facial Recognition Company Clearview AI

• How AI-Powered Robots Fulfill Your Online Orders

• AI and the Metaverse

February

February saw the use of Deep Reinforcement Learning to control plasma shape in nuclear fusion reactor. This is a very impressive and impactful real-world application of Deep RL, which often gets criticized for its lack of real-world uses. Meanwhile, several stories highlighted that facial recognition remains one of the most concerning trends related to AI. Fortunately, progress on AI regulation around the world continued to accelerate.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• Accelerating fusion science through learned plasma control

• OpenAI rolls out new text-generating models that it claims are less toxic

• Meta Aims to Build the World's Fastest AI Supercomputer

• Deepfake Regulations in EU, UK, and US and China

• Celestial AI lands $56M to develop light-based AI accelerator chips

• DeepMind claims its new code-generating system is competitive with human programmers

• The EU and U.S. are starting to align on AI regulation

• A New Trick Lets Artificial Intelligence See in 3D

• An ancient language has defied translation for 100 years. Can AI crack the code?

• Lawmakers Warn Clearview AI Could End Public Anonymity if Feds Don't Ditch It

• Texas Sues Meta Over Facebook’s Facial-Recognition Practices

• AI giant SenseTime expands tech application to manufacturing sector

Editorial:

• Generating AI Art from Text with Google Colab

• Neural nets are not "slightly conscious," and AI PR can do with less hype

March

Mikael Thalen @MikaelThalen

A deepfake of Ukrainian President Volodymyr Zelensky calling on his soldiers to lay down their weapons was reportedly uploaded to a hacked Ukrainian news website today, per @Shayan86

Russia began its invasion of Ukraine in late February, and AI-powered technology has played a part in this war. Among other things, we saw perhaps for the first time a malicious wartime use of deepfake technology that showed the Ukrainian president giving a surrender speech. This would and will not be the last time AI is used in war.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• Deepfake Zelenskyy surrender video is the 'first intentionally used' in Ukraine war

• Russia’s AI Army: Drones, AI-Guided Missiles and Autonomous Tanks

• Europe Is in Danger of Using the Wrong Definition of AI

• Can A.I. Help Casinos Cut Down on Problem Gambling?

• The secret police: Cops built a shadowy surveillance machine in Minnesota after George Floyd's murder

• Microsoft, OpenAI may have solved a fundamental AI bottleneck

• AI suggested 40,000 new possible chemical weapons in just six hours

• Meta AI’s Sparse All-MLP Model Doubles Training Efficiency Compared to Transformers

• Nvidia shows off AI model that turns a few dozen snapshots into a 3D-rendered scene

• Ukraine uses facial recognition to identify dead Russian soldiers, minister says

• US-China collaboration in AI papers drops amid ongoing tech war, Stanford report shows

Editorials:

• A Summary of Concerns About Clearview AI's Facial Recognition Product

• Redefining "Inventor"- Should AI Systems be Granted Patents?

April

This year for AI really kicked off with OpenAI’s release of DALL-E 2, the first really good text-to-image AI model. For many in the general public, DALL-E 2 would be their first contact with really powerful generative AI, but more impressive models turned out to be just around the corner. At the same time, Google’s massive Pathways Language Model demonstrated that text-generating AI will keep making huge progress alongside their text-to-image peers.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• OpenAI's new DALL.E model turns your words into pieces of art

• Pathways Language Model (PaLM): Scaling to 540 Billion Parameters for Breakthrough Performance

• AI-Directed Algae Blooms Boost Biofuel Prospects

• The military wants AI to replace human decision-making in battle

• Face scanner Clearview AI aims to branch out beyond police

• Venture funding for chip startups has doubled in the last five years thanks to AI

• How Hospitals Are Using AI to Save Lives

• Panera Bread is testing automated coffee brewing with Miso Robotics

• South Africa’s private surveillance machine is fueling a digital apartheid

• GM Patents Autonomous Tech to Train New Drivers Sans Instructor

• Actors launch campaign against AI 'show stealers'

Editorial: Kamikaze Drones in Russia’s War Against Ukraine Point to Future "Killer Robots"

May

DeepMind’s Gato model convinced many that transformers may be “the only thing that you need” in deep learning - this one model, trained on text, images, and audio. Hugging Face, the “GitHub of machine learning”, reached enormous evaluations that once again demonstrated the growth of AI’s importance in industrial applications, and AI’s growing footprint on the economy in general.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4 | Week 5

Highlights:

• DeepMind’s new AI can perform over 600 tasks, from playing games to controlling robots

• Autonomous cargo ship completes 500 mile voyage, avoiding hundreds of collisions

• Scientists Discover Method to Break Down Plastic in Days, Not Centuries

• Machine Learning Helps See into a Volcano’s Depths

• Anthropic’s quest for better, more explainable AI attracts $580M

• An algorithm that screens for child neglect raises concerns

• Singaporean wins $100k prize in challenge to build AI models that detect deepfakes

• AI delivers real-time data for smarter farming

• Mayo researchers use AI to detect weak heart pump via patients' Apple Watch ECGs

• Hugging Face reaches $2 billion valuation to build the GitHub of machine learning

• A quick guide to the most important AI law you’ve never heard of

• How A.I. Is Changing Hollywood

• AI may be searching you for guns the next time you go out in public

• Royal Mail is Doing the Right Thing with Drone Delivery

• Undetectable Backdoors Plantable In Any Machine Learning Algorithm

• AI and machine learning are improving weather forecasts, but they won’t replace human experts

Editorials:

• Foundation Models and the Future of Multi-Modal AI

• Russia May Have Used Autonomous Drones. Now What?

June

The field of Embodied AI - AI that must act within an environment to complete temporally extended tasks - also picked up steam this year with many research groups using Minecraft as a diverse, open-world, and multi-task AI testbed. This VPT work is only one such example. Also this month, one of the biggest AI stories of the year emerged: the claim that Google’s LaMDA model is sentient. As we wrote, this was not the case, but the story spread like wildfire regardless.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• Learning to Play Minecraft with Video PreTraining (VPT)

• Google Sidelines Engineer Who Claims Its A.I. Is Sentient

• DALL·E mini - Generate Images from Any Text Prompt

• Reddit Bans ‘SFW’ Deepfake Community

• Instagram is testing an AI tool that verifies your age by scanning your face

• The AI Defending Ukraine

• Amazon launches CodeWhisperer, a GitHub Copilot-like AI pair programming tool

• Can Machine Learning Translate Ancient Egyptian Texts?

• Papercup raises $20M for AI that automatically dubs videos

• UCI Researchers: Autonomous Vehicles Can Be Tricked Into Dangerous Driving Behavior

• Satellites and AI Can Help Solve Big Problems—If Given the Chance

• Google bans deepfake-generating AI from Colab

Editorial: LaMDA’s Sentience is Nonsense - Here’s Why

July

Perhaps somewhat quietly, and amid all the tech layoffs and downsizing, the industry of robo-taxis is slowly maturing and gaining ground. Cruise was the first company to widely deploy its robo-taxi service in San Francisco, with Waymo soon following soon. These self-driving cars are not perfect and are susceptible to mistakes like causing the traffic jam above. However, it’s undeniable that these services “work”, and robo-taxis may actually be soon upon us.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• Cruise’s Robot Car Outages Are Jamming Up San Francisco

• Open-source language AI challenges big tech’s models

• Cerebras Systems sets record for largest AI models ever trained on one device

• Fake Friends and the Real Threat of AI-Generated Influencers

• The Fight Over Which Uses of AI Europe Should Outlaw

• Autonomous Drones Challenge Human Champions in First “Fair” Race

• Commercial image-generating AI raises all sorts of thorny legal issues

• From ‘Barbies scissoring’ to ‘contorted emotion’: the artists using AI

• Sony’s racing AI destroyed its human competitors by being nice (and fast)

August

If there’s been one trend that defined AI progress in 2022, it was the advancements in text-conditioned generative models. Only a few months after OpenAI’s DALL-E 2, Google, Facebook, and others developed their own incredible image generation models, with some progress also being made in creating entire videos as well. The key driver to all this progress is scale — using larger models, more data, and more computational power to train the same simple models.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4 | Week 5

Highlights:

• After Text-to-Image, Now it’s Text-to-Video

• Meta’s AI chatbot says Trump will always be president and repeats anti-Semitic conspiracies

• Documents Reveal Advanced AI Tools Google Is Selling to Israel

• Police can use facial recognition again after ban in New Orleans, home to sprawling surveillance

• How machine learning could help save threatened species from extinction

• New algorithm aces university math course questions

• How a femtech app is using A.I. to fill in the gaps for women’s health care

• Machine Learning Is Causing a ‘Reproducibility Crisis’ in Science

• TrashBot uses AI to sort recyclables

• China drafts rules on use of self-driving vehicles for public transport

• AI May Come to the Rescue of Future Firefighters

• Tesla wants to take machine learning silicon to the Dojo

• The State of State AI Policy

Editorials:

• The AI Scaling Hypothesis

• In the Era of Large Language Models, What's Next for AI Chatbots ?

September

September proved to be a fairly quiet month, with a mix of news about new exciting research, business moves, growth in the AI ecosystem, progress on regulations, and AI’s growing impact on the media we consume.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• Using AI to decode speech from brain activity

• DALL-E can now use AI to extend images as a human artist might

• Robots can be used to assess children’s mental wellbeing, study suggests

• Uber Eats and Nuro sign a 10-year deal to do robot food delivery in California and Texas

• A.I. Is Making It Easier Than Ever for Students to Cheat

• Announcing the PyTorch Foundation: A new era for the cutting-edge AI framework

• The FTC Is Closing In on Runaway AI

• There’s no Tiananmen Square in the new Chinese image-making AI

• OpenAI open-sources Whisper, a multilingual speech recognition system

• NVIDIA's new AI model quickly generates objects and characters for virtual worlds

• Darth Vader’s Voice Emanated From War-Torn Ukraine

Editorials:

• Can We Use AI to Communicate With Animals?

• More Claims that AI is Sentient are (Probably) Coming

October

Ever the hype man, Elon Musk promised a lot when it came to the reveal of the TeslaBot on “AI Day 2”. Many robotics researchers were skeptical, but the general consensus after the event was that Tesla had achieved a lot with their prototype in just one year. Still, it is clearly far from a finished product. Aside from that notable highlights, many stories this month familiar themes: text-to-X generation, self driving cars, regulation, and new exciting applications of AI.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• Google’s newest AI generator creates HD video from text prompts

• Meta’s new text-to-video AI generator is like DALL-E for video

• DeepMind unveils first AI to discover faster matrix multiplication algorithms

• Using AI to Translate Speech For a Primarily Oral Language

• State of AI Report 2022

• Elon Musk Reveals Tesla's Optimus Robot, Says It 'Can Actually Do a Lot More' as It Walks and Waves on Stage

• House Democrats debut new bill to limit US police use of facial recognition

• Even After $100 Billion, Self-Driving Cars Are Going Nowhere

• FedEx abandons its last-mile delivery robot program

• AI in Medicine Is Overhyped

• AI-generated imagery is the new clip art as Microsoft adds DALL-E to its Office suite

Editorial: Tesla's AI Day Was a Success

November

November was the month that many artists made their disdain for AI-generated art known, and many platforms had to figure out their policies regarding AI-generated images. Text generation tools also grew in notoriety, with more and more op-eds being written about their impact on schoolwork; how do you assign students to write essays, when an AI can write a convincing unique essay on just about any topic? Meanwhile, the AI community went through some drama as Meta released an AI model that purported to be capable of writing papers, but was soon demonstrated to often output silly things.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4 | Week 5

Editorial: Robots That Write Their Own Code

Highlights:

• CICERO: An AI agent that negotiates, persuades, and cooperates with people

• How Amazon Robotics researchers are solving a “beautiful problem”

• Google wants robots to generate their own code

• TSMC Said to Suspend Work for Chinese Chip Startup Amid US Curbs

• Shutterstock will start selling AI-generated stock imagery with help from OpenAI

• Ford, VW-backed Argo AI is shutting down

• Meta layoffs hit entire ML research team focused on infrastructure

• AlphaFold’s new rival? Meta AI predicts shape of 600 million proteins

• A teacher allows AI tools in exams – here’s what he learned

• Welp, There Goes Twitter's Ethical AI Team, Among Others as Employees Post Final Messages

• Mythic bet big on analog AI but has run out of cash

• Why Meta’s latest large language model survived only three days online

• Anime Convention Bans All AI-Generated Art

• This World Cup is wired and fueled by AI

• Can an AI-powered insect trap solve a $220 billion pest problem?

Editorial: Robots That Write Their Own Code

December

The year ended with ChatGPT capturing the spotlight and making it more widely known than ever how advanced text-generating AI has become. At the same time, the AI-powered portrait generation app Lensa went viral, making it known how advanced image-generating AI has become. As we had to 2023, one thing is for certain: the scale of AI models, the public’s awareness of AI, and the number of AI research advancements will all continue to grow.

Newsletters: Week 1 | Week 2 | Week 3 | Week 4

Highlights:

• ChatGPT: Optimizing Language Models for Dialogue

• The DOJ is reportedly investigating rent-setting software company RealPage

• San Francisco debates letting police deploy robots that kill

• How AI That Powers Chatbots and Search Queries Could Discover New Drugs

• The smallest robotic arm you can imagine is controlled by artificial intelligence

• San Francisco reverses plans to allow police robots to kill suspects

• How does GPT Obtain its Ability? Tracing Emergent Abilities of Language Models to their Sources

• Wheels up for Waymo as we expand our 24/7 rider-only territories

• Stability AI plans to let artists opt out of Stable Diffusion 3 image training

• Enjoy Chatbots While They're Free

• TikTok Wants to Be More Transparent With Its Algorithm

• A New Chat Bot Is a ‘Code Red’ for Google’s Search Business

Editorial: AI's Year of Text-to-Everything

---

Closing Note

Throughout the last few years, Last Week in AI has been a side project of three people —- Andrey, Jacky, and Daniel. We are continuously encouraged to find the time and energy for it because of all of you, the readers! So thank you. Have a great 2023!

PS: feel free to chat with us here!

Copyright © 2023 Skynet Today, All rights reserved.