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!

Get 40% off forever

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:

Editorials:

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:

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:

Editorials:

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:

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:

Editorials:

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:

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:

August

Meta and Facebook logos are seen in this illustration taken February 15, 2022. Photo: Reuters

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:

Editorials:

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:

Editorials:

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:

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:

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:

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.

Last Week in AI

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