Messages
5/3/2021
11 : 4
Import AI 247: China makes its own GPT3; the AI hackers have arrived; four fallacies in AI research.
How might different alien intelligences conceive of AI? If - or perhaps when - we meet aliens, will they have also developed things that seem like neural networks? How much diversity is possible in the
4/26/2021
10 : 14
Import AI 246: Generating data via game engines; the FTC weighs in on AI fairness; Waymo releases a massive self-driving car dataset.
In the same way 'just-in-time' manufacturing revolutionized global capitalism, how much 'just-in-time' automatic data gathering speed up the OODA loop of model development and
4/12/2021
11 : 4
Import AI 244: NVIDIA makes better fake images; DeepMind gets better at weather forecasting; plus 5,000 hours of speech data.
If you had to design a Von Neumann probe, what would be the part of the design you'd be happiest to cheap-out on? View this email in your browser Welcome to Import AI, a newsletter about artificial
4/5/2021
11 : 4
Import AI 243: Training AI with fractals, RL-trained walking robots; and the European AI fund makes grants to 16 organizations
What will be some of the most sacred memories of an AI system built two hundred years from now? View this email in your browser Welcome to Import AI, a newsletter about artificial intelligence. Forward
3/29/2021
11 : 4
Import AI 242: ThreeDWorld, 209 delivery drone flights, Spotify transcripts versus all the words in New York City
We're almost ten years out from the ImageNet 2012 result. What's the most important and dramatic system that has broken a benchmark since? From my perspective, there's been more incremental
3/22/2021
10 : 34
Import AI 241: The $2 million dataset; small GPT-3 replications; ImageNet gets a face-blur update
Prediction: By 2030, most computation on planet earth will be "restricted" and "unrestricted computation" will be associated with fringe actors and nationstate/megacorp proxies.
3/15/2021
12 : 14
Import AI 240: The unbeatable MATH benchmark; an autonomous river boat dataset; robots for construction sites
In a few decades, the only true records of certain pieces of art or media will be in neural networks, rather than the original input media (which will have been lost). How might 'neural
3/8/2021
12 : 4
Import AI 239: China trains a massive 10b model, Vicarious does pick and place; the GCHQ publishes some of its thoughts on AI
What could be the physically largest neural network you could build? And what would it be made of? People moving in formations? Water flowing through vessels? Great temperature fluctuations being
3/1/2021
16 : 14
Import AI 238: Robots that fold clothes; how Bytedance censors its product; a differentiable simulator.
Instead of inventing time travel, we could just invent a bunch of high-fidelity simulations of the past and beam ourselves into them. View this email in your browser Welcome to Import AI, a newsletter
2/22/2021
10 : 4
Import AI 237: GPT3 at 5X the speed; 6 hours of AI breakbeats; NeuralMMO++
GPT3 + Videogames = entertaining, babbling AI-infused NPCs View this email in your browser Welcome to Import AI, a newsletter about artificial intelligence. Forward this email to give your chums an AI
2/15/2021
10 : 4
Import AI 236: EfficientNet++; why robots are hard; AI2 makes a harder ARC
In what year will we get the '3D game simulator' equivalent of AI Dungeon, where you can generate and unroll radically different gameworlds in response to some user priming? View this email in
2/8/2021
11 : 4
Import AI 235: Use GEM to test language models; the four eras of facial recognition; and how the US can measure its robot fleet
How big will the market for 'reality simulators' eventually become? View this email in your browser Welcome to Import AI, a newsletter about artificial intelligence. Forward this email to give
2/1/2021
12 : 4
Import AI 234: Pre-training with fractals; compute&countries; GANS for good
In what year will we see the emergence of the first religion entirely centered around AI? View this email in your browser Welcome to Import AI, a newsletter about artificial intelligence. Forward this
1/25/2021
10 : 4
Import AI 233: AI needs AI designers; estimating COVID risk with AI; the dreams of an old computer programmer.
In what year will we be able to record the 'dreams' of AI models in sufficient definition to make diagnoses from their content? View this email in your browser Welcome to Import AI, a
1/18/2021
11 : 4
Import AI 232: Google trains a trillion parameter model; South Korean chatbot blows up; AI doesn't use as much electricity as you think
How might far-future historians describe the growth of the world's data centers? How strange, to have a people build large buildings with increasingly intricate vents and interfaces to the outside
1/11/2021
10 : 4
Import AI 231: US army builds nightvision facial recognition; 800GB of text for training GPT-3 models; fighting COVID with a mask detector
Back in the 2010s, people were obsessed with graphene, making grand proclamations about how the material was about to upend the semiconductor industry. Graphene found many uses, but it hasn't
12/28/2020
11 : 4
Import AI 229: Apple builds a Hypersim dataset; ways to attack ML; Google censors its research
A somewhat short issue, due to the holiday season. I hope you are all well and spending time with loved ones. View this email in your browser Welcome to Import AI, a newsletter about artificial
12/21/2020
11 : 4
Import AI 228: Alibaba uses AI to spot fake brands; China makes AI whale songs; what 36 experts think is needed for fair AI in India
How might the current discussions about fairness in machine learning relate to future policy work by regulators? Will we see different definitions of fairness emerge and be regulated in specific parts
12/14/2020
10 : 4
Import AI 227: MAAD-Face; GPT2 and Human Brains; Facebook detects Hateful Memes
How might an AI system conceptualize time travel? Would they perhaps see the ability to slow time as being (somewhat) comparable to acquiring more compute power and developing more efficient algorithms
12/7/2020
11 : 4
Import AI 226: AlphaFold; a Chinese GPT2; Google fires Timnit Gebru
What would it look like to be able to train ML models via a touch-based UI on a mobile phone? And how realistic would it be to let users finetune their own models from generic pre-trained ones? View