Links for 2023-03-01
Language Is Not All You Need: Aligning Perception with Language Models https://arxiv.org/abs/2302.14045
Highlights:
KOSMOS-1 is a Multimodal Large Language Model (MLLM) that is capable of perceiving multimodal input, following instructions, and performing in-context learning for multimodal tasks.
“A big convergence of language, multimodal perception, action, and world modeling is a key step toward artificial general intelligence…”
We also show that MLLMs can benefit from cross-modal transfer, i.e., transfer knowledge from language to multimodal, and from multimodal to language.”
Raven IQ test: “The results indicate that KOSMOS-1 is able to perceive abstract conceptual patterns in a nonverbal context, and then deduce the following element across multiple choices.”
More links:
“Introducing a big update to Windows 11 making the everyday easier including bringing the new AI-powered Bing to the taskbar” https://blogs.windows.com/windowsexperience/2023/02/28/introducing-a-big-update-to-windows-11-making-the-everyday-easier-including-bringing-the-new-ai-powered-bing-to-the-taskbar/
“...we demonstrate that training an RL agent at scale leads to a general in-context learning algorithm that can adapt to open-ended novel embodied 3D problems as quickly as humans...We believe our results lay the foundation for increasingly general and adaptive RL agents that perform well across ever-larger open-ended domains.” https://www.lesswrong.com/posts/GLrnyH4ChFhMqsy4v/powerful-mesa-optimisation-is-already-here
Pre-training generalist agents using offline reinforcement learning https://ai.googleblog.com/2023/02/pre-training-generalist-agents-using.html
Introducing CACTI — a framework for scalable multi-task multi-scene imitation learning: “Our agents show generalization to unseen distractors and are extremely sample efficient, capable of 10 unique real-world tasks with just 1k demonstrations.” https://cacti-framework.github.io/
New approach retrains deep neural networks to deal with changes in complex systems https://physicsworld.com/a/new-approach-retrains-deep-neural-networks-to-deal-with-changes-in-complex-systems/
The US Copyright Office Says You Can’t Copyright Midjourney AI-Generated Images https://www.reddit.com/r/COPYRIGHT/comments/1197ylf/comment/j9lmvkt/
“Bitcoin’s Future Depends on a Handful of Mysterious Coders” [Wall Street Journal] https://archive.is/c7HIe
"As of 2021, 21% of Americans say it is unacceptable for police to use force against a person which is attacking them! 40% disagree with using force against an escaped suspect" https://lefineder.substack.com/p/dont-tase-men-bro
“How Did the Taliban Win? The short answer is that they auditioned to replace the state across the spectrum of control — including punitive violence, but also the pedestrian tasks of recordkeeping and adjudication and governance. They wove their legitimacy into ordinary people’s water rights, their inheritances, their personal disputes...” https://extradeadjcb.substack.com/p/how-did-the-taliban-win
Spacecraft Scale Magnetospheric Protection from Galactic Cosmic Radiation https://www.nasa.gov/directorates/spacetech/niac/2018_Phase_I_Phase_II/Spacecraft_Scale_Magnetospheric_Protection_from_Galactic_Cosmic_Radiation/
The Unreasonable Effectiveness of Pronouns https://vectors.substack.com/p/the-unreasonable-effectiveness-of
“"Ending Medical Reversal" is an essential book for medical students, physicians, and anyone even peripherally involved in medicine; for everybody else, it's merely highly recommended.” https://twitter.com/Willyintheworld/status/1252338734843879430
Immigration to Denmark: impact on public finances and crime. Non-Western immigrants and descendants are almost 40% of rape convicts, but just 10% of the population. https://inquisitivebird.substack.com/p/the-effects-of-immigration-in-denmark
The U.S. Navy recently announced that it would be lowering the cut-off score for the IQ test it administers to new recruits. This has been tried before with disastrous consequences. https://gwern.net/review/mcnamara
De facto blasphemy laws in Great Britain https://whyevolutionistrue.com/2023/02/27/de-facto-blasphemy-laws-in-great-britain/
It seems pretty clear now that we're witnessing an AI "arms race" between the big tech companies. Will governments join in? I predict they will. At the latest when they realize that GPT-5 is capable of automating warfare.
A good reminder that we already have the technology to easily pass the Turing test of the average normie. But I doubt anyone will put all the pieces together before new multimodal models make such duct-tape AI obsolete.