Links for 2023-04-27
HOSNeRF: Dynamic Human-Object-Scene Neural Radiance Fields from a Single Video https://showlab.github.io/HOSNeRF/
We introduce HOSNeRF, a novel 360° free-viewpoint rendering method that reconstructs neural radiance fields for dynamic human-object-scene from a single monocular in-the-wild video. Our method enables pausing the video at any frame and rendering all scene details (dynamic humans, objects, and backgrounds) from arbitrary viewpoints.
Metaculus prediction market: less than a year from AGI to superintelligence https://www.metaculus.com/questions/4123/after-an-agi-is-created-how-many-months-will-it-be-before-the-first-superintelligence/
WizardLM: An LLM fine-tuned with synthetic data that is preferred to ChatGPT https://arxiv.org/abs/2304.12244
LLM+P: Empowering Large Language Models with Optimal Planning Proficiency https://arxiv.org/abs/2304.11477
📢AudioGPT: Understanding and Generating Speech, Music, Sound, and Talking Head https://arxiv.org/abs/2304.12995 (examples: https://twitter.com/RongjieH/status/1651054998594400256)
Generating and editing novel high-quality 3D natural scenes using a single example without any training. http://weiyuli.xyz/Sin3DGen/
TextMesh: Generation of Realistic 3D Meshes From Text Prompts https://arxiv.org/abs/2304.12439
LangChain explained - The hottest new Python framework https://www.youtube.com/watch?v=RoR4XJw8wIc
Fundamental Limitations of Alignment in Large Language Models https://arxiv.org/abs/2304.11082
RedPajama, a project to create leading open-source models, starts by reproducing LLaMA training dataset of over 1.2 trillion tokens https://www.together.xyz/blog/redpajama
A selection of reviews of the arguments for AGI ruin — that artificial general intelligence represents an existential risk to humanity. https://blog.jessriedel.com/2023/04/24/notable-reviews-of-arguments-for-agi-ruin/
The Brain is Not Close to Thermodynamic Limits on Computation https://www.lesswrong.com/posts/mW7pzgthMgFu9BiFX/the-brain-is-not-close-to-thermodynamic-limits-on
“…a dubious academic study claiming evidence that wealthy people are selfish and greedy. Academia loved it! 1300 citations! Journalists too...pre-registered replication: we find no evidence of a positive relationship between SES and unethical/selfish behavior...This was their "evidence": They paid some student assistants to observe if upper-class cars wait their turn at a 4-way intersection….Lesson: we know much of academia's output on race, sex, and sexuality is not to be trusted. Same is true on class.” https://twitter.com/whyvert/status/1647435796125810690
“So far it's astonishing how 1-sided my previous understanding of McCarthy was. He was clearly pointing at a real problem, the infiltration of, at the very least, Communist-sympathetic personnel throughout the State Department, from the 1930's onwards, & official inaction.” https://twitter.com/Willyintheworld/status/1645266024504205312
“This talk explains the big data problem from tiny neural implant sensors, and it also updates our research status of the ERC funded project "Intranet of Neurons".” https://youtu.be/sMUT72RaCm4
“But just pushing around the symbols won't get you there, to solve this problem you have to actually know something about calculus.” https://blog.plover.com/2023/04/15/#calculus-exercise
An elegant proof of Laplace’s rule of succession https://ericneyman.wordpress.com/2020/11/29/an-elegant-proof-of-laplaces-rule-of-succession/
Institutions that are seen as ideologically slanted, whether to the left (Hollywood, professors) or to the right (Catholic Church, cops) are less trusted even to some extent by people who share the ideological lean. https://psyarxiv.com/sfubr