Links for 2023-04-26
Track-Anything is a flexible and interactive tool for video object tracking and segmentation. It is developed upon Segment Anything, can specify anything to track and segment via user clicks only. During tracking, users can flexibly change the objects they wanna track or correct the region of interest if there are any ambiguities.
Project page: https://github.com/gaomingqi/Track-Anything
Speed Is All You Need: On-Device Acceleration of Large Diffusion Models via GPU-Aware Optimizations — “We present a series of implementation optimizations for large diffusion models that achieve the fastest reported inference latency to-date (under 12 seconds for Stable Diffusion 1.4 without int8 quantization on Samsung S23 Ultra for a 512x512 image with 20 iterations) on GPU-equipped mobile devices.” https://arxiv.org/abs/2304.11267
Toby Ord: What a simple toy could tell the past about their future http://www.tobyord.com/writing/a-childs-plaything
Tech Billionaires Bet on Fusion as Holy Grail for Business [Wall Street Journal] https://archive.is/JZ7lX
“I've heard people be somewhat optimistic about this AI guideline from China...No one in China is talking about it. They're talking about how much the Baidu LLM sucks in comparison to ChatGPT.” https://www.lesswrong.com/posts/EAwe7smpmFQi2653G/my-assessment-of-the-chinese-ai-safety-community
Pen and Paper Exercises in Machine Learning: A handy collection of pen-and-paper exercises in machine learning. Covers foundational topics related to machine learning such as linear algebra, optimization, graphical models and etc... https://arxiv.org/abs/2206.13446
Inducing anxiety in large language models increases exploration and bias https://arxiv.org/abs/2304.11111
1% of the population are accountable for 63% of all violent crime convictions (and 3.9% of the population are accountable for 100% of all violent crime convictions) https://link.springer.com/article/10.1007/s00127-013-0783-y
“Nearly a third of all shoplifting arrests in New York City last year involved just 327 people, the police said. Collectively, they were arrested and rearrested more than 6,000 times.” [The New York Times] https://archive.is/GWkwe
When few do great harm. Power laws in criminal behavior. https://inquisitivebird.substack.com/p/when-few-do-great-harm
Calls grow for drastic steps to address Japan's sliding birth rate: 'To raise the number of births sharply, [economist] Oguro proposes giving parents 10 million yen if they have a third child, and 10 million for each after that as an incentive to increase the size of families.' https://mainichi.jp/english/articles/20230410/p2g/00m/0na/017000c
“There are lots of articles about the low birthrate in Korea. I will probably never have kids, and my reason is simple: I don’t want to go through what my parents went through. Like spending their salary on hagwon fees to nunchi game at work balancing long work hrs w child caring” https://twitter.com/juwonreports/status/1646494684427218945
“It’s interesting to note that, empirically, results with lower p-values are more likely to be genuine effects (i.e., not false positives). I looked at results for 325 psychology study replications, and when the original study p-value was at most 0.01, about 72% replicated. When p>0.01, only 48% did.” https://www.spencergreenberg.com/2022/12/demystifying-p-values/
4 different meanings of p-value (and how my thinking has changed) https://statmodeling.stat.columbia.edu/2023/04/14/4-different-meanings-of-p-value-and-how-my-thinking-has-changed/
Ranking Fields by p-Value Suspiciousness https://cremieux.substack.com/p/ranking-fields-by-p-value-suspiciousness
Preventing common misconceptions about Bayes Factors http://daniellakens.blogspot.com/2023/04/preventing-common-misconceptions-about.html
“Ignore Bark's "I'm done with this input" token and tell Bark to just keep generating more audio anyway.” https://twitter.com/jonathanfly/status/1650001584485552130
A new optical illusion by Akiyoshi Kitaoka: illusory departure