hey look, you can train GPT-2 in under an hour*
* using about $40M worth of hardware. still impressive, though
https://devblogs.nvidia.com/training-bert-with-gpus/

what is color, really?
https://medium.com/@rawwerks/what-is-structural-color-536ee6fe46d4

https://steganography.live/info
Neural Linguistic Steganography - encode your message with a secret key into innocent-looking text with GPT-2!
— I wonder how secure does that look from the cryptanalysis point of view; though one can always cover the message with AES first :)

The TimeViz Browser —
A Visual Survey of Visualization Techniques for Time-Oriented Data
https://vcg.informatik.uni-rostock.de/~ct/timeviz/timeviz.html

sentient amoeba collectives
https://www.nature.com/news/2008/080123/full/451385a.html

Symmetric Satellite Swarms and Choreographic Crystals
https://arxiv.org/pdf/1407.5876v2.pdf

an optical metamaterial, finally! now they just need to add a gradient to fully transparent plastic on the edges and it would be great #gits #camo

if you though it's just rich text input that's complicated...
https://lord.io/blog/2019/text-editing-hates-you-too/

studying wildlife with drones!
https://twitter.com/HerdHover?s=20

Writing userspace USB drivers for abandoned devices

> a hell of a roller coaster in USB protocol details, kernel debugging/module reverse engineering, and general video format decoding!
https://blog.benjojo.co.uk/post/userspace-usb-drivers

that's why defense in depth is important
https://twitter.com/hashbreaker/status/1198395661403725824

MMORPG's are awesome emergent behavior playgrounds
https://www.rockpapershotgun.com/2019/11/18/planet-zoo-is-temporarily-a-game-about-mass-producing-knackered-warthogs/

Nice to see that even in 2019, there's still room for new short, fast, and practical algorithms. Though note that xor filters aren't a replacement for bloom filters, they're more of an optimization of one particular use case.
https://lemire.me/blog/2019/12/19/xor-filters-faster-and-smaller-than-bloom-filters/

Finally, a book that bridges ML with control theory and dynamical systems (and a bit of chaos theory).

Well written, comprehensive overview of the field. Got quite a few insights and filled the gaps.

Also, I was totally surprised to find that dynamical systems research is nowhere near the 4th paradigm (data-intensive scientific discovery). I thought it's my Google-fu is failing me, but it seems there's really no (published) math yet that enables one to robustly identify, describe and simulate anything beyond simple quadratic or periodic processes :(

Now longing for something similar connecting dynamical systems to game theory, catastrophe theory, emergent behaviors, and multi-agent modeling. And something about time-varying systems/processes :)

https://www.goodreads.com/book/show/40714461-data-driven-science-and-engineering

#controltheory #book #dynamicalsystems #bookreview

I haven't dived into underlying math yet, but it certainly looks interesting
https://research.kudelskisecurity.com/2019/11/25/forget-homomorphic-encryption-here-comes-functional-encryption/

TLDR: when people can't compare themselves to other people they cooperate and create wealth. when status is codified, they play status games instead.
https://alexdanco.com/2020/01/23/social-capital-in-silicon-valley/