Posts Tagged ‘quote’

Robo-writers: the rise and risks of language-generating AI

April 17, 2021

https://www.nature.com/articles/d41586-021-00530-0

GPT3

QT:{{”
A neural network’s size — and therefore its power — is roughly measured by how many parameters it has. These numbers define the strengths of the connections between neurons. More neurons and more connections means more parameters; GPT-3 has 175 billion. The next-largest language model of its kind has 17 billion (see ‘Larger language models’). (In January, Google released a model with 1.6 trillion parameters, but it’s a ‘sparse’ model, meaning each parameter does less work. In terms of performance, this is equivalent to a ‘dense’ model that has between 10 billion and 100 billion parameters, says William Fedus, a researcher at the University of Montreal, Canada, and Google.)
“}}

Double-Trouble Dominoes

April 6, 2021

Thought this was a great video of geometric growth in action by @uoftphysics Prof. Stephen Morris
https://www.youtube.com/watch?v=y97rBdSYbkg Starting with a ~5mm domino you get to a one ~1 m in size in 13 domino topples. Overall, this chain reaction represents a 2 billion-fold amplification!

QT:{{”
Dominoes are little toy rectangle tiles with dots on them. People like to stand them up on end in a long row, so when the first domino falls over, it knocks over the next domino, which knocks over the
next…pretty soon you have a rippling wave of falling dominoes. In this simple but amazing video, Stephen Morris shows that a little domino can knock over another one that’s 1 1/2 times as big in each direction. Then that one can tip over one that ‘s 1 1/2 times as big again. In this domino chain, the first one is only 1/4 inch tall, but the 13th domino weighs more than 100 pounds! If he kept going, the 29th domino would be as tall as the Empire State Building (1,454 feet). We’d all better get out of the way!
“}}

Creator behind viral Tom Cruise deepfakes says they’re a warning

March 29, 2021

https://news.yahoo.com/deepfake-videos-tom-cruise-went-150232443.html Quote: “Think about the implications for national security… about the implications if I create a video of Jeff Bezos saying that $AMZN stock profits are down 20% — how much can I move the markets?”

The Plague Year | The New Yorker

February 14, 2021

Nice discussion on the mistakes on aerosols + a vaccine development chronology

QT:{{
During the study’s initial stages, in February and March, the researchers were discomfited by the implications of their data. “The rapidity and degree of spread suggested it wasn’t a series of one-to-one-to-one transmissions,” Dr. Jacob Lemieux, a lead author, told me. Rather, it was “one-to-many transmission events.” That raised the question of airborne transmission. “At the time, the idea was heretical,” Lemieux said. “We were afraid to consider it, because it implied a whole different approach to infection control”—one in which masks played a central role, especially indoors. But the W.H.O. had repeatedly proclaimed that large respiratory droplets—as from a sneeze or a cough—drove the spread. This wasn’t based on data about the new virus, Lemieux said: “It was received wisdom based on how previous respiratory viruses had behaved. The global public-health
infrastructure has egg on its face. There’s a component of human nature that, until you get burned, you don’t know how hot the fire is.”

Until recently, one of the main imaging tools used by vaccinologists, the cryogenic electron microscope, wasn’t powerful enough to visualize viral proteins, which are incredibly tiny. “The whole field was referred to as blobology,” McLellan said. As a work-around, he developed expertise in X-ray crystallography. …McLellan showed me an “atomistic interpretation” of the F protein on the RSV virus—the visualization looked like a pile of Cheetos. It required a leap of imagination, but inside that murky world Graham and McLellan and their team manipulated the F protein, essentially by cloning it and inserting mutations that kept it strapped down. McLellan said, “There’s a lot of art to it.”

In 2013, Graham and McLellan published “Structure-Based Design of a Fusion Glycoprotein Vaccine for Respiratory Syncytial Virus,” in Science, demonstrating how they had stabilized the F protein in order to use it as an antigen—the part of a vaccine that sparks an immune response. Antibodies could now attack the F protein, vanquishing the virus. Graham and McLellan calculated that their vaccine could be given to a pregnant woman and provide enough antibodies to her baby to last for its first six months—the critical period. The paper opened a new front in the war against infectious disease. In a subsequent paper in Science, the team declared that it had established “clinical proof of concept for structure-based vaccine design,” portending “an era of precision vaccinology.”

Within a day after Graham and McLellan downloaded the sequence for sars-CoV-2, they had designed the modified proteins. The key accelerating factor was that they already knew how to alter the spike proteins of other coronaviruses. On January 13th, they turned their scheme over to Moderna, for manufacturing. Six weeks later, Moderna began shipping vials of vaccine for clinical trials. The development process was “an all-time record,” Graham told me. Typically, it takes years, if not decades, to go from formulating a vaccine to making a product ready to be tested: the process privileges safety and cost over speed.

After the vaccine was tested in animals, it became clear that Graham’s design choices had been sound. The first human trial began on March 16th. A week later, Moderna began scaling up production to a million doses per month.
“}}

https://www.newyorker.com/magazine/2021/01/04/the-plague-year

Checkerboard Chocolates

January 29, 2021

https://www.foodandwine.com/lifestyle/look-some-worlds-most-modern-and-beautiful-chocolates

A Look at Some of the World’s Most Modern and Beautiful Chocolates

QT:{{”
When we saw the photo of these amazing chocolates, the first thing we asked was, “Would anyone want to eat them?” How can you mess up something that looks so cool? These perfectly shaped mathematical wonders were made by Japanese company Nendo for the big Maison & Objet design show in Paris (that’s French for “House and Object”). The candies were made by pouring chocolate into a “mold,” a hollow shape cut so that when the candies cool down, they pop out looking like these. Suddenly a flat chocolate bar doesn’t seem as exciting as a spiky crown or checkerboard chunk — but in any shape, it all tastes good.
“}}

When a Virus Is the Cure | The New Yorker

January 26, 2021

https://www.newyorker.com/magazine/2020/12/21/when-a-virus-is-the-cure

QT:{{”
To make matters worse, fears of antibiotic resistance have, in recent decades, created a perverse incentive in medical research: new antibiotics, to remain effective, must be used sparingly, as so-called antibiotics of last resort. As a result, it is almost impossible to recoup the cost of developing them.
“}}

Great article!… But I don’t see how phages will solve the economic conundrum: the “perverse incentive [that]…new antibiotics, to remain effective, must be used sparingly…As a result, it is almost impossible to recoup the cost of developing them.”

‘Sanitizing’ functional genomics data may prevent privacy breaches | Spectrum | Autism Research News

January 16, 2021

QT:{{”

The new data ‘sanitization’ technique obscures regions of a
participant’s genome in a dataset to secure her privacy, and may encourage more people to participate in genetic studies, says lead investigator Mark Gerstein, professor of biomedical informatics at Yale University.

“If someone hacks into your email, you can get a new email address; or if someone hacks your credit card, you can get a new credit card,” Gerstein says. “If someone hacks your genome, you can’t get a new one.”

To determine which information and how much of it should remain private to prevent a linkage attack, Gerstein and his colleagues performed linkage attacks on existing genetic datasets. In one sample attack, they compared two publicly available databases and RNA sequencing results to successfully identify 421 individuals.

In another linkage attack, Gerstein’s team sequenced the RNA of two volunteers and shuffled these data into a larger dataset. They then obtained DNA samples from the volunteers’ used coffee cups and sequenced their genomes. Again, they could link the two individuals to their genomes with a high degree of certainty.

Based on what they learned from the mock linkage attacks, Gerstein’s team developed a technique to mask some variants from a person’s genetic data while preserving where those variants are located in the genome. To do this, they replace the genetic variant of concern with one from a reference genome; which variants are removed depend on the genetic conditions or predispositions someone’s genetic data reveals.

Introducing too many of these privacy-masking variants can decrease the usefulness of the data. But Gerstein’s team struck a balance that enables researchers to obtain data on gene-expression values but also enables study participants to dictate how much of their genetic information they wish to keep hidden.

“}}

https://www.spectrumnews.org/news/sanitizing-functional-genomics-data-may-prevent-privacy-breaches/

Top 20 Alexa commands you’ll wish you knew before

January 4, 2021

QT:{{”

To use it, just say “Alexa, drop in on all devices” to reach every Alexa-enabled speaker or “Alexa, drop in on [device name]” to target just one. Then just wait for the confirmation tone before speaking. ….
For example, “Alexa, remember I have a doctor’s appointment at 9 a.m. on Wednesday morning” or “Alexa, remember I put the check in my bedside table.”
“}}

https://www.komando.com/tech-tips/top-20-alexa-tricks/746804/

Monday night

December 21, 2020

QT:{{”
For the past year, Jupiter and Saturn have been dancing ever closer in the night sky. On the evening of Dec. 21, the very nadir of winter, they will be so close — one-tenth of one angular degree — that if your eyes are as bad as mine, they will appear as one blurry, bright planet. With a little optical aid you should be able to discern them as separate orbs, almost kissing, although Jupiter will be 450 million miles in front of the ringed Saturn.

Go out and look southwest in the hour after sunset. According to astronomers, the two planets have not appeared this close to each other in the sky since 1623 — but the sun’s glare then would have rendered them invisible. To find a conjunction that humans could see, you must skip all the way back to 1226, or ahead to March 15, 2080. You might wonder who will be around to witness that event
“}}
https://www.nytimes.com/2020/12/18/science/christmas-star-jupiter-saturn-conjunction.html This Solstice, Solace for the Darkness
A rare conjunction of planets serves as a reminder that there is more to the universe than just ourselves.

How to watch tonight’s ‘great conjunction’ of Jupiter and Saturn https://www.livescience.com/how-to-watch-great-conjunction-jupiter-saturn.html

visible near the horizon

Dressing for the Surveillance Age | The New Yorker

December 13, 2020

https://www.newyorker.com/magazine/2020/03/16/dressing-for-the-surveillance-age

Much of this @jmseabrook article about facial recognition will soon be applicable to genomic privacy & individuals’ attempts to protect themselves in this sphere as well…

QT:{{”
adversarialfashion.com

Adversarial examples demonstrate that deep-learning-based C.V. systems are only as good as their training data, and, because the data sets don’t contain all possible images, we can’t really trust them. In spite of the gains in accuracy and performance since the switch to deep learning, we still don’t understand or control how C.V. systems make decisions. “You train a neural network on inputs that represent the world a certain way,” Goldstein said. “And maybe something comes along that’s different—a lighting condition the system didn’t expect, or clothing it didn’t expect. It’s important that these systems are robust and don’t fail catastrophically when they stumble on something they aren’t trained on.”

The early work on adversarial attacks was done in the digital realm, using two-dimensional computer-generated images in a simulation. Making a three-dimensional adversarial object that could work in the real world is a lot harder, because shadows and partial views defeat the attack by introducing nuisance variables into the input image. A Belgian team of researchers printed adversarial images on
two-dimensional boards, which made them invisible to yolo when they held the boards in front of them. Scientists at Northeastern University and at the M.I.T.-I.B.M. Watson A.I. Lab created an adversarial design that they printed on a T-shirt. Goldstein and his students came up with a whole line of clothes—hoodies, sweatshirts, T-shirts.

“}}