In short An AI system provided nearly 1,300 human competitors in the annual American Crossword Puzzle Tournament to achieve the highest score.
The computer, named Dr. Fill, was the brainchild of computer scientist Matt Ginsberg, who designed its software to automatically fill in crosswords using a mix of “good old-fashioned AI” and more modern learning techniques. machine, according to on Slate.
It is able to solve multiple words quickly with fewer errors than its opponents. However, Dr. Fill was not eligible for the $ 3,000 cash prize, which instead went to the best player, someone named Tyler Hinman, which may not feel quite redundant.
Ginsberg’s machine contains a computer running a 64-core CPU and two GPUs, and is trained on tons of text scraped from Wikipedia to figure out words, and a database of crossword clues and their answers to sweep the competition questions. You can watch it in action below.
Google defends large language models such as those used by Google
In a new paper, researchers from Google and the University of California, Berkeley have outlined various ways to reduce the environmental impact of the large amount of energy consumed in the practice of text generation models such as used by Google.
Large language models are a particularly controversial area for The Chocolate Factory. AI Ethics research team co-leads Timnit Gebru and Margaret Mitchell, was expelled this year in a role detailed the power consumption and financial costs of these models as well as concerns over their undetectable nature.
Today, Google published a counter-study. Large language models are not as big of a carbon footprint if they are trained using resources from data centers that run efficiently in countries that use renewable energy, the internet giant argues. You can read the whole thing here.
The paper written by Gebru, professor of computational linguistics Emily M. Bender, and others were shot by Google for allegedly not including enough references to relevant research. What is sad here is that Google’s latest paper fails to mention or reference the role of Gebru and Bender in their study. One of the researchers confirmed that they would add a hat-tip to the pair in an updated version of their study.
Oops, thanks for pointing out. We have several texts discussing Bender et al. in an earlier draft, but looks like an editing pass to reconstruct some text from a few weeks ago by accident. We will upload a new version soon with that discussion restored.
– Quoc Le (@quocleix) April 22, 2021
Be careful with satellite imagery
Academics warn of the potential dangers of fake satellite images created by AI.
A team of geographers led by the University of Washington in the US demonstrated how machine learning algorithms can be trained to spit out fake geospatial images. The outputs can be used to interfere with applications that rely on satellite imagery, such as Google Earth or even military software.
“It’s not just the things in Photoshopping. It makes the data look unrealistic,” said Bo Zhao, assistant professor of geography at UW and lead author of the study published in the journal Cartography and Geographic Information Science, this week. “The strategies are already there. We are only trying to expose the possibility of using both techniques, and of the need to develop a coping strategy for it. “
Zhao showed examples of how real images from cities can be manipulated by pasting on fake buildings to create townhouses or adding false fire to mimic natural disasters. While it will take more than deepfakes to attack real software systems, researchers are stepping up here now in the hope that they could be one step ahead of the threat.
iGiant to create new AI jobs
Apple has pledged to invest $ 430bn in the US to employ 20,000 new staff focused on emerging technologies, such as AI on new chips, over the next five years. Apple also plans to spend $ 1bn to launch a new campus in North Carolina, with nearly 3,000 employees working on advanced research and development.
“In this moment of recovery and rebuilding, Apple has doubled our commitment to U.S. innovation and manufacturing with a generational investment that reaches communities in all 50 states,” the Apple CEO. Tim Cook, announced this Sunday.
“We’re creating jobs in a wide range of fields – from 5G to silicon engineering to artificial intelligence – investing in the next generation of innovative new businesses, and in all our work, building towards a greener and more fair future. “
New SiFive AI chip made by Samsung coming soon
An AI accelerator system-on-chip developed in collaboration between SiFive and a mystery partner is set to be made by the Samsung chip.
Not much is known about the chip, except that it is based on a 14nm FinFET design and contains SiFive RISC-V cores as well as PCIe Gen connectivity. 4 and 32-bit LPDDR4 quad-channel memory.
SiFive did not disclose to whom the chip was for or when it would be shipped for mass production.
“The partnership with Samsung Foundry has accelerated SiFive’s ability to deliver our highly efficient and configurable approach for SoC design and implementation,” Yunsup Lee, CTO of SiFive, said in a statement.
“We are excited to continue to partner with Samsung Foundry as we launch our latest SiFive Intelligence products to accelerate the development of the next generation of AI SoCs with Samsung’s advanced process technology.” ®