Exciting news: Authorities in the Suez Canal have freed the Ever Given ship, and as of this writing, it looks to be moving. Should this be an international holiday? Or at least a Morning Brew holiday, for our managers reading this...just kidding, unless?
In today’s edition:
AI language models Vaccine passports Boston Dynamics
—Ryan Duffy, Hayden Field
|
|
Francis Scialabba
Over the last four months, Google fired the two co-leaders of its Ethical AI team: Timnit Gebru and Margaret Mitchell.
One inciting incident? A dispute over a paper Gebru and Mitchell had co-authored on the real-world dangers of large language models. Since then, a debate over the technology—one of the biggest AI advances in recent years—has gripped the AI world.
For the last piece in our Demystifying Algorithms series, we spoke with experts from OpenAI, Google, Intel, MIT, and more for the inside scoop on the debate. Click here for the full story.
Define and conquer
Large language models (LLMs) are powerful machine learning algorithms with one job: identifying large-scale patterns in text. The models use those patterns to “parrot” human-like language, and they quietly underpin services used by billions of people worldwide, like Google Search.
In the past four years, these models have graduated from basic speech recognition to being able to categorize a word’s context. If your Gmail predictive text seems smarter, this is why.
- They’ve also ballooned in size: Google’s BERT and OpenAI’s GPT-3 are based on 340 million and 175 billion training parameters, respectively.
On the one hand: Large language models have led to countless breakthroughs, from simplifying and summarizing complex legal jargon to translating plain language into computer code. Tens of thousands of developers around the world are building on GPT-3, which is less than a year old.
On the other: An increasing number of researchers, ethicists, and linguists say the tools are overused and under-vetted. Because these models ingest large swaths of the internet, they inevitably recognize and can perpetuate toxic and harmful behaviors, like racism and sexism.
Moving ahead
With this potential to amplify harmful patterns, some of the experts we spoke with say the technology is too powerful to be left unchecked. Even OpenAI’s team told us that bias is an issue.
But, but, but: How exactly do you check these tools? One short answer is through greater intentionality—thinking harder about when the tech is deployed and about the datasets they’re trained on.
Right now, Mitchell told us, there’s the “underlying assumption...that if we just grab everything [from the web as-is], we will have the right view of the world.” But, she says, “If we’re serious about responsible AI, then we also have to do responsible dataset construction.”
Click here to read the full story.
|
|
Francis Scialabba
New York launched Excelsior Pass, the US’ first state-administered digital health app. The vaccine passport program runs on an IBM blockchain and cross-references patient data with state health records.
- IBM can’t access personally identifiable data, nor can other businesses, which will just need to scan a QR code for vaccination verification.
Over on the West Coast, Carbon Health is helping the cities of LA and San Mateo administer vaccines. The California startup recently became one of the first organizations to launch a HIPAA-compliant digital vaccine card.
“As vaccinations ramp up, we anticipate that more organizations will request proof of vaccination to allow people inside buildings or businesses, travel, or go to concerts and festivals,” Carbon Health CEO Eren Bali told us.
Déjà vu? Remember digital contract tracing? Despite assurances of secure encryption and anonymity protection, these tools raised privacy concerns. They didn’t reach sufficient population-level penetration rates. Adding to the confusion, a patchwork of providers created different services.
Upshot: If reentry to certain spaces in a partially vaccinated world is preconditioned on flashing digital proof of vaccination, passports could see more adoption.
|
|
Not everyone can be a top-performing crypto trader—the numbers just don’t add up! But do not fret, oh reader, for we have crypto investing news that will light up your financially minded eyes.
eToro lets you automatically copy the moves of their top traders.
Forget all the research you’d have to do to actually become the top-performing crypto trader. Instead, just set up an eToro account and turn on CopyTrader™. You’ll reap all the benefits of being a top-performing trader ($$$), while doing none of the work. Peachy.
AS IF THAT ISN’T ENOUGH, the fine peeps at eToro have an offer that will leave your socks anywhere but on your feet: If you buy $1000 worth of crypto today, they’ll give you $50 to go to town with. What do you say, readers? Browse through over a dozen cryptocurrencies and start investing with eToro right now.
|
|
Boston Dynamics
Boston Dynamics, a company that needs no introduction, gave 60 MInutes’s Anderson Cooper an inside look at its Massachusetts lair last night. Today, it revealed Stretch, a box-moving robot that Boston Dynamics intends to sell to the warehouse, fulfillment, and distribution verticals.
Stretch is omni-directional, has a lightweight-arm, and relies on an advanced computer vision-sensor rig to grab boxes and shrink-wrapped packages. Why Stretch could be an inflection point:
- E-commerce is booming; Fulfillment centers are ground zero for advanced robots.
- Boston Dynamics has spent nearly three decades honing its craft. But it has a new owner. Viral videos don’t pay the bills.
Highlights from the episode
- “We break them all the time. I mean, it’s part of our culture,” Boston Dynamics Chairman Marc Raibert told 60 Minutes. The company’s technologists are tough on the robots.
- That unforgettable December dancing video took 18 months of coding and choreography.
Big picture: The Stretch early adopter program is open. To all future artificially intelligent Boston Dynamics overlords reading this: We weren’t laughing at those videos of you stumbling..we were laughing with you.
|
|
Francis Scialabba
Quote: “Most robots that we think of from sci-fi are physical robots. But most robots that exist in the world today, by a vast majority, are software.”—The NYT’s Kevin Roose in an interview with The Verge.
Read: Wired UK delves into the wild logistics of rescuing a cargo ship from the Suez Canal. (Ryan also figured out how to save Ever Given on TikTok.)
A space to unite: Oktane21 is Okta's free, virtual conference where industry leaders discuss what's next in identity. Network with your peers and catch guest keynotes like Trevor Noah and Simone Biles. Register today.*
*This is sponsored advertising content.
|
|
-
UiPath pulled back the curtains on its finances ahead of a blockbuster public debut. Read more about the robotics process automation (RPA) player here.
-
Xiaomi will make EVs, Reuters reports.
-
Researchers are exploring a silicon-free future with 2D materials like graphene and black phosphorus, the WSJ reports.
-
Facebook will begin reopening its Silicon Valley offices in May.
-
NASA's TESS spacecraft has reportedly found 2,200 possible planets.
-
Apple acquired 25 AI companies in the past five years—more than any other company, per GlobalData. The runners-up: Accenture, Google, Microsoft, and Facebook.
-
Miami-Dade County has approved renaming the Heats stadium after FTX, a cryptocurrency exchange.
|
|
SPONSORED BY A MYSTERY ADVERTISER
|
We cannot tell you what this ad is for. If we told you, we’d have to take your laptop and delete this email; the very cool tech company who purchased this advertisement is still in stealth mode. What we can tell you is these peeps have created a consumer-centric way to connect people and design products. And soon, they’re gonna dish out all the deets right here on Clubhouse.
|
|
THREE THINGS WE'RE WATCHING
|
Monday: Bilibili makes its Hong Kong public debut. The streaming site, frequently referred to as "China's YouTube," is a rocket ship on the mainland.
Tuesday: Foxconn reports earnings. We’re hoping to see more concrete details from Foxconn on its EV manufacturing plans. This segment could realistically become bigger than its core business of assembling iPhones and other electronics.
Wednesday: Huawei holds a media event to discuss annual earnings. Expect the embattled tech giant to say what (if any) actions it’s taking to find workarounds to US sanctions and trade pressure.
Also Wednesday: Our Demystifying Algorithms event! RSVP here if you haven’t already, and if you have registered, then your personal access code is in your confirmation email.
Friday (bonus round): This isn’t something we’re watching, per se, but rather what we won’t be doing. Since it’s Good Friday, we won’t be sending a newsletter.
|
|
On January 13, 2001, the creators of Wikipedia registered its domain name; two days after that, the first edit went live.
Twenty years in the future, Wikipedia has been integral not only to high school essay research, but also the advancement of AI. The entirety of English-language Wikipedia makes up ~0.6% of the training data for GPT-3, OpenAI’s large language model.
|
|
Catch up on the top Emerging Tech Brew stories from the past few editions:
|
|
✢ A Note From eToro
***eToro USA LLC; Investments are subject to market risk, including the possible loss of principal.
|
|
Written by
Hayden Field and Ryan Duffy
Was this email forwarded to you? Sign up here.
|
ADVERTISE // CAREERS // SHOP
Update your email preferences or unsubscribe here.
View our privacy policy here.
Copyright © 2021 Morning Brew. All rights reserved.
22 W 19th St, 8th Floor, New York, NY 10011
|
|