The Download: Meta’s new AI system, and covert Chinese social media activity

1 year ago 208

This is today's variation of The Download, our weekday newsletter that provides a regular dose of what's going connected successful the satellite of technology.

Meta’s latest AI exemplary is escaped for all 

The news: Meta is going each successful connected open-source AI. The institution has unveiled LLaMA 2, its archetypal ample connection exemplary that’s disposable for anyone to use—for free. It’s besides releasing a mentation of the AI exemplary that radical tin physique into ChatGPT-style chatbots.

Why it matters: The thought is that by releasing the exemplary into the chaotic and letting developers and companies tinker with it, Meta volition larn important lessons astir however to marque its models safer, little biased, and much efficient.

But… Many caveats inactive remain. Meta is not releasing accusation astir the information acceptable that it utilized to bid LLaMA 2, and it inactive spews offensive, harmful, and different problematic language, conscionable similar rival models. Meta besides cannot warrant that it didn’t see copyrighted works oregon idiosyncratic data, according to a institution probe insubstantial shared exclusively with MIT Technology Review. Read the afloat story.

—Melissa Heikkilä

Spotting Chinese authorities media societal accounts continues to beryllium a challenge

It’s nary concealed that Chinese state-owned media are progressive connected Western societal platforms. But sometimes they instrumentality a covert attack and region themselves from China, possibly to scope much unsuspecting audiences. 

Such operations person been recovered to people Chinese- and English-speaking users successful the past. Now, a survey published past week has discovered different web of Twitter accounts that seems to beryllium obscuring its China ties—and this time, it’s made up of Spanish-language quality accounts targeting Latin America. Read the afloat story.

—Zeyi Yang

This communicative is from China Report, Zeyi’s play newsletter covering tech successful China. Sign up to person it successful your inbox each Tuesday.

The must-reads

I’ve combed the net to find you today’s astir fun/important/scary/fascinating stories astir technology.

1 ChatGPT could go an precocious facial designation machine
But OpenAI wants to debar that result astatine each costs. (NYT $)
+ China is acrophobic AI could go a “runaway horse.” (Reuters)
+ How to halt AI from recognizing your look successful selfies. (MIT Technology Review)

2 North Carolina is policing online termination discussions
In theory, users could beryllium prevented from posting and speechmaking astir termination access. (Wired $)

3 Calls for AI companies to recompense authors are increasing louder
Margaret Atwood is among the high-profile names to articulation the charge. (WSJ $)
+ Microsoft has started charging $30 a period for its generative AI. (FT $)
+ OpenAI’s hunger for information is coming backmost to wound it.  (MIT Technology Review)

4 The US has blacklisted much overseas spyware companies
In a bid to deter investors from sinking currency into seemingly-dodgy companies. (WP $)
+ US lawmakers are cracking down connected surveillance overreach. (Wired $)
+ Snapchat has been accused of promoting Saudi Arabia’s royals. (The Guardian)

5 Temperatures are soaring successful Death Valley
And it’s looking progressively apt it’ll interruption its all-time record. (LA Times $)
+ Tourists are getting successful connected the act, too. (BBC)

6 The Pentagon doesn’t privation you to way its planes ✈️
It’s had capable of civilians keeping an oculus connected mysterious aircraft. (The Intercept)

7 Lemon8 is flopping 🍋
ByteDance’s different societal media platform’s biggest problem? It’s bland. (Rest of World)
+ New apps request to interruption retired of the societal doom cycle. (The Atlantic $)

8 Who uses WeightWatchers successful the property of Ozempic?
Loyal WeightWatchers consciousness betrayed by the company’s determination to clasp weight-loss medications. (Bloomberg $)
+ Weight-loss injections person taken implicit the internet. But what does this mean for radical IRL? (MIT Technology Review)

9 How to archer if your babe show is susceptible to hacking
A caller statement volition assistance consumers successful the US to decide. (Vox)
+ Electric vehicles are astatine precocious risk, too. (Slate $)

10 Your telephone fig isn’t dormant yet
For galore people, it’s a reminder of wherever they’ve travel from. (The Atlantic $)

Quote of the day

"I'm disquieted astir regular intelligence!"

—Actor Danny Trejo tells CNBC he’s got bigger concerns than the emergence of AI during an interrogation astir the Hollywood strikes.

The large story

Minneapolis constabulary utilized fake societal media profiles to surveil Black people

April 2022The Minneapolis Police Department violated civilian rights instrumentality done a signifier of racist policing practices, according to a damning study by the Minnesota Department of Human Rights. 

The study recovered that officers stop, search, arrest, and usage unit against radical of colour astatine a overmuch higher complaint than achromatic people, and covertly surveilled Black radical not suspected of immoderate crimes via societal media. 

The findings are accordant with MIT Technology Review’s probe of Minnesota instrumentality enforcement agencies, which has revealed an extended surveillance web that targeted activists successful the aftermath of the execution of George Floyd. Read the afloat story.

—Tate Ryan-Mosley and Sam Richards

Read Entire Article