Galactica was intended to help scientists by summarizing academic papers, and solving math problems, among other tasks. But outsiders swiftly prompted the model to provide “scientific research” on the benefits of homophobia, anti-Semitism, suicide, eating glass, being white, or being a man—demonstrating not only how its botched launch was premature, but just how insufficient AI researchers’ efforts to make large language models safer have been. Read the full story.
This story is from The Algorithm, our weekly newsletter giving you the inside track on all things AI. Sign up to receive it in your inbox every Monday.
The must-reads
I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.
1 Verified anti-vax Twitter accounts are spreading health misinformation
And perfectly demonstrating the problem with charging for verification in the process. (The Guardian)
+ Maybe Twitter wasn’t helping your career as much as you thought it was. (Bloomberg $)
+ A deepfake of FTX’s founder has been circulating on Twitter. (Motherboard)
+ Some of Twitter’s liberal users are refusing to leave. (The Atlantic $)
+ Twitter’s layoff bloodbath is over, apparently. (The Verge)
+ Twitter’s potential collapse could wipe out vast records of recent human history. (MIT Technology Review)
2 NASA’s Orion spacecraft has completed its lunar flyby 🌒
Paving the way to humans returning to the moon. (Vox)
3 Amazon’s warehouse-watching algorithms are trained by humans
Poorly-paid workers in India and Costa Rica are reviewing thousands of hours of mind-numbing footage. (The Verge)
+ The AI data labeling industry is deeply exploitative. (MIT Technology Review)
4 How to make sense of climate change
Accepting the hard facts is the first step towards avoiding the grimmest ending for the planet. (New Yorker $)
+ The world’s richest nations have agreed to pay for global warming. (The Atlantic $)
+ These three charts show who is most to blame for climate change. (MIT Technology Review)
5 Apple uncovered a cybersecurity startup’s dodgy dealings
It compiled a document that illustrates the extent of Corellium’s relationships, including with the notorious NSO Group. (Wired $)
+ The hacking industry faces the end of an era. (MIT Technology Review)
6 The crypto industry is still feeling skittish
Shares in its largest exchange have dropped to an all-time low. (Bloomberg $)
+ The UK wants to crack down on gamified trading apps. (FT $)
7 The criminal justice system is failing neurodivergent people
Mimicking an online troll led to an autistic man being sentenced to five and a half years in jail. (Economist $)
8 Your workplace could be planning to scan your brain 🧠
All in the name of making you a more efficient employee. (IEEE Spectrum)
9 Facebook doesn’t care if your account is hacked
A series of new solutions to rescue accounts doesn’t seem to have had much effect. (WP $)
+ Parent company Meta is being sued in the UK over data collection. (Bloomberg $)
+ Independent artists are building the metaverse their way. (Motherboard)
10 Why training image-generating AIs on generated images is a bad idea
The ‘contaminated’ images will only confuse them. (New Scientist $)
+ Facial recognition software used by the US government reportedly didn’t work. (Motherboard)
+ The dark secret behind those cute AI-generated animal images. (MIT Technology Review)
Quote of the day
“It feels like they used to care more.”