It’s not simple being one in all Silicon Valley’s favourite benchmarks.
SWE-Bench (pronounced “swee bench”) launched in November 2024 as a approach to consider an AI mannequin’s coding ability. It has since shortly develop into one of the in style exams in AI. A SWE-Bench rating has develop into a mainstay of main mannequin releases from OpenAI, Anthropic, and Google—and outdoors of basis fashions, the fine-tuners at AI corporations are in fixed competitors to see who can rise above the pack.
Regardless of all of the fervor, this isn’t precisely a truthful evaluation of which mannequin is “higher.” Entrants have begun to sport the system—which is pushing many others to wonder if there’s a greater approach to really measure AI achievement. Learn the total story.
—Russell Brandom
Did solar energy trigger Spain’s blackout?
At roughly noon on Monday, April 28, the lights went out in Spain. The grid blackout, which prolonged into components of Portugal and France, affected tens of hundreds of thousands of individuals—flights had been grounded, cell networks went down, and companies closed for the day.
Over every week later, officers nonetheless aren’t completely positive what occurred, however some have prompt that renewables could have performed a job, as a result of simply earlier than the outage occurred, wind and photo voltaic accounted for about 70% of electrical energy technology. Others, together with Spanish authorities officers, insist that it’s too early to assign blame.
It’ll take weeks to get the total report, however we do know a couple of issues about what occurred. Listed here are a couple of takeaways that would assist our future grid.
—Casey Crownhart
This text is from The Spark, MIT Expertise Assessment’s weekly local weather publication. To obtain it in your inbox each Wednesday, join right here.
The must-reads
I’ve combed the web to seek out you in the present day’s most enjoyable/necessary/scary/fascinating tales about expertise.
1 The Trump administration will repeal some international chip curbs
It’s drawing up new guidelines that prioritize direct negotiations with numerous nations. (Bloomberg $)
+ The curbs have at all times been leaky anyway. (Economist $)
2 India and Pakistan have accused one another of in a single day drone assaults
The battle between the 2 nations is quickly escalating. (The Guardian)
+ Pakistan claims to have shot down 25 drones in its airspace. (Reuters)
+ Mass-market army drones have modified the best way wars are fought. (MIT Expertise Assessment)
3 The FDA is thinking about utilizing AI for drug analysis
And has met with OpenAI to listen to extra about methods to do it. (Wired $)
+ An AI-driven “manufacturing facility of medication” claims to have hit a giant milestone. (MIT Expertise Assessment)
4 The US is pushing nations dealing with its tariffs to undertake Starlink
Authorities officers in India and different nations have quick tracked approvals. (WP $)
+ India not too long ago introduced new guidelines for satellite tv for pc web suppliers. (Remainder of World)
5 Apple is overhauling its Safari browser to give attention to AI search
Its search quantity is down for the primary time in 22 years. (The Verge)
+ Apple exec Eddy Cue thinks AI search will change conventional engines like google. (Bloomberg $)
+ AI means the top of web search as we’ve identified it. (MIT Expertise Assessment)
6 Mark Zuckerberg is betting huge on AI chatbots
He’s on a media appeal offensive to persuade us that AI associates are the long run. (WSJ $)
+ The AI relationship revolution is already right here. (MIT Expertise Assessment)
7 College students can’t wean themselves off ChatGPT
And consultants concern that they’ll emerge into the workforce basically illiterate. (NY Magazine $)
+ Some educators consider that AI highlights how the methods we train want to alter. (MIT Expertise Assessment)
8 We don’t actually understand how reminiscence works 🧠
However these researchers are doing their finest to seek out out. (Quanta Journal)
9 The overwhelming majority of the ocean depths are nonetheless unexplored
What lies beneath is a thriller. (New Scientist $)
+ Meet the divers attempting to determine how deep people can go. (MIT Expertise Assessment)
10 Pet psychics are taking up TikTok 🔮
However does your furry good friend have something to say?(NYT $)
+ People are nonetheless higher than AI at futuregazing—for now. (Vox)
+ How DeepSeek turned a fortune teller for China’s youth. (MIT Expertise Assessment)
Quote of the day
“It’s like dwelling in hell.”
—Elizabeth Martorana, a Virginia resident, describes what it’s prefer to stay in a growth zone for Amazon, Microsoft, and Google knowledge facilities, Semafor reviews.
Another factor
How Antarctica’s historical past of isolation is ending—due to Starlink
“This is likely one of the least visited locations on planet Earth and I received to open the door,” Matty Jordan, a building specialist at New Zealand’s Scott Base in Antarctica, wrote within the caption to the video he posted to Instagram and TikTok in October 2023.
Within the video, he guides viewers by way of the hut, declaring the place the boys of Ernest Shackleton’s 1907 expedition lived and labored.
The video has racked up hundreds of thousands of views from all around the world. It’s additionally type of a miracle: till very not too long ago, those that lived and labored on Antarctic bases had no hope of speaking so readily with the skin world.
That’s beginning to change, due to Starlink, the satellite tv for pc constellation developed by Elon Musk’s firm SpaceX to service the world with high-speed broadband web. Learn the total story.
—Allegra Rosenberg
We are able to nonetheless have good issues
A spot for consolation, enjoyable and distraction to brighten up your day. (Acquired any concepts? Drop me a line or skeet ’em at me.)
+ Does Boston nonetheless drink? Not in the identical approach it used to.
+ The place within the US you need to arrange camp to stargaze proper now.
+ Wow: this New Zealand snail lays eggs from its neck. 🐌
+ Jurassic World Rebirth is coming: and it appears to be like suitably bonkers.
