04-09 00:00 · 硬件监管,嵌入式系统,供应链开放,法律合规,产业政策
John Deere to pay $99M in right-to-repair settlement
Farmers have been fighting John Deere for years over the right to repair their equipment, and this week, they finally reached a landmark settlement.
While the agricultural manufacturing giant pointed out in a statement that this is no admission of wrongdoing, it agreed to pay $99 million into a fund for farms and individuals who participated in a class action lawsuit. Specifically, that money is available to those involved who paid John Deere’s authorized dealers for large equipment repairs from January 2018. This means that plaintiffs will recover somewhere between 26% and 53% of overcharge damages, according to one of the court documents—far beyond the typical amount, which lands between 5% and 15%.
The settlement also includes an agreement by Deere to provide “the digital tools required for the maintenance, diagnosis, and repair” of tractors, combines, and other machinery for 10 years. That part is crucial, as farmers previously resorted to hacking their own equipment’s software just to get it up and running again. John Deere signed a memorandum of understanding in 2023 that partially addressed those concerns, providing third parties with the technology to diagnose and repair, as long as its intellectual property was safeguarded. Monday’s settlement seems to represent a much stronger (and legally binding) step forward.
Ripple effects of this battle have been felt far beyond the sales floors at John Deere dealers, as the price of used equipment skyrocketed in response to the infamous service difficulties. Even when the cost of older tractors doubled, farmers reasoned that they were still worth it because repairs were simpler and downtime was minimized. $60,000 for a 40-year-old machine became the norm.
A judge’s approval of the settlement is still required, though it seems likely. Still, John Deere isn’t out of the woods yet. It still faces another lawsuit from the United States Federal Trade Commission, in which the government organization accuses Deere of harmfully locking down the repair process.
It’s difficult to overstate the significance of this right-to-repair fight. While it has obvious implications for the ag industry, others like the automotive and even home appliance sectors are looking on. Any court ruling that might formally condemn John Deere of wrongdoing may set a precedent for others to follow. At a time when manufacturers want more and more control of their products after the point of sale, every little update feels incredibly high-stakes.
Got a tip or question for the author? Contact them directly: caleb@thedrive.com
▸ 展开全文
04-09 00:00 · 大模型训练,硬件效率,技术突破,成本降低,开源生态
MegaTrain: Full Precision Training of 100B+ Parameter LLMs on a Single GPU
Computer Science > Computation and Language
Title:MegaTrain: Full Precision Training of 100B+ Parameter Large Language Models on a Single GPU
View PDF HTML (experimental)Abstract:We present MegaTrain, a memory-centric system that efficiently trains 100B+ parameter large language models at full precision on a single GPU. Unlike traditional GPU-centric systems, MegaTrain stores parameters and optimizer states in host memory (CPU memory) and treats GPUs as transient compute engines. For each layer, we stream parameters in and compute gradients out, minimizing persistent device state. To battle the CPU-GPU bandwidth bottleneck, we adopt two key optimizations. 1) We introduce a pipelined double-buffered execution engine that overlaps parameter prefetching, computation, and gradient offloading across multiple CUDA streams, enabling continuous GPU execution. 2) We replace persistent autograd graphs with stateless layer templates, binding weights dynamically as they stream in, eliminating persistent graph metadata while providing flexibility in scheduling. On a single H200 GPU with 1.5TB host memory, MegaTrain reliably trains models up to 120B parameters. It also achieves 1.84$\times$ the training throughput of DeepSpeed ZeRO-3 with CPU offloading when training 14B models. MegaTrain also enables 7B model training with 512k token context on a single GH200.
Bibliographic and Citation Tools
Code, Data and Media Associated with this Article
Demos
Recommenders and Search Tools
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
▸ 展开全文
04-08 00:01 · 中东,局势,AlJazeera,中东
The US president nevertheless warned that the suspension was dependent on Iran reopening the Strait of Hormuz.
United States President Donald Trump has agreed to suspend his planned bombing of Iran for two weeks, following his threat that “a whole civilization will die tonight” if Tehran did not comply with his demands.
On Tuesday evening, within hours of the planned attack, Trump took to his platform Truth Social to announce he had reversed course.
Recommended Stories
list of 3 items- list 1 of 3Iran warns region and beyond as tension builds ahead of US ultimatum
- list 2 of 3Trump announces two-week ceasefire as Iran agrees to reopen Hormuz Strait
- list 3 of 3Trump agrees to pause attacks on Iran if Strait of Hormuz opens
He credited Pakistan for mediating the settlement but warned it came with conditions — namely that Iran re-open the Strait of Hormuz.
“Based on conversations with Prime Minister Shehbaz Sharif and Field Marshal Asim Munir, of Pakistan, and wherein they requested that I hold off the destructive force being sent tonight to Iran, and subject to the Islamic Republic of Iran agreeing to the COMPLETE, IMMEDIATE, and SAFE OPENING of the Strait of Hormuz, I agree to suspend the bombing and attack of Iran for a period of two weeks,” Trump wrote.
The message was posted online at 6:32pm US Eastern time (22:00 GMT), just under one and a half hours before Trump’s 8pm (00:00 GMT) deadline for the attack.
Shortly after Trump’s message, Iran’s Foreign Minister Seyed Abbas Araghchi confirmed that a tentative agreement had been reached.
“If attacks against Iran are halted, our Powerful Armed Forces will cease their defensive operations,” Araghchi wrote.
“For a period of two weeks, safe passage through the Strait of Hormuz will be possible via coordination with Iran’s Armed Forces and with due consideration of technical limitations.”
Araghchi likewise thanked Pakistan for its last-minute appeal to suspend the US bombing campaign. Iran’s Supreme National Security Council meanwhile hinted in a statement that the ceasefire could be extended beyond its initial two weeks, if negotiations proceed favourably.
Further talks are expected to take place in Islamabad over the coming weeks.
Dire warnings
In the lead-up to the Thursday evening announcement, there had been speculation that Trump might extend the deadline, as he has multiple times in recent weeks.
But tensions had mushroomed on Tuesday as the deadline inched closer — and after Trump himself posted a menacing message on social media that morning.
“A whole civilization will die tonight, never to be brought back again,” Trump had written at the time. “I don’t want that to happen, but it probably will.”
He added that the evening deadline would mark “one of the most important moments in the long and complex history of the World”.
Trump had previously threatened to blow up power plants, bridges and other civilian infrastructure, actions that many legal experts said would amount to a war crime.
In his Tuesday evening message, the US president once again claimed victory over Iran and said that Tehran had delivered a feasible ceasefire proposal.
“This will be a double sided CEASEFIRE,” Trump wrote.
“The reason for doing so is that we have already met and exceeded all Military objectives, and are very far along with a definitive Agreement concerning Longterm PEACE with Iran, and PEACE in the Middle East. We received a 10 point proposal from Iran, and believe it is a workable basis on which to negotiate.”
Trump explained that the next two weeks would be spent finalising the agreement with Iran. But he voiced optimism that any past disagreements had been resolved.
“Almost all of the various points of past contention have been agreed to between the United States and Iran,” Trump said.
“It is an Honor to have this Longterm problem close to resolution,” he added.
‘The one who lit the fuse’
Al Jazeera correspondent Osama Bin Javaid noted that Trump’s latest announcement was likely to be welcomed by US allies in the Middle East, who had been bracing for a heavy attack on Iran and the possibility of counter-attacks.
“That is going to be a big sigh of relief, collectively from the region and beyond, because the alternative was so horrific,” Bin Javaid said.
“The level of anxiety around the region was extraordinary, and Donald Trump was the only person who could defuse it, because he’s the one who lit the fuse in the first place.”
Trump’s announcement was met with scepticism and confusion on the ground in Tehran, where some questioned if the war as a whole had been paused.
“It’s not clear, at least not to me, whether he’s suspending what is already going on since the beginning of the war entirely, or he’s extending the deadline for that ultimate destruction of the power plants,” Al Jazeera correspondent Mohamed Vall said from Tehran.
The war has been raging for more than five weeks, since Israel and the US launched a joint military offensive against Iran on February 28.
Trump and Israeli Prime Minister Benjamin Netanyahu have argued that the offensive was necessary to eliminate Iran
▸ 展开全文
04-08 00:01 · 中东,局势,AlJazeera,中东
Pakistan’s PM has asked Trump to extend the deadline for two weeks and for Iran to open the Strait of Hormuz.
Iran says it has accepted a two-week ceasefire, with negotiations to begin on Friday in Pakistan’s Islamabad, after US President Donald Trump said he would suspend attacks subject to Tehran agreeing to fully reopen the Strait of Hormuz.
Iran’s foreign minister says safe passage through the key waterway will be possible for a period of two weeks via coordination with Iranian armed forces.
▸ 展开全文
04-08 00:01 · 中东,局势,BBC,中东
Israel's military says troops fired at a vehicle driven by the Palestinian because they believed it posed "an immediate threat".
WHO suspends Gaza medical evacuations after contractor killed by Israeli troops
The World Health Organization has suspended medical evacuations from Gaza to Egypt via the Rafah crossing after a contractor was killed by Israeli troops on Monday.
WHO chief Tedros Adhanom Ghebreyesus said the contractor – named locally as Majdi Aslan, 54 – was killed in a "security incident" and that two staff were present but not injured. He gave no further details.
Gaza's Hamas-run health ministry said the contractor was driving a WHO-rented vehicle in southern Gaza when it was targeted by Israeli forces.
Israel's military said its troops "identified an unmarked vehicle approaching them and the Yellow Line, posing an immediate threat", referring to the line marking Israeli-held territory.
"In response, the troops fired warning shots. The vehicle continued to accelerate toward the troops, who then responded with additional fire that hit the vehicle," it added.
"The incident is under review."
Tedros said the WHO had suspended medical evacuations of sick and wounded Palestinians from Gaza via Rafah to Egypt until further notice.
"We call for the protection of civilians and humanitarian workers," he added.
Hanan Balkhy, the WHO's regional director, described the killing of the contractor as "a devastating loss" and warned that the suspension of medical evacuations was "cutting off a critical pathway to care for patients".
At al-Aqsa hospital in central Gaza, a colleague of Majdi Aslan told reporters the vehicle he was driving was clearly marked with the WHO's logo on all sides, and that it had been at the front of a co-ordinated convoy transporting patients to the Rafah crossing when it came under fire from an Israeli tank.
"The tank came out directly and targeted the driver so that he would stop. There were three people in the vehicle; the driver is the one who paid the price for the situation we are living in," Raed Aslan said. "What ceasefire is this? Every day we have someone killed."
A medic at the hospital, Dr Fathi al-Lulu, said Majdi Aslan had been targeted while "co-ordinating the transfer of patients from the Gaza Strip to the Egyptian side".
The Rafah crossing was reopened for the limited movement of people in both directions in early February as part of October's ceasefire deal between Israel and Hamas, having been largely closed since the Gazan side was captured by Israeli forces in May 2024.
Israeli authorities shut the crossing on 28 February following the start of the US-Israeli war with Iran, but it reopened again after 20 days.
Israeli military body Cogat says approximately 1,150 people have left Gaza via the crossing over the past two months and that the same number have entered Gaza.
More than 18,000 Palestinian patients are waiting to leave Gaza for treatment, according to local hospitals and the WHO.
Israel and Hamas have accused each other of repeatedly violating the ceasefire which they agreed almost six months ago.
At least 733 Palestinians have been killed in Gaza since a ceasefire between Israel and Hamas began on 10 October, according to the territory's health ministry.
The Israeli military has said five of its soldiers have been killed in attacks by Palestinian groups over the same period.
▸ 展开全文
04-08 00:01 · 航天,科技政策,国际合作,基础设施,研发投入
The Artemis II mission has been near flawless to date, but has the test flight shown Nasa is ready to send humans to the lunar surface?
Has Artemis II shown we can land on the Moon again?
Nasa's Artemis II mission has passed every major test since its launch on 1 April, with its rocket, spacecraft and crew performing better than engineers had dared to hope for.
The mission's first six days have shown that the Orion capsule works as designed with people on board for the first time - something no simulator could prove.
Perhaps its greatest achievement, though, is through the actions of the Artemis crew, which have generated hope, agency and optimism for a world appearing to be in desperate need of inspiration.
But the bigger question remains - is a Moon landing by 2028, as Nasa and President Trump want, now really an achievable goal?
What Artemis II has taught us so far
A few days after Nasa's Space Launch System (SLS) reached the launch pad at Kennedy Space Centre, the most important lesson about Artemis II had already been learned.
After two scrubbed launches in February and again in March because of separate technical issues, Nasa Administrator Jared Isaacman said "launching a rocket as important and as complex as SLS every three years is not a path to success."
The previous uncrewed Artemis I mission took off in November 2022.
The agency, he said, had to stop treating each rocket "like a work of art" and start launching with the frequency of a programme that means serious business.
It was, in effect, a declaration that relearning the same lessons every three years had to stop.
That matters, because it reframes everything that has followed. And judged against that ambition, what has the mission shown us in the six days since Reid Wiseman, Victor Glover, Christina Koch and Jeremy Hansen lifted off on April 1st?
The short answer is more than even the optimists dared hope for.
A Rocket that did the job
The SLS generated 8.8 million pounds of thrust at liftoff and, by every measure engineers care about, performed to plan. Each phase of the ascent was, in the understated language of mission control, "nominal": maximum dynamic pressure, main engine cut-off and booster separation.
Two of the three planned course corrections on the way to the Moon were scrapped because the trajectory was already so accurate they were not needed. As Dr Simeon Barber, space scientist at the Open Univertsity, put it: "Credit to them - they got it right the first time."
A day after launch came the critical moment. Orion fired its main engine for five minutes and fifty five seconds - known as the translunar injection burn - putting the spacecraft on a looping path to the Moon with no further major manoeuvres required.
The powerful engine burn was "flawless" according to the head of the Artemis programme, Dr Lori Glaze.
Humans in the machine
The official purpose of this mission is to put people inside Orion and find out what happens - not just to the spacecraft, but to the interaction between crew and machine. What has unfolded is precisely what was anticipated, and precisely what could not have been learned in a simulator.
There have been toilet problems. A water dispenser issue requiring the crew to bag water as a precaution. A minor redundancy loss in one of the helium systems was mentioned at an early press conference and quietly resolved.
As Barber observed: "This is all about putting humans in the loop - these pesky humans that press buttons and breathe carbon dioxide and want air conditioning and want to use the toilet. It was all about how the system works with those guys on board."
The engineers monitoring Orion's CO2 removal system through back-to-back exercise sessions, or testing how the spacecraft handles with thrusters deliberately disabled, are building the case that this vehicle is safe enough to carry people to the surface of the Moon.
Barber's overall assessment was direct: "Orion itself seems to have worked pretty well, actually - certainly all the propulsion stuff, which is the real critical stuff."
Great science or Nasa hype?
NASA has talked up the scientific returns. The crew made extensive observations during their flyby - around 35 geological features noted in real time, colour variations that could reveal mineral composition, and a solar eclipse from deep space that pilot Victor Glover said "just looks unreal."
One image stood out: the Orientale basin, a 600-mile crater near the Moon's far side, seen in full by human eyes for the first time.
And yet the science is not the main point. Professor Chris Lintott of Oxford, co-host of The Sky at Night, was blunt: "The artistic value of the images returned from Artemis and its crew is significant, but their scientific value is limited."
India's Chandrayaan-3 landed near the south pole in 2023. China's Chang'e-6 retrieved samples from the far side in 2024. Robotic probes have mapped this terrain in extraordinary detail.
The most affecting moment came not from any instrument, but from the crew. As the astronauts broke the distance record set by the stricken Apollo 13 crew in 1970, Mission Specialist Jere
▸ 展开全文
04-08 00:01 · 娱乐产业,社会舆论,活动取消,国际新闻,品牌风险
West was due to headline the festival in July but drew criticism over past antisemitic comments.
Wireless Festival cancelled after Kanye West blocked from coming to UK
This summer's Wireless Festival has been cancelled after headliner Kanye West was blocked from coming to the UK.
The government refused permission for West, now known as Ye, to travel to the UK after backlash to his planned set at the London festival this summer.
For the last several years, West has caused outrage for a string of antisemitic, racist and pro-Nazi comments.
In a statement, Wireless Festival said it was cancelled and refunds would be issued to all ticket holders. It added that "multiple stakeholders" had been consulted ahead of booking West, "and no concerns were highlighted at the time".
It continued: "Antisemitism in all its forms is abhorrent, and we recognise the real and personal impact these issues have had.
"As Ye said today, he acknowledges that words alone are not enough, and in spite of this still hopes to be given the opportunity to begin a conversation with the Jewish community in the UK."
Presale tickets for Wireless Festival were released at midday on Tuesday and are believed to have sold out, while the general sale was due to open at midday on Wednesday.
Earlier on Tuesday, the Home Office told the BBC the rapper had made an application on Monday to travel to the UK via an Electronic Travel Authorisation (ETA).
It said the decision to refuse permission was made on the grounds that his presence would not be conducive to the public good.
Visitors to the UK need an ETA if they do not need a visa for short stays of up to six months, or do not already have a UK immigration status.
West has faced criticism for his previous comments. In 2022, he posted on social media saying he would go "death con 3 On Jewish people".
Later the same year, he appeared on a podcast hosted by conspiracy theorist Alex Jones and said: "I see good things about Hitler."
In May 2025, West released a song called Heil Hitler and sold T-shirts featuring swastikas.
The rapper has been seeking a return to mainstream public view after apologising for his actions in a lengthy statement published in the Wall Street Journal in January.
Just last week, he played two sold-out concerts in Los Angeles, telling the crowd at the SoFi Stadium: "Tonight we're going to put all this behind us, ain't that right LA?"
West had said on Tuesday that he "would be grateful" to meet members of the Jewish community in the UK in person "to listen" after controversy over his booking.
"I know words aren't enough," added the star.
"I'll have to show change through my actions. If you're open, I'm here."
Melvin Benn, the managing director of Festival Republic, the company behind Wireless, previously said attempts had been made to reach out to Jewish groups after the concerts were announced, but "they have refused a meeting". He said that the approach had been made over the past couple of days.
In response, a Board of Deputies spokesperson told the BBC: "Neither the Board of Deputies nor, we understand, the Jewish Leadership Council has refused any request to meet with the Wireless festival organisers.
"When the Board of Deputies received a letter from Melvin Benn on 6 April, proposing to meet, in response to a letter we sent setting out our concerns, we responded positively."
The board said that, regardless of any meeting, they had been clear that the invitation for West to perform "should be rescinded".
That decision has now been taken out of West's hands, and those of Festival Republic, which runs other major UK festivals including Reading and Leeds.
"It may be that the home secretary does rescind his visa... if she does, the issue is over," he told the Today programme.
Benn, who is a major UK music industry figure and event promoter, agreed that West's past behaviour was "abhorrent" and "disgusting".
But the promoter highlighted the role that the star's mental health may have played, and asked for understanding.
"Mental health is not something that disappears overnight," Benn said.
"People suffer psychotic behaviour, suffer bipolar behaviour, for many, many years... And I think people are forgetting that."
However, Jewish groups criticised Benn's support for the star.
The Community Security Trust (CST), which aims to protect British Jews from antisemitism, described the government's decision to stop West travelling to the UK as "a sensible outcome to what has been yet another bruising episode for British Jews".
Its statement added: "Anti-Jewish hatred should have no place in society and cultural leaders have a role to play in ensuring that is the case.
"People who show genuine and meaningful remorse for previous antisemitic behaviour will always receive a sympathetic hearing from the Jewish community, but that process must come before this kind of public rehabilitation."
Gideon Falter, chief executive of Campaign Against Antisemitism, has called for an apology from the organisers of Wireless.
He told the BBC that the booking of West represented a pursuit of "profit over
▸ 展开全文
04-08 00:01 · 国际,新闻,BBC,国际
Attacks have targeted bridges, steel plants and pharmaceutical facilities, verified videos show.
Tracking recent US-Israeli strikes on Iranian infrastructure
President Donald Trump has pledged to target infrastructure across Iran unless the country reaches an "acceptable" deal to end the war with the US and Israel by Tuesday night.
Promising to bomb the country "back to the Stone Ages", Trump said US forces would target bridges and power plants across Iran. He also posted to social media on Tuesday that a "whole civilization will die tonight" if an agreement is not struck.
There has been a spate of attacks on infrastructure critical to ordinary Iranians since the conflict began, with schools and hospitals damaged.
BBC Verify has confirmed that US and Israeli strikes have targeted at least two steel plants, three bridges and a pharmaceutical plant over the past two weeks.
Some senior Democrats in the US Congress and UN officials have warned that strikes like these could amount to war crimes. But in a news conference on Monday Trump dismissed those concerns.
What civilian infrastructure has been hit by strikes?
Footage confirmed by BBC Verify showed at least two strikes on the bridge. A large gap in the bridge can be seen in the aftermath of the attack, with construction cranes visible either side of it.
A number of steel facilities have also been targeted. On 27 March, verified footage showed smoke billowing from the Isfahan Mobarakeh Steel company, forcing workers to suspend operations. The plant - which is Iran's largest steel manufacturer - exported around $860m (£649m) between March 2025 and January 2026, according to company linked-reporting.
Satellite images also showed damage at the Khuzestan Steel Company - the country's second largest manufacturer. Local officials said that it could take up to a year to repair the damage caused to the plant by the attacks.
Israeli Prime Minister Benjamin Netanyahu has claimed that the strikes have disrupted up to 70% of Iran's steel manufacturing capability. Arman Mahmoudian, research fellow at the University of South Florida's Global and National Security Institute, said the damage could seriously impact the Iranian economy.
"Steel is a cornerstone of Iran's non-oil economic capacity," he said. "If Israeli strikes have indeed dismantled around 70% of Iran's steel production capacity, this would place nearly 20 million tons of output at risk, potentially affecting around 3–3.5% of Iran's GDP."
Mahmoudian also noted that attacks on Iran's pharmaceutical industry could pose major issues for the healthcare system. On 31 March, an Israel Defense Forces (IDF) spokesperson said it had carried out a strike on the Tofigh Daru Research & Engineering Company - one of Iran's largest pharmaceutical companies which produces anaesthetic and cancer drugs.
The IDF said in a statement that it had carried out the strike and alleged that the company had transferred "chemical substances, including fentanyl, that were used for research and development of chemical weapons". The BBC cannot independently verify that claim.
While pharmaceutical goods represent only a small part of Iran's economy, Mahmoudian noted that strikes against the sector could impact access to medicine at a time of need and undermine Tehran's "medical independence". Iranian media has previously claimed that over 90% of pharmaceuticals are produced domestically, but BBC Verify cannot confirm that.
Educational facilities have been heavily damaged in recent strikes, verified footage and photos suggest. On Saturday, images showed debris around the exterior of Shahid Beheshti University in Tehran and parts of the building were destroyed after a reported attack.
Images also showed damage at the Sharif University of Technology in Tehran on Monday after strikes on the capital.
Religious buildings have also been hit. In the central city of Zanjan an attack levelled parts of the Husseinya Mosque, photos showed. Local officials said two people were killed in the strike, which destroyed a clinic and a library within the complex.
The IDF also said on Tuesday that it had bombed 10 "key" parts of the Iranian railway. Verified footage from Aminabad village in central Iran showed a collapsed railway bridge said to be caused by the attacks.
Another video posted by the Iranian Red Crescent and which we verified showed paramedics carrying an injured man away from a railway line near Karaj. It's unclear what caused his injuries.
One railway worker in Tehran who spoke to the BBC expressed frustration at the attacks on the lines. "I'm really angry," the worker said. "Everything is falling apart."
The strikes came after IDF warned Iranians early on Tuesday against travelling. Writing in Farsi in a post to social media, it warned civilians that their "presence on trains and near railway lines endangers your life".
Could the US-Israeli strikes amount to war crimes?
The recent strikes are just the latest in a wave of attacks which have damaged civilian infrastructure.
Legal experts and some senior US and UN officials have questioned wheth
▸ 展开全文
04-08 00:00 · AI应用案例,文化遗产数字化,技术社区,轻量自动化,数据增强
AI helps add 10k more photos to OldNYC
Over the past two years I’ve quietly rebuilt major parts of the OldNYC photo viewer. The result: 10,000 additional historic photos on the map, more accurate locations, and a site that’s cheaper and easier to run—thanks to modern AI tools and the OpenStreetMap ecosystem.
OldNYC had about 39,000 photos in 2016. Today it has 49,000.
Most of these changes happened in 2024, but I’m only writing about them now in 2026. (I got distracted by an unrelated project.) If you haven’t visited OldNYC in a while, take a look—you might find some photos you missed.
Here are the three biggest improvements: better geolocation, dramatically improved OCR, and a switch to an open mapping stack.
OldNYC works by geocoding historical descriptions—turning text like “Broad Street, south from Wall Street” into a latitude and longitude.
Originally this mostly meant extracting cross streets from titles and sending them to the Google Maps Geocoding API. That worked well when the streets still existed—but many historical intersections don’t.
Two changes in 2024 improved this dramatically.
Some images include useful location details only in the description. I now use the OpenAI API (gpt-4o
) to extract locations from that text.
Public Schools - Brooklyn - P.S. 143. 1930
Havemeyer Street, west side, from North 6th to North 7th Streets, showing Public School No. 143. The view is north from North 6th Street.
The school no longer exists, so the title alone can’t be geocoded. From the description, GPT extracted:
Both intersections exist in OpenStreetMap, so OldNYC places the image at the first one.
Tasks like this require a surprising amount of interpretation: GPT understands that “North 6th” means “North 6th Street” and extracts the relevant intersections while ignoring irrelevant phrases like “west side”. Computers have historically had trouble with this type of task, but the newer AI models nail it.
Using GPT located about 6,000 additional photos. Today OldNYC can locate roughly 87% of photos with usable location data, and about 96% of mapped images appear in the correct location.
I also replaced the Google Maps geocoder with OpenStreetMap and historical street datasets.
Brooklyn: Fulton Street – Nassau Street
These streets intersected in Brooklyn in the 1930s but no longer do today. Google geocodes this to Manhattan, where streets with those names still intersect.
OldNYC now incorporates data from the NYPL’s [historical streets project], which includes the original Brooklyn intersection. The photo now appears in the correct location.
Most OldNYC photos include descriptions from the NYPL catalog—but on the NYPL site these are scanned typewriter images, not text.
When I launched OldNYC in 2015, converting these images to text (OCR) was the hardest technical problem. I built a custom pipeline using Ocropus that achieved over 99% character accuracy. Even so, the errors were noticeable when reading.
To fix mistakes I added a “Fix Typos” feature that let users correct transcriptions. This triggered New Yorkers’ collective OCD and users submitted thousands of edits.
In 2024 I rebuilt the OCR system using gpt-4o-mini
.
The results were much better:
Here’s a dramatic example where the old OCR produced complete gibberish due to an unusual font:
GPT transcribes it perfectly.
A few lessons from rebuilding the pipeline:
Overall, tools like OpenAI mean that OCR is a much easier problem in 2024 than it was in 2015.
When OldNYC launched, Google Maps was the default choice for web mapping, and it was free to use. But over time, Google’s pricing model changed. In late 2024 they replaced their $200/month free credit with separate quotas for individual APIs. Under the new system, rather than being free, OldNYC would have cost about $35/month.
Instead of paying Google indefinitely for a hobby project, I migrated the site to OpenStreetMap vector tiles and MapLibre.
The new stack has some nice benefits:
For example, I can remove anachronisms like highways and tunnels that didn’t exist in the 1930s.
Look, no Brooklyn-Battery Tunnel!
There’s still plenty to improve.
AI could extract additional information from images—identifying people, buildings, or indoor/outdoor scenes. I’d also like to incorporate photographs from other collections.
I’ve also started contributing to OpenHistoricalMap, the history-focused cousin of OpenStreetMap. If it eventually includes full historical street grids for NYC, locating photos could become dramatically easier.
Finally, I’d love to make it easier for developers to build OldNYC-style sites for other cities. If you’re interested, please reach out.
📪 If you’d like to be informed of OldNYC updates, please subscribe to the new mailing list! If you subscribed before 2026, you’ll need to subscribe again. Sorry, MailChimp deleted the old list. 😡
Please reach out if you have any feedback.
▸ 展开全文
04-08 00:00 · AI安全,技术标准,软件供应链,合规风险,开发者生态
Project Glasswing: Securing critical software for the AI era
Project Glasswing
Securing critical software for the AI era
Introduction
Today we’re announcing Project Glasswing1, a new initiative that brings together Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks in an effort to secure the world’s most critical software.
We formed Project Glasswing because of capabilities we’ve observed in a new frontier model trained by Anthropic that we believe could reshape cybersecurity. Claude Mythos2 Preview is a general-purpose, unreleased frontier model that reveals a stark fact: AI models have reached a level of coding capability where they can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.
Mythos Preview has already found thousands of high-severity vulnerabilities, including some in every major operating system and web browser. Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely. The fallout—for economies, public safety, and national security—could be severe. Project Glasswing is an urgent attempt to put these capabilities to work for defensive purposes.
As part of Project Glasswing, the launch partners listed above will use Mythos Preview as part of their defensive security work; Anthropic will share what we learn so the whole industry can benefit. We have also extended access to a group of over 40 additional organizations that build or maintain critical software infrastructure so they can use the model to scan and secure both first-party and open-source systems. Anthropic is committing up to $100M in usage credits for Mythos Preview across these efforts, as well as $4M in direct donations to open-source security organizations.
Project Glasswing is a starting point. No one organization can solve these cybersecurity problems alone: frontier AI developers, other software companies, security researchers, open-source maintainers, and governments across the world all have essential roles to play. The work of defending the world’s cyber infrastructure might take years; frontier AI capabilities are likely to advance substantially over just the next few months. For cyber defenders to come out ahead, we need to act now.
Cybersecurity in the age of AI
The software that all of us rely on every day—responsible for running banking systems, storing medical records, linking up logistics networks, keeping power grids functioning, and much more—has always contained bugs. Many are minor, but some are serious security flaws that, if discovered, could allow cyberattackers to hijack systems, disrupt operations, or steal data.
We have already seen the serious consequences of cyberattacks for important corporate networks, healthcare systems, energy infrastructure, transport hubs, and the information security of government agencies across the world. On the global stage, state-sponsored attacks from actors like China, Iran, North Korea, and Russia have threatened to compromise the infrastructure that underpins both civilian life and military readiness. Even smaller-scale attacks, such as those where individual hospitals or schools are targeted, can still inflict substantial economic damage, expose sensitive data, and even put lives at risk. The current global financial costs of cybercrime are challenging to estimate, but might be around $500B every year.
Many flaws in software go unnoticed for years because finding and exploiting them has required expertise held by only a few skilled security experts. With the latest frontier AI models, the cost, effort, and level of expertise required to find and exploit software vulnerabilities have all dropped dramatically. Over the past year, AI models have become increasingly effective at reading and reasoning about code—in particular, they show a striking ability to spot vulnerabilities and work out ways to exploit them. Claude Mythos Preview demonstrates a leap in these cyber skills—the vulnerabilities it has spotted have in some cases survived decades of human review and millions of automated security tests, and the exploits it develops are increasingly sophisticated.
Ten years after the first DARPA Cyber Grand Challenge, frontier AI models are now becoming competitive with the best humans at finding and exploiting vulnerabilities. Without the necessary safeguards, these powerful cyber capabilities could be used to exploit the many existing flaws in the world’s most important software. This could make cyberattacks of all kinds much more frequent and destructive, and empower adversaries of the United States and its allies. Addressing these issues is therefore an important security priority for democratic states.
Although the risks from AI-augmented cyberattacks are serious, there is reason for optimism: the same capabilities that make AI models dangerous in the wrong hands make them invaluable for findin
▸ 展开全文