1. Claude Code Unpacked : A visual guide (ccunpacked.dev)

What actually happens when you type a message into Claude Code? The agent loop, 40+ tools, multi-agent orchestration, and unreleased features, mapped from source.

6. CERN levels up with new superconducting karts (home.cern)

Following on from the robotic mice, CERN engineers have now developed a super-charged kart to enable workers to race through the Large Hadron Collider (LHC) underground tunnel during the upcoming major works, starting this summer. The karts promise a power boost to activities during this period, known as Long Shutdown 3 (LS3), which will see the LHC transformed into the High-Luminosity LHC. These vehicles will replace the bicycles that were used until now to travel through the 27-km underground tunnel, enabling engineers and technicians to speed to areas where improvements to the accelerator are required. β€œEach kart is turbo-boosted by 64 superconducting engines,” explains project leader Mario Idraulico. β€œWhen the engines are cooled to below their critical temperatures, the Meissner effect levitates the karts, allowing them to zip through the tunnels at high speeds and, mamma mia, they’re super!” Early tests have been promising, and the next steps involve testing different kart designs in an underground race. Safety coordinator Luigi Fratello has ensured that each driver will be issued with Safety and Health Equipment for Long and Limited Stays (SHELLS), although his response to drivers wanting bananas in the tunnel was β€œOh no!” These karts, although developed to support CERN’s fundamental research programme, show clear applications for society. CERN’s Knowledge Transfer Group has begun discussions with European startup company Quantum Mushroom to explore aerospace applications and powering for next-generation anti-gravity vehicles. Surprisingly, the kart project began from a collaboration between CERN engineers and onsite nursery school children – one example of CERN’s commitment to inspiring future generations. β€œWe’re thrilled that the children’s kart designs were the inspiration for the engineered karts,” exclaimed schoolteacher Yoshi Kyouryuu, mid-way through painting spots on eggs for an Easter egg hunt. β€œAs educators, we promote curiosity from a young age, which is why we paint question marks all over our yellow school walls,” explained school director, Rosalina Pfirsich, looking up from her storybook. β€œWith all the contributions the children have made to the upcoming High-Luminosity LHC project, we’ve taken to calling them Luma!” Find out more about the High-Luminosity LHC project.

13. Is BGP Safe Yet? No. Test Your ISP (isbgpsafeyet.com)

On the Internet, network devices exchange routes via a protocol called BGP (Border Gateway Protocol). Unfortunately, issues with BGP have led to malicious actors being able to hijack and misconfigure devices leading to security problems which have the potential to cause widespread problems. BGP security can be greatly improved by using technologies such as RPKI to sign Internet routes. This page attempts to track the progress of major Internet players (ISPs, transit operators, and content providers) in their progress to adopt RPKI and other technologies.

17. TinyLoRA – Learning to Reason in 13 Parameters (arxiv.org)

Recent research has shown that language models can learn to \textit{reason}, often via reinforcement learning. Some work even trains low-rank parameterizations for reasoning, but conventional LoRA cannot scale below the model dimension. We question whether even rank=1 LoRA is necessary for learning to reason and propose TinyLoRA, a method for scaling low-rank adapters to sizes as small as one parameter. Within our new parameterization, we are able to train the 8B parameter size of Qwen2.5 to 91\% accuracy on GSM8K with only 13 trained parameters in bf16 (26 total bytes). We find this trend holds in general: we are able to recover 90\% of performance improvements while training $1000x$ fewer parameters across a suite of more difficult learning-to-reason benchmarks such as AIME, AMC, and MATH500. Notably, we are only able to achieve such strong performance with RL: models trained using SFT require $100-1000x$ larger updates to reach the same performance.

19. AI for American-Produced Cement and Concrete (engineering.fb.com)

Meta is continuing its long-term roadmap to help the construction industry leverage AI to produce high-quality and more sustainable concrete mixes, as well as those exclusively produced in the Unit…

21. Chess in SQL (www.dbpro.app)

What if I told you SQL could play chess? No JavaScript, no frameworks - just pure SQL rendering a full chess board in your browser.

25. Intuiting Pratt Parsing (louis.co.nz)

You already know that a + b * c + d is calculated as a + (b * c) + d. But how do you encode that knowledge precisely enough for a machine to act on it?

Article Details

Browse by Date

Type to search HN and engineering blog articles
Press Escape to close  |  ↑↓ to navigate  |  Enter to open