Hasty Briefsbeta

Bilingual

Japan implements language proficiency requirements for certain visa applicants
2 hours ago
- Japan is introducing a language proficiency requirement for specific foreign nationals applying for the common 'Engineer/Specialist in Humanities/International Services' visa, effective immediately.
- Applicants whose work involves language skills must now prove their ability, with the benchmark set at CEFR B2 level, demonstrable through JLPT N2 certification or a BJT score of 400 or higher.
- The change applies to roles such as interpreters, company workers, and hotel staff, as outlined by the Justice Ministry under new guidelines.
Where the DOGE Operatives Are Now
2 hours ago
- Young technologists in Elon Musk's Department of Government Efficiency (DOGE) caused significant disruption across US government agencies, leading to over 300,000 federal workers fired and agency closures like USAID.
- DOGE members have moved into influential roles in government and private sector, with examples including Edward Coristine at the National Design Studio and Gavin Kliger as Department of Defense chief data officer.
- DOGE's actions, justified by claims of targeting 'waste, fraud, and abuse,' have had lasting impacts, such as increased wait times at the Social Security Administration and cuts to humanitarian aid.
- Several DOGE alumni founded or joined private companies, often with ties to Musk or Silicon Valley, such as Nate Cavanaugh and Justin Fox's AI startup Special, and Adam Ramada and Brooks Morgan's venture firm Banner VC.
- The article highlights ongoing legal and ethical concerns, including whistleblower complaints about data mishandling and litigation over agency takeovers, with some members' current whereabouts unknown.
A look at the J language: the fine line between genius and insanity (2012)
a day ago
- The author was searching for a time-series database (TSDB) and explored APL family languages, eventually focusing on J and JDB as a free alternative to KDB+.
- J is characterized by its array-oriented programming, using a unique syntax with verbs, nouns, and adverbs, which requires high attention but can lead to concise and powerful code.
- The J community is described as friendly, helpful, and knowledgeable, offering strong support through interactive development environments (jgtk and JHS) and comprehensive tutorials.
- Key features of J include built-in sparse arrays, efficient memory management, and the ability to handle large datasets effectively, with packages for plotting and other utilities.
- While the author sees potential in JDB for tick data storage, they note that J may not replace R for specialized tasks like Hidden Markov models but is valuable for numerics and different programming paradigms.
Show HN: Flint – A 30B model fine-tuned for less repetition
18 hours ago
- Flint is a language model designed specifically for inspiration and creative tasks, not for giving correct answers.
- It addresses the issue of output convergence in frontier LLMs, which leads to repetitive and similar responses, hindering divergent thinking.
- Flint achieves a dramatic increase in output diversity on creative tasks without degrading performance in other areas like accuracy or responsible AI benchmarks.
- On NoveltyBench, Flint scores 7.47 mean distinct responses out of 10, significantly outperforming SOTA models such as Gemini 3.1 Pro (3.19), GPT-5.4 (2.54), and Claude 4.6 Sonnet (1.83).
- In intra-model similarity tests, Flint has a mean similarity of 0.721 (lower is better), compared to higher scores for other models (e.g., GPT-5.4: 0.864, Gemini 3.1 Pro: 0.871, Claude 4.6 Sonnet: 0.905).
- Flint also shows lower inter-model similarity (0.672) than other models, making it the most distinctive in the comparison.
- It maintains performance on benchmarks like MMLU-STEM (78.9%), TruthfulQA MC1 (34.4%), and ToxiGen (59.6%), showing divergence tuning does not compromise capability.
- Flint is not a replacement for frontier models but acts as a multiplier in creative workflows, generating range while larger models provide depth and reasoning, with humans applying judgment.
- The model is available in alpha version via the Springboards app.
The API Tooling Crisis: Why developers are abandoning Postman and its clones?
6 hours ago
- The article criticizes the shift of API tools like Postman from developer-friendly to corporate-driven products, focusing on enshittification and performance issues.
- Key grievances include Postman's removal of offline mode, forced cloud synchronization, and bloated performance, alongside similar patterns in Insomnia and Thunder Client.
- Developers are turning to alternatives like Bruno for local-first, Git-integrated workflows, though these tools lack polish and some enterprise features.
- An ideal API tool is described as local-first, Git-native, high-performance, extensible, and free from mandatory logins or cloud locks.
- An update mentions Postman's introduction of Native Git features, but criticizes it as a surface-level change that still ties collaboration to a paid cloud model.
Turning a Chinese IoT camera into an owl livestream
a day ago
- The article describes a project to create a livestream of a Great Horned Owl nest using a Chinese IoT camera.
- Spencer, an artist and conservationist, set up an S4 Pro camera from Premium Invention (a rebranded Nice Intelligence Y5 Camera) in a forest.
- The NiView app used with the camera required app downloads and link sharing for viewing, which didn't scale well and drained battery.
- Initial attempts to extract a web viewer or stream file (e.g., .m3u8) via the manufacturer's websites were unsuccessful.
- Using mitmproxy to intercept network traffic revealed encrypted API requests to Tencent Cloud but no direct video stream URLs.
- The NiView app was decrypted from a jailbroken iPhone using bfdecrypt, allowing inspection of its code and resources.
- Debugging with lldb and Hopper identified key classes like NVLivePlayerView and IVVideoRender for video rendering.
- The video feed used a P2P relay via Tencent's servers (list.iotvideo.tencentcs.com), making direct reverse-engineering complex.
- A jailbreak tweak was developed using Theos to hook into IVVideoRender's _renderVideoFrame method, capture YUV frames, convert them to JPEGs, and stream them via a local HTTP server on port 8080.
- The tweak composited images (cropping and downsampling) to show the zoomed view with overlays like timestamps, and sent MJPEG streams to clients.
- A Python script and PHP wrapper were created to cache and rebroadcast the stream from an EC2 server, reducing load on the iPhone.
- Pitfalls included memory leaks in the tweak (fixed with static buffers and autorelease pools), image compositing challenges, and camera battery drain (solved temporarily with an external battery).
- The project highlights the technical hurdles of reverse-engineering proprietary IoT systems and offers a workaround for similar livestreaming needs, suggesting cameras with RTSP support for future projects.
The American Experiment Has Been Infected by Oligarchs
3 hours ago
- The article warns of an 'oligarchy virus' in the U.S. tax system, where extreme wealth concentration distorts democracy and culture.
- Wealth inequality has surged, with the top 0.00001% (19 billionaires) holding over $3 trillion, more than 12% of national income, and billionaire political spending skyrocketing.
- The tax system's flaw lies in not taxing unrealized gains (Haig-Simons income) and taxing investment gains at lower rates than wages, favoring the wealthy.
- Historical progressive taxation and strong labor policies once prevented oligarchy, but since the 1980s, wage stagnation, deregulation, and tax cuts have enabled wealth reconcentration.
- Proposed solutions include raising the minimum wage, enforcing antitrust laws, taxing investment gains as ordinary income, and implementing a wealth tax via the Oligarch Act of 2025.
- The oligarchy cycle (wealth begetting power) makes reform challenging, but confronting it is urgent to restore democratic balance.
KLM cancels 160 flights due to fuel shortage
3 hours ago
- The IMF warns the global economy faces a large shock due to a conflict blocking 20% of oil and gas in the Strait of Hormuz, impacting everyone, especially low-income countries.
- IMF chief advises against 'generous' subsidies for energy to avoid increasing demand and worsening the crisis, urging policies that shrink demand instead.
- France's central bank chief notes inflation is lower than in 2022 but stresses central banks will act to prevent second-round inflation effects, though not hastily.
- Saudi Arabia's finance minister warns that even with a ceasefire, recovery in energy markets will take weeks or months due to operational and insurance delays.
- KLM cancels 160 European flights due to rising fuel costs, while oil prices rise amid US threats to Iran's energy infrastructure.
- A Nigerian mega-refinery boosts jet fuel supplies to Europe, helping offset shortages caused by the conflict and benefiting from local crude purchases.
- Economic impacts include reduced growth forecasts for Greece due to high energy prices and a drop in US industrial output, signaling broader economic strain.
What's new in Claude Opus 4.7
5 hours ago
- Claude Opus 4.7 is the most capable generally available model with high autonomy and excels at long-horizon agentic work, knowledge work, vision tasks, and memory.
- It introduces high-resolution image support up to 2576px/3.75MP, enhancing vision-heavy workloads like document understanding, with pixel-accurate coordinates and no scale-factor math required.
- The model features a new effort parameter to tune intelligence vs. token spend, with 'xhigh' recommended for coding and agentic use cases, and 'high' for intelligence-sensitive tasks.
- Task budgets allow setting an advisory token limit for full agentic loops, helping the model prioritize work, with a minimum of 20k tokens, distinct from the hard max_tokens cap.
- Breaking changes include removal of extended thinking budgets (adaptive thinking only), non-default temperature/top_p/top_k parameters causing errors, and thinking content omitted by default unless opted in.
- A new tokenizer may use 1x to 1.35x more tokens, affecting token counts, and prompts may need updates for max_tokens and cost control, with improvements in knowledge-worker tasks and memory use.
Artifacts: Versioned storage that speaks Git
6 hours ago
- Agents are driving a massive increase in code generation, overwhelming traditional source control systems built for humans.
- Artifacts is a new distributed, versioned filesystem designed primarily for agents, compatible with Git and accessible via APIs.
- It enables creating repositories on-demand for agent sessions, sandboxes, and forks, with support for REST and Workers APIs.
- Artifacts leverages Git's data model for state tracking, time travel, and persistence, beyond just source control.
- Built on Cloudflare's Durable Objects, R2, and KV, it uses a custom Zig-based Git implementation compiled to WebAssembly for efficiency.
- ArtifactFS, an open-source filesystem driver, accelerates cloning large repos by hydrating files asynchronously, reducing startup time.
- Planned features include expanded metrics, event subscriptions, SDKs for multiple languages, and search APIs.
- Pricing is based on storage and operations, with free tiers and beta access starting privately before a public launch in early May 2026.
Laravel raised money and now injects ads directly into your agent
4 hours ago
- Laravel secured a $57M investment from Accel and now integrates promotion for Laravel Cloud into its official agent library.
- Developers criticize a PR change that exclusively suggests Laravel Cloud for deployment, removing mentions of alternatives, as a form of 'enshittification'.
- User feedback indicates this change 'poisons' agents by defaulting to Laravel Cloud, even in irrelevant scenarios, but the CEO prioritizes commercial benefits.
- Despite Laravel Cloud already receiving praise from agents like ChatGPT and Claude, the forced promotion risks community trust and raises ethical concerns.
- The situation sparks a broader discussion on advertising to AI agents and the potential need for 'agent-ad-block' tools as these practices become more discreet.
Launch HN: Kampala (YC W26) – Reverse-Engineer Apps into APIs
4 hours ago
- Kampala enables reverse engineering of websites, mobile apps, and desktop apps.
- It provides full traffic interception to monitor HTTP/S requests in real time.
- Features include auth chain tracing, flow replay with automation export, and fingerprint preservation.
- Windows support is planned with a waitlist available for notification.
OpenSSL 4.0.0 Released
21 hours ago
- OpenSSL 4.0.0 released with new features and incompatible changes.
- Support added for Encrypted Client Hello (ECH) and SM2 signature algorithms.
- Removed support for deprecated protocols like SSLv2, SSLv3, and engines.
- Key changes include standardized hexadecimal dumps and API function signature updates.
- FIPS self-tests can now be deferred with new installation options.
- OpenSSL 4.0 will be supported until May 14, 2027.
The becquerel as an SI unit for request rate
11 hours ago
- Request rate refers to the number of requests arriving, serviced, or leaving within a specified period.
- It's crucial to specify the period length when measuring request rate, often overlooked in dashboards.
- Requests per second is recommended as the standard period length for consistency across different dashboard views.
- Two SI units relevant to request rate: Hertz (Hz) for periodic events and Becquerel (Bq) for average frequency events.
- Hz implies regular, predictable timing, while Bq accounts for irregular, average-based occurrences.
- Hz suits regular load testing; Bq better represents organic traffic with average request rates.
- Using 'kBq' simplifies expressing high request rates compared to verbose 'requests per second'.
- Becquerel is officially for radioactive decay, but applying it to request rates is suggested for future adoption.
Rust 1.95.0
6 hours ago
- Rust 1.95.0 has been released, introducing new features and improvements.
- The new cfg_select! macro provides compile-time configuration selection similar to the cfg-if crate.
- If-let guards are now allowed in match expressions, enabling conditionals based on pattern matching.
- Multiple APIs have been stabilized, including MaybeUninit, Cell, Atomic types, and core::range.
- Support for JSON target specs on stable has been removed, requiring nightly for custom targets.
- Future releases can be tested by switching to beta or nightly channels using rustup.
Alibaba open-sources Qwen3.6-35B-A3B, a 35B MoE model with 3B active parameters
5 hours ago
- Qwen3.6-35B-A3B is a post-trained model released by the Qwen team, offering improved stability and utility based on community feedback.
- It features 35B total parameters (3B activated), a context length of up to 262,144 tokens (extendable to 1,010,000 with YaRN), and supports multimodal inputs including images and videos.
- Key enhancements include better agentic coding for frontend workflows and repository-level reasoning, along with thinking preservation to retain reasoning context from historical messages.
- The model is compatible with various inference frameworks such as SGLang, vLLM, KTransformers, and Hugging Face Transformers for deployment and serving.
- It achieves strong benchmark results in coding (e.g., SWE-bench, Terminal-Bench), knowledge (e.g., MMLU-Pro, C-Eval), and vision-language tasks (e.g., MMMU, MathVista).
- Usage involves API integration with recommended sampling parameters for different modes (thinking vs. non-thinking) and tasks, plus support for tool calling and agent applications via Qwen-Agent.
The Internet's Most Powerful Archiving Tool Is in Peril
6 hours ago
- USA Today's investigative report utilized the Wayback Machine to uncover ICE's delayed disclosure on detention policies, highlighting the tool's critical role in public interest journalism.
- Several major news organizations, including USA Today Co., The New York Times, and Reddit, are blocking the Internet Archive's crawler, citing concerns over AI misuse and copyright violations.
- Journalists and advocacy groups are rallying to support the Wayback Machine, emphasizing its importance for historical research, fact-checking, and preserving digital records amid declining traditional archives.
- Publishers blocking the tool argue it risks enabling AI companies to train models on their content without permission, a concern central to ongoing legal battles over copyright and AI training data.
- The trend threatens the Wayback Machine's mission of preserving web history, potentially eroding access to early digital records and impacting accountability journalism and legal evidence.
Claude Opus 4.7
5 hours ago
- Claude Opus 4.7 is now generally available, offering significant improvements in advanced software engineering and handling complex, long-running tasks with consistency and self-verification.
- It features substantially better vision with higher resolution image support (up to ~3.75 megapixels) and excels in professional tasks like creating interfaces, slides, and documents.
- Opus 4.7 demonstrates notable performance gains across benchmarks, including a 13% lift on a 93-task coding benchmark and improved efficiency in multi-step workflows.
- Despite being less capable than Claude Mythos Preview, it is the first model released with cybersecurity safeguards, and security professionals can join the Cyber Verification Program for legitimate use.
- It shows strong feedback from testers, with improvements in autonomy, reasoning, tool-call accuracy, and error recovery, making it effective for agentic and enterprise workflows.
- Key updates include a new 'xhigh' effort level, task budgets in the API, and features like '/ultrareview' in Claude Code, along with changes in tokenizer affecting token usage.
There's yet another study about how bad AI is for our brains
5 hours ago
- AI assistance initially improves performance but at a cognitive cost, reducing persistence and independent skills.
- Study on 350 Americans solving equations showed AI group's performance dropped sharply when access was cut, with similar results in larger experiments.
- Researchers warn AI in education could dilute human innovation and creativity by eroding motivation and long-term learning.
- Caveats: study not peer-reviewed; using AI for hints vs. answers had varying impacts on post-AI performance.
- Other studies link AI to 'AI brain fry' in workers, increased fatigue, and poorer test outcomes in students.
Fly Drones from the Browser
4 days ago
- FPVSIM is a browser-based drone flight simulator accessible online.
- Includes lap timer and V3 features, with manuals and how-tos for setup.
- Supports iOS/Android, OTA flashing, multi-device setup, and internet connectivity.
- Offers performance tips, P/D tuning, sync/diff with Betaflight/INAV, PID/rates from top pilots.
- Provides auto quad configurations, upgrade via authorization code, and beginner guides.
- Features tools like blackbox analyzer and radio latency checker.
- Requires login/signup with Google or email for access, with performance tips on initial load.

first prev0next

About|Login

Bilingual

Japan implements language proficiency requirements for certain visa applicants
2 hours ago
- Japan is introducing a language proficiency requirement for specific foreign nationals applying for the common 'Engineer/Specialist in Humanities/International Services' visa, effective immediately.
- Applicants whose work involves language skills must now prove their ability, with the benchmark set at CEFR B2 level, demonstrable through JLPT N2 certification or a BJT score of 400 or higher.
- The change applies to roles such as interpreters, company workers, and hotel staff, as outlined by the Justice Ministry under new guidelines.
Where the DOGE Operatives Are Now
2 hours ago
- Young technologists in Elon Musk's Department of Government Efficiency (DOGE) caused significant disruption across US government agencies, leading to over 300,000 federal workers fired and agency closures like USAID.
- DOGE members have moved into influential roles in government and private sector, with examples including Edward Coristine at the National Design Studio and Gavin Kliger as Department of Defense chief data officer.
- DOGE's actions, justified by claims of targeting 'waste, fraud, and abuse,' have had lasting impacts, such as increased wait times at the Social Security Administration and cuts to humanitarian aid.
- Several DOGE alumni founded or joined private companies, often with ties to Musk or Silicon Valley, such as Nate Cavanaugh and Justin Fox's AI startup Special, and Adam Ramada and Brooks Morgan's venture firm Banner VC.
- The article highlights ongoing legal and ethical concerns, including whistleblower complaints about data mishandling and litigation over agency takeovers, with some members' current whereabouts unknown.
A look at the J language: the fine line between genius and insanity (2012)
a day ago
- The author was searching for a time-series database (TSDB) and explored APL family languages, eventually focusing on J and JDB as a free alternative to KDB+.
- J is characterized by its array-oriented programming, using a unique syntax with verbs, nouns, and adverbs, which requires high attention but can lead to concise and powerful code.
- The J community is described as friendly, helpful, and knowledgeable, offering strong support through interactive development environments (jgtk and JHS) and comprehensive tutorials.
- Key features of J include built-in sparse arrays, efficient memory management, and the ability to handle large datasets effectively, with packages for plotting and other utilities.
- While the author sees potential in JDB for tick data storage, they note that J may not replace R for specialized tasks like Hidden Markov models but is valuable for numerics and different programming paradigms.
Show HN: Flint – A 30B model fine-tuned for less repetition
18 hours ago
- Flint is a language model designed specifically for inspiration and creative tasks, not for giving correct answers.
- It addresses the issue of output convergence in frontier LLMs, which leads to repetitive and similar responses, hindering divergent thinking.
- Flint achieves a dramatic increase in output diversity on creative tasks without degrading performance in other areas like accuracy or responsible AI benchmarks.
- On NoveltyBench, Flint scores 7.47 mean distinct responses out of 10, significantly outperforming SOTA models such as Gemini 3.1 Pro (3.19), GPT-5.4 (2.54), and Claude 4.6 Sonnet (1.83).
- In intra-model similarity tests, Flint has a mean similarity of 0.721 (lower is better), compared to higher scores for other models (e.g., GPT-5.4: 0.864, Gemini 3.1 Pro: 0.871, Claude 4.6 Sonnet: 0.905).
- Flint also shows lower inter-model similarity (0.672) than other models, making it the most distinctive in the comparison.
- It maintains performance on benchmarks like MMLU-STEM (78.9%), TruthfulQA MC1 (34.4%), and ToxiGen (59.6%), showing divergence tuning does not compromise capability.
- Flint is not a replacement for frontier models but acts as a multiplier in creative workflows, generating range while larger models provide depth and reasoning, with humans applying judgment.
- The model is available in alpha version via the Springboards app.
The API Tooling Crisis: Why developers are abandoning Postman and its clones?
6 hours ago
- The article criticizes the shift of API tools like Postman from developer-friendly to corporate-driven products, focusing on enshittification and performance issues.
- Key grievances include Postman's removal of offline mode, forced cloud synchronization, and bloated performance, alongside similar patterns in Insomnia and Thunder Client.
- Developers are turning to alternatives like Bruno for local-first, Git-integrated workflows, though these tools lack polish and some enterprise features.
- An ideal API tool is described as local-first, Git-native, high-performance, extensible, and free from mandatory logins or cloud locks.
- An update mentions Postman's introduction of Native Git features, but criticizes it as a surface-level change that still ties collaboration to a paid cloud model.
Turning a Chinese IoT camera into an owl livestream
a day ago
- The article describes a project to create a livestream of a Great Horned Owl nest using a Chinese IoT camera.
- Spencer, an artist and conservationist, set up an S4 Pro camera from Premium Invention (a rebranded Nice Intelligence Y5 Camera) in a forest.
- The NiView app used with the camera required app downloads and link sharing for viewing, which didn't scale well and drained battery.
- Initial attempts to extract a web viewer or stream file (e.g., .m3u8) via the manufacturer's websites were unsuccessful.
- Using mitmproxy to intercept network traffic revealed encrypted API requests to Tencent Cloud but no direct video stream URLs.
- The NiView app was decrypted from a jailbroken iPhone using bfdecrypt, allowing inspection of its code and resources.
- Debugging with lldb and Hopper identified key classes like NVLivePlayerView and IVVideoRender for video rendering.
- The video feed used a P2P relay via Tencent's servers (list.iotvideo.tencentcs.com), making direct reverse-engineering complex.
- A jailbreak tweak was developed using Theos to hook into IVVideoRender's _renderVideoFrame method, capture YUV frames, convert them to JPEGs, and stream them via a local HTTP server on port 8080.
- The tweak composited images (cropping and downsampling) to show the zoomed view with overlays like timestamps, and sent MJPEG streams to clients.
- A Python script and PHP wrapper were created to cache and rebroadcast the stream from an EC2 server, reducing load on the iPhone.
- Pitfalls included memory leaks in the tweak (fixed with static buffers and autorelease pools), image compositing challenges, and camera battery drain (solved temporarily with an external battery).
- The project highlights the technical hurdles of reverse-engineering proprietary IoT systems and offers a workaround for similar livestreaming needs, suggesting cameras with RTSP support for future projects.
The American Experiment Has Been Infected by Oligarchs
3 hours ago
- The article warns of an 'oligarchy virus' in the U.S. tax system, where extreme wealth concentration distorts democracy and culture.
- Wealth inequality has surged, with the top 0.00001% (19 billionaires) holding over $3 trillion, more than 12% of national income, and billionaire political spending skyrocketing.
- The tax system's flaw lies in not taxing unrealized gains (Haig-Simons income) and taxing investment gains at lower rates than wages, favoring the wealthy.
- Historical progressive taxation and strong labor policies once prevented oligarchy, but since the 1980s, wage stagnation, deregulation, and tax cuts have enabled wealth reconcentration.
- Proposed solutions include raising the minimum wage, enforcing antitrust laws, taxing investment gains as ordinary income, and implementing a wealth tax via the Oligarch Act of 2025.
- The oligarchy cycle (wealth begetting power) makes reform challenging, but confronting it is urgent to restore democratic balance.
KLM cancels 160 flights due to fuel shortage
3 hours ago
- The IMF warns the global economy faces a large shock due to a conflict blocking 20% of oil and gas in the Strait of Hormuz, impacting everyone, especially low-income countries.
- IMF chief advises against 'generous' subsidies for energy to avoid increasing demand and worsening the crisis, urging policies that shrink demand instead.
- France's central bank chief notes inflation is lower than in 2022 but stresses central banks will act to prevent second-round inflation effects, though not hastily.
- Saudi Arabia's finance minister warns that even with a ceasefire, recovery in energy markets will take weeks or months due to operational and insurance delays.
- KLM cancels 160 European flights due to rising fuel costs, while oil prices rise amid US threats to Iran's energy infrastructure.
- A Nigerian mega-refinery boosts jet fuel supplies to Europe, helping offset shortages caused by the conflict and benefiting from local crude purchases.
- Economic impacts include reduced growth forecasts for Greece due to high energy prices and a drop in US industrial output, signaling broader economic strain.
What's new in Claude Opus 4.7
5 hours ago
- Claude Opus 4.7 is the most capable generally available model with high autonomy and excels at long-horizon agentic work, knowledge work, vision tasks, and memory.
- It introduces high-resolution image support up to 2576px/3.75MP, enhancing vision-heavy workloads like document understanding, with pixel-accurate coordinates and no scale-factor math required.
- The model features a new effort parameter to tune intelligence vs. token spend, with 'xhigh' recommended for coding and agentic use cases, and 'high' for intelligence-sensitive tasks.
- Task budgets allow setting an advisory token limit for full agentic loops, helping the model prioritize work, with a minimum of 20k tokens, distinct from the hard max_tokens cap.
- Breaking changes include removal of extended thinking budgets (adaptive thinking only), non-default temperature/top_p/top_k parameters causing errors, and thinking content omitted by default unless opted in.
- A new tokenizer may use 1x to 1.35x more tokens, affecting token counts, and prompts may need updates for max_tokens and cost control, with improvements in knowledge-worker tasks and memory use.
Artifacts: Versioned storage that speaks Git
6 hours ago
- Agents are driving a massive increase in code generation, overwhelming traditional source control systems built for humans.
- Artifacts is a new distributed, versioned filesystem designed primarily for agents, compatible with Git and accessible via APIs.
- It enables creating repositories on-demand for agent sessions, sandboxes, and forks, with support for REST and Workers APIs.
- Artifacts leverages Git's data model for state tracking, time travel, and persistence, beyond just source control.
- Built on Cloudflare's Durable Objects, R2, and KV, it uses a custom Zig-based Git implementation compiled to WebAssembly for efficiency.
- ArtifactFS, an open-source filesystem driver, accelerates cloning large repos by hydrating files asynchronously, reducing startup time.
- Planned features include expanded metrics, event subscriptions, SDKs for multiple languages, and search APIs.
- Pricing is based on storage and operations, with free tiers and beta access starting privately before a public launch in early May 2026.
Laravel raised money and now injects ads directly into your agent
4 hours ago
- Laravel secured a $57M investment from Accel and now integrates promotion for Laravel Cloud into its official agent library.
- Developers criticize a PR change that exclusively suggests Laravel Cloud for deployment, removing mentions of alternatives, as a form of 'enshittification'.
- User feedback indicates this change 'poisons' agents by defaulting to Laravel Cloud, even in irrelevant scenarios, but the CEO prioritizes commercial benefits.
- Despite Laravel Cloud already receiving praise from agents like ChatGPT and Claude, the forced promotion risks community trust and raises ethical concerns.
- The situation sparks a broader discussion on advertising to AI agents and the potential need for 'agent-ad-block' tools as these practices become more discreet.
Launch HN: Kampala (YC W26) – Reverse-Engineer Apps into APIs
4 hours ago
- Kampala enables reverse engineering of websites, mobile apps, and desktop apps.
- It provides full traffic interception to monitor HTTP/S requests in real time.
- Features include auth chain tracing, flow replay with automation export, and fingerprint preservation.
- Windows support is planned with a waitlist available for notification.
OpenSSL 4.0.0 Released
21 hours ago
- OpenSSL 4.0.0 released with new features and incompatible changes.
- Support added for Encrypted Client Hello (ECH) and SM2 signature algorithms.
- Removed support for deprecated protocols like SSLv2, SSLv3, and engines.
- Key changes include standardized hexadecimal dumps and API function signature updates.
- FIPS self-tests can now be deferred with new installation options.
- OpenSSL 4.0 will be supported until May 14, 2027.
The becquerel as an SI unit for request rate
11 hours ago
- Request rate refers to the number of requests arriving, serviced, or leaving within a specified period.
- It's crucial to specify the period length when measuring request rate, often overlooked in dashboards.
- Requests per second is recommended as the standard period length for consistency across different dashboard views.
- Two SI units relevant to request rate: Hertz (Hz) for periodic events and Becquerel (Bq) for average frequency events.
- Hz implies regular, predictable timing, while Bq accounts for irregular, average-based occurrences.
- Hz suits regular load testing; Bq better represents organic traffic with average request rates.
- Using 'kBq' simplifies expressing high request rates compared to verbose 'requests per second'.
- Becquerel is officially for radioactive decay, but applying it to request rates is suggested for future adoption.
Rust 1.95.0
6 hours ago
- Rust 1.95.0 has been released, introducing new features and improvements.
- The new cfg_select! macro provides compile-time configuration selection similar to the cfg-if crate.
- If-let guards are now allowed in match expressions, enabling conditionals based on pattern matching.
- Multiple APIs have been stabilized, including MaybeUninit, Cell, Atomic types, and core::range.
- Support for JSON target specs on stable has been removed, requiring nightly for custom targets.
- Future releases can be tested by switching to beta or nightly channels using rustup.
Alibaba open-sources Qwen3.6-35B-A3B, a 35B MoE model with 3B active parameters
5 hours ago
- Qwen3.6-35B-A3B is a post-trained model released by the Qwen team, offering improved stability and utility based on community feedback.
- It features 35B total parameters (3B activated), a context length of up to 262,144 tokens (extendable to 1,010,000 with YaRN), and supports multimodal inputs including images and videos.
- Key enhancements include better agentic coding for frontend workflows and repository-level reasoning, along with thinking preservation to retain reasoning context from historical messages.
- The model is compatible with various inference frameworks such as SGLang, vLLM, KTransformers, and Hugging Face Transformers for deployment and serving.
- It achieves strong benchmark results in coding (e.g., SWE-bench, Terminal-Bench), knowledge (e.g., MMLU-Pro, C-Eval), and vision-language tasks (e.g., MMMU, MathVista).
- Usage involves API integration with recommended sampling parameters for different modes (thinking vs. non-thinking) and tasks, plus support for tool calling and agent applications via Qwen-Agent.
The Internet's Most Powerful Archiving Tool Is in Peril
6 hours ago
- USA Today's investigative report utilized the Wayback Machine to uncover ICE's delayed disclosure on detention policies, highlighting the tool's critical role in public interest journalism.
- Several major news organizations, including USA Today Co., The New York Times, and Reddit, are blocking the Internet Archive's crawler, citing concerns over AI misuse and copyright violations.
- Journalists and advocacy groups are rallying to support the Wayback Machine, emphasizing its importance for historical research, fact-checking, and preserving digital records amid declining traditional archives.
- Publishers blocking the tool argue it risks enabling AI companies to train models on their content without permission, a concern central to ongoing legal battles over copyright and AI training data.
- The trend threatens the Wayback Machine's mission of preserving web history, potentially eroding access to early digital records and impacting accountability journalism and legal evidence.
Claude Opus 4.7
5 hours ago
- Claude Opus 4.7 is now generally available, offering significant improvements in advanced software engineering and handling complex, long-running tasks with consistency and self-verification.
- It features substantially better vision with higher resolution image support (up to ~3.75 megapixels) and excels in professional tasks like creating interfaces, slides, and documents.
- Opus 4.7 demonstrates notable performance gains across benchmarks, including a 13% lift on a 93-task coding benchmark and improved efficiency in multi-step workflows.
- Despite being less capable than Claude Mythos Preview, it is the first model released with cybersecurity safeguards, and security professionals can join the Cyber Verification Program for legitimate use.
- It shows strong feedback from testers, with improvements in autonomy, reasoning, tool-call accuracy, and error recovery, making it effective for agentic and enterprise workflows.
- Key updates include a new 'xhigh' effort level, task budgets in the API, and features like '/ultrareview' in Claude Code, along with changes in tokenizer affecting token usage.
There's yet another study about how bad AI is for our brains
5 hours ago
- AI assistance initially improves performance but at a cognitive cost, reducing persistence and independent skills.
- Study on 350 Americans solving equations showed AI group's performance dropped sharply when access was cut, with similar results in larger experiments.
- Researchers warn AI in education could dilute human innovation and creativity by eroding motivation and long-term learning.
- Caveats: study not peer-reviewed; using AI for hints vs. answers had varying impacts on post-AI performance.
- Other studies link AI to 'AI brain fry' in workers, increased fatigue, and poorer test outcomes in students.
Fly Drones from the Browser
4 days ago
- FPVSIM is a browser-based drone flight simulator accessible online.
- Includes lap timer and V3 features, with manuals and how-tos for setup.
- Supports iOS/Android, OTA flashing, multi-device setup, and internet connectivity.
- Offers performance tips, P/D tuning, sync/diff with Betaflight/INAV, PID/rates from top pilots.
- Provides auto quad configurations, upgrade via authorization code, and beginner guides.
- Features tools like blackbox analyzer and radio latency checker.
- Requires login/signup with Google or email for access, with performance tips on initial load.

Hasty Briefsbeta

Japan implements language proficiency requirements for certain visa applicants

Where the DOGE Operatives Are Now

A look at the J language: the fine line between genius and insanity (2012)

Show HN: Flint – A 30B model fine-tuned for less repetition

The API Tooling Crisis: Why developers are abandoning Postman and its clones?

Turning a Chinese IoT camera into an owl livestream

The American Experiment Has Been Infected by Oligarchs

KLM cancels 160 flights due to fuel shortage

What's new in Claude Opus 4.7

Artifacts: Versioned storage that speaks Git

Laravel raised money and now injects ads directly into your agent

Launch HN: Kampala (YC W26) – Reverse-Engineer Apps into APIs

OpenSSL 4.0.0 Released

The becquerel as an SI unit for request rate

Rust 1.95.0

Alibaba open-sources Qwen3.6-35B-A3B, a 35B MoE model with 3B active parameters

The Internet's Most Powerful Archiving Tool Is in Peril

Claude Opus 4.7

There's yet another study about how bad AI is for our brains

Fly Drones from the Browser

Japan implements language proficiency requirements for certain visa applicants

Where the DOGE Operatives Are Now

A look at the J language: the fine line between genius and insanity (2012)

Show HN: Flint – A 30B model fine-tuned for less repetition

The API Tooling Crisis: Why developers are abandoning Postman and its clones?

Turning a Chinese IoT camera into an owl livestream

The American Experiment Has Been Infected by Oligarchs

KLM cancels 160 flights due to fuel shortage

What's new in Claude Opus 4.7

Artifacts: Versioned storage that speaks Git

Laravel raised money and now injects ads directly into your agent

Launch HN: Kampala (YC W26) – Reverse-Engineer Apps into APIs

OpenSSL 4.0.0 Released

The becquerel as an SI unit for request rate

Rust 1.95.0

Alibaba open-sources Qwen3.6-35B-A3B, a 35B MoE model with 3B active parameters

The Internet's Most Powerful Archiving Tool Is in Peril

Claude Opus 4.7

There's yet another study about how bad AI is for our brains

Fly Drones from the Browser