Hasty Briefsbeta

Bilingual

Claude Code may be burning your limits with invisible tokens
7 hours ago
- Anthropic's Claude Code tool appears to inject around 20,000 extra tokens into requests starting from version 2.1.100, causing users to hit usage limits faster.
- Users, even on the $200 monthly Max plan, report depleting their quotas within hours, sometimes as quickly as 90 minutes, despite Anthropic acknowledging the issue without explanation.
- A developer's proxy test revealed that v2.1.100 billed 69,922 tokens for the same task that used 49,726 tokens in v2.1.98, with the inflation occurring server-side and invisible to users.
- The extra tokens may be linked to new session memory features, potentially diluting custom instructions and degrading output quality in long sessions.
- Community workaround: downgrade to v2.1.98 using 'npx [email protected]'. This issue adds to user frustration after Anthropic recently restricted subscription limits for third-party tools.
- RunPod advertises cheaper, on-demand GPU access for AI/ML tasks, offering hardware like H100 and RTX 4090, billed per second with quick setup.
The Importance of Benchmarks
31 minutes ago
- Aha! Develop team found a sprint report taking 15 minutes to load due to a 1,000-day sprint, triggering a performance investigation.
- The performance bottleneck was identified in sprint progress parsing, affecting all sprint reports like retrospective and burndown.
- The original algorithm cloned data for each sprint day, including empty days, causing memory and speed issues.
- Hypothesis: memory usage scaled with days, making later days slower to clone as more records accumulated.
- Solution #1 introduced a new data structure (ParsedSprintProgressData) that stores empty days as null, reducing memory usage by up to 90%.
- Benchmarking revealed that memory improvements only led to a 20% runtime gain, disproving the initial hypothesis.
- Profiling identified two hot paths: lodash.cloneDeep() and moment.format(), which were major performance overheads.
- Solution #2 replaced cloneDeep() with explicit duplication and moment.format() with a custom date formatting function.
- Combining both approaches resulted in faster parsing, less memory usage, and minimal impact from empty days.
- The problematic report now parses in a fraction of a second, a three-order-of-magnitude improvement, though rendering still takes seconds.
Ascending into the Realm of Japanese Charts
3 hours ago
- The author discovers a 1920s Japanese statistical chart collection through online antiquarian markets, aided by AI for translation and research.
- AI tools help navigate and scale the search, uncovering digitized archives like the National Diet Library, transforming a simple purchase into a deep historical exploration.
- The collection reveals a coordinated public campaign in 1920s Japan to popularize statistical thinking, linked to events like the Great Kantō Earthquake and industrialization.
- Over 19 volumes are acquired through a logistical process involving Japanese dealers and a forwarding service, highlighting the challenges and successes of cross-border collecting.
- The essay argues that Western-centric graphic history overlooks significant Asian works due to language barriers, uneven digitization, and archival depth, emphasizing the value of curiosity and taste in uncovering hidden gems.
N-Day-Bench – Can LLMs find real vulnerabilities in real codebases?
an hour ago
- N-Day-Bench measures LLMs' ability to discover real-world vulnerabilities disclosed after their knowledge cut-off.
- The benchmark is adaptive, with monthly updates to test cases and model versions.
- All traces are publicly accessible, providing transparency into the evaluation process.
- Leading models include GPT-5.4, GLM-5.1, and Claude Opus-4.6, with GPT-5.4 achieving the highest average score of 83.93.
Meta creating AI version of Mark Zuckerberg so staff can talk to the boss
6 hours ago
- Meta is developing an AI version of Mark Zuckerberg trained on his mannerisms, tone, and public statements to help employees feel more connected.
- The AI character, using Zuckerberg's images and voice, is part of Meta's broader AI integration to enhance efficiency, lower costs, and accelerate work.
- Meta is investing billions in AI, including advanced models like Muse Spark, to remain competitive, despite facing legal issues over platform safety and addiction concerns.
Ukraine Tested Space-Launch Capabilities in Middle of War Twice
2 hours ago
- Ukrainian intelligence successfully conducted two rocket launches into space during the war, with altitudes of over 100 km and 204 km.
- The launches were combat missions under military intelligence leadership, using hypersonic missiles capable of striking up to 500 km.
- Ukraine also launched a rocket carrier from a transport aircraft at 8 km altitude, a first in Europe and record-setting.
- This air-launch technology could evolve into an 'airborne spaceport', enhancing range and efficiency for civilian and military use.
- Ukraine's UASAT LEO satellite constellation, with about 300 satellites, is set to launch its first satellite via SpaceX in October.
The Perils of an Over-Optimized Life
2 hours ago
- Practiced a morning routine called POWER for 15 years, but once created an unsustainable 'Ultimate Morning Routine' waking at 3am for intense study, prayer, and exercise.
- The over-ambitious routine failed immediately due to lack of sleep and mental strain, highlighting the importance of consistency over idealism.
- Observes a pattern among productivity-minded Christians of over-optimizing routines into impossibility, stressing that fruitfulness comes from simple habits, not machine-like precision.
- Suggests focusing on key habits like daily devotions, healthy living, planning, and curated information as the 80/20 for effective productivity, advising to avoid overcomplication.
(AMD) Build AI Agents That Run Locally
4 hours ago
- GAIA is an open-source framework for building AI agents that run locally on Python and C++.
- All processing is on-device with no cloud dependency, API keys, or data leaving the device.
- It supports AMD optimization with NPU and GPU acceleration on Ryzen AI.
- Features include an Agent UI for desktop chat, document Q&A (RAG), speech-to-speech, code generation, image generation, MCP integration, and agent routing.
- Installation is simple via npm or commands, with quick start guides for both Python and C++.
- Resources include SDK reference, GitHub source, component specs, glossary, and Discord community.
Stanford report highlights growing disconnect between AI insiders and everyone
2 hours ago
- AI experts and the public increasingly disagree on AI's impact, with experts more optimistic than the general public.
- Public anxiety about AI is growing, especially in the U.S., focusing on jobs, healthcare, and the economy, while experts worry more about AGI risks.
- Negative sentiment is led by Gen Z, with many young people becoming less hopeful and more angry despite high AI usage.
- Online reactions to events like attacks on AI leaders reflect public frustration, drawing comparisons to incidents involving corporate executives.
- Survey data shows a large gap: 56% of experts foresee positive AI impacts in 20 years, versus only 10% of Americans being more excited than concerned.
- Divergence is stark in specific areas: 84% of experts see AI benefiting healthcare, compared to 44% of the public; similar gaps exist for jobs and the economy.
- Americans distrust government regulation of AI, with only 31% trusting responsible oversight, the lowest among nations surveyed.
- Globally, perception of AI benefits slightly increased from 55% to 59%, but nervousness about AI also rose from 50% to 52%.
Civilian volunteers across Europe 3D print weapon parts for Ukraine
3 hours ago
- A volunteer network across Europe uses 3D printers to produce drone parts and bomb casings for the Ukrainian military.
- These operations, run from private spaces like garages and spare rooms, work around the clock to fill supply gaps when official channels fail.
- The printed items include mechanical safety switches, LED torches, antennas, and lightweight plastic casings that extend drone range.
- One network, involving about 400 volunteers, produced 100 tonnes of plastic items last year, sent directly to soldiers after verification via a restricted website.
- Volunteers, who are civilians, print only the non-explosive components; the military adds explosives and detonators to complete the bombs.
- Similar groups exist in the UK, with individuals like Etienne Paresys joining networks such as Print Army to contribute despite having no prior 3D printing experience.
- Experts note this grassroots, bottom-up mobilization is emblematic of modern Ukrainian society, enabling decentralized manufacturing that reduces supply chain demands.
New Mexico governor signs nation's first universal child care law
4 hours ago
- Governor Lujan Grisham signed a universal child care law, making New Mexico the first state in the U.S. to guarantee no-cost child care for families.
- A second law removed local zoning and HOA barriers to expand child care availability in the state, supporting the governor's 'cradle to career' education plan.
- Universal child care aims to help parents work, improve family quality of life, and boost the state's economy, removing the financial burden of child care costs.
- The program has already enrolled thousands of additional families and children, with tens of thousands expected to receive free care this year.
- Funding includes up to $700 million from the Early Childhood Education and Care Fund through 2031, with fiscal safeguards for sustainability.
- Investments have increased child care wages significantly and grown the number of professionals in the field, addressing workforce needs.
- New Mexico now offers both tuition-free college and universal child care, positioning it as a national model for other states to follow.
Zuckerberg Didn't Think He Was Robotic Enough So Now He's Using AI
3 hours ago
- Mark Zuckerberg's company Meta is developing an AI clone of him using his image, voice, mannerisms, and strategic thoughts for employee interactions.
- This initiative aims to help Meta catch up to competitors like Google and OpenAI in AI technology, despite challenges in scaling 3D AI avatars due to high computing power requirements.
- The project draws parallels to Meta's failed metaverse venture, a costly VR endeavor that didn't gain traction despite significant investment and company rebranding.
- Critics view this AI clone effort as another questionable move in the name of 'efficiency,' potentially impacting jobs, data center resources, and digital safety.
New Orleans's Car-Crash Conspiracy
5 hours ago
- Tractor-trailers are a major risk on U.S. highways, accounting for 10% of fatal accidents despite being 5% of vehicles.
- A staged accident fraud scheme in New Orleans involved intentional collisions with trucks to file lucrative lawsuits.
- Cornelius Garrison was a key 'slammer' who orchestrated crashes, working with lawyers like Vanessa Motta and Sean Alfortish.
- The fraud exploited high insurance limits on trucks, with lawyers and doctors profiting while passengers risked injury.
- Garrison cooperated with the FBI, but was murdered in 2020, revealing ties to Red Harris and others in the conspiracy.
- Vanessa Motta was convicted of fraud and related charges in 2025, while Alfortish faces murder conspiracy charges.
- The scheme highlights systemic issues in Louisiana's legal and insurance systems, with high auto-insurance rates and minimal reform.
Just Enough Chimera Linux
3 hours ago
- Chimera Linux is a community-built Linux distribution that uses musl as its C library, dinit for system initialization, and FreeBSD core tools for its userland.
- This guide details the installation of Chimera Linux with OpenZFS filesystem and ZFSBootMenu bootloader on a single disk using a two-partition layout (ESP and ZFS pool) with native encryption and zram for swap.
- Key steps include preparing the live environment, partitioning the disk, creating an encrypted ZFS pool, installing the base system, configuring the OS, setting up ZFSBootMenu, and enabling zram swap.
- Assumptions for the installation: x86_64 architecture with UEFI boot, disabled Secure Boot, wired network connection, and no hibernation support.
- Post-installation configuration covers setting root password, creating a user with superuser privileges, installing packages, configuring console font/keyboard, timezone, hostname, services, and ESP partition.
GitHub Stacked PRs
2 hours ago
- Stacked PRs allow arranging pull requests in an ordered stack to merge them all in one click.
- Each PR represents one focused layer of a change, reviewed independently and landed together.
- GitHub provides a UI for navigating between PRs in a stack and checking each layer's status.
- A powerful CLI (`gh stack`) facilitates creating stacks, performing rebases, pushing branches, creating PRs, and navigating layers.
- AI agents can be integrated to teach them how to work with stacks using `npx skills add github/gh-stack`.
- Stacked PRs address issues with large pull requests by breaking big changes into small, focused PRs.
- A stack consists of a chain of PRs where each targets the branch of the PR below it, ultimately landing on the main branch.
- GitHub supports stacks end-to-end with UI features like stack maps, branch protection enforcement, and CI runs.
- The CLI handles local workflow, while GitHub's PR UI provides context for reviewers.
- Merging can be done for all or part of the stack, with automatic rebasing of remaining PRs after a merge.
Picasso's Guernica (Gigapixel)
3 days ago
- Recent photographic technical studies on Guernica by the Reina Sofía Museum's restoration team have revealed important conclusions for researchers and restorers.
- Changes in Picasso's painting composition are documented by Dora Maar's photographs and now visible through infrared reflectography.
- Key findings include: the soldier's head was originally upside down and later rotated; hidden eyes in the bull's head; and alterations from the canvas's multiple mounting and unmounting.
- The artwork has suffered craquelados (cracks), fissures, and loss of material due to handling and rolling, with craquelados being more severe.
- Wax accumulations on the surface result from a 1957 restoration at MoMA using a wax-resin mixture applied from the reverse of the canvas.
- Residues from a 1974 vandalism act with red acrylic spray paint are still visible as micro-residues, though the original painting was protected by a varnish applied in 1962.
- A glossary details alterations like abrasion, craquelado, and repinte, as well as technical data including changes in composition and underlying drawing.
What We Learned Building a Rust Runtime for TypeScript
3 days ago
- Encore's TypeScript support involved creating a new runtime in Rust instead of extending the Go runtime or using TypeScript directly, to avoid reimplementing infrastructure for each new language and to overcome Node.js's single-threaded limitations.
- The Rust runtime handles all infrastructure concerns like HTTP lifecycle, database connections, pub/sub, and tracing, leaving TypeScript for business logic only, which improves performance by leveraging multi-threading.
- A prototype using Go as a sidecar with Node.js was discarded due to high latency from IPC overhead and operational complexity from managing multiple processes.
- The runtime is integrated into the Node.js process via Rust NAPI bindings, allowing direct interaction and avoiding serialization, with enhancements to handle JavaScript promises and cancellation guards.
- Application metadata (generated at compile time) and runtime configuration (generated at deploy time) are separated, enabling the same application to run in different environments without code changes.
- An embedded gateway using Pingora handles routing, auth, and more within the same process, allowing user-defined TypeScript auth handlers to execute directly without IPC.
- Pub/sub and object storage systems abstract cloud provider differences through trait objects, supporting NSQ, GCP, and AWS with consistent interfaces.
- A custom binary trace protocol with efficient serialization and time anchoring provides detailed, sampled tracing without overhead, though an OpenTelemetry adapter is planned.
- A TypeScript parser built on SWC extracts infrastructure declarations from code to generate metadata, enabling features like architecture diagrams and API docs.
- Key learnings include investing in structured error types, prioritizing OpenTelemetry support, using more snapshot testing, and exploring a unified runtime for Go and Rust.
- Benchmarks show Encore.ts outperforms frameworks like Express.js with significantly higher throughput and lower latency, especially with validation enabled.
An Introduction to Obsidian
2 days ago
- Obsidian is a local markdown-based note-taking app, ensuring file ownership and protection from enshittification.
- It offers powerful features like interlinking notes, extensive plugin support, and consistent functionality across desktop and mobile.
- Users are advised to explore Obsidian strategically, avoiding FOMO, and keeping their setup simple for effective knowledge work.
- Common uses include content creation, knowledge management, project tracking, and media organization with a preference for minimal plugins.
- Syncing is achievable via tools like Google Drive and GitHub, while features like graph view and canvas are used sparingly.
- While Obsidian is a primary tool, alternatives like Notion (collaboration) and Logseq (journaling) serve complementary roles.
Put your SSH keys in your TPM chip
a day ago
- SSH private keys can be stored in a TPM chip to enhance security, preventing extraction and keeping keys off the filesystem.
- TPMs are less secure than portable HSMs like Yubikey since they are device-bound and don't require physical presence.
- Install tpm2-tools and related software, then create a PKCS#11 token in a persistent directory like ~/.tpm2_pkcs11.
- Import an existing SSH key into the TPM by converting it to PEM format and using tpm2_ptool with a user PIN.
- Use the TPM SSH key by setting TPM2_PKCS11_SO environment variable and configuring SSH to use the PKCS#11 provider.
B-trees and database indexes (2024)
6 hours ago
- B-trees are foundational data structures in DBMS like MySQL and MongoDB, used for efficient data lookups via indexes.
- B-trees store ordered key/value pairs in nodes, with each node sized to match disk blocks for optimized persistent storage.
- B+trees, a variant used in MySQL, store values only in leaf nodes, have linked lists for in-order traversal, and allow more keys in inner nodes.
- MySQL's InnoDB engine stores table data in a B+tree with the primary key as the tree key; secondary indexes use additional B+trees.
- Choosing a sequential primary key (e.g., auto-incrementing integer) over random keys like UUIDs reduces I/O and improves insert/query performance.
- Primary key size matters: smaller keys (e.g., BIGINT vs. UUID) allow more keys per node, leading to shallower trees and faster lookups.
- InnoDB uses a buffer pool to cache pages in memory, reducing disk I/O, but minimizing page visits remains crucial for performance.

first prev0next

About|Login

Bilingual

Claude Code may be burning your limits with invisible tokens
7 hours ago
- Anthropic's Claude Code tool appears to inject around 20,000 extra tokens into requests starting from version 2.1.100, causing users to hit usage limits faster.
- Users, even on the $200 monthly Max plan, report depleting their quotas within hours, sometimes as quickly as 90 minutes, despite Anthropic acknowledging the issue without explanation.
- A developer's proxy test revealed that v2.1.100 billed 69,922 tokens for the same task that used 49,726 tokens in v2.1.98, with the inflation occurring server-side and invisible to users.
- The extra tokens may be linked to new session memory features, potentially diluting custom instructions and degrading output quality in long sessions.
- Community workaround: downgrade to v2.1.98 using 'npx [email protected]'. This issue adds to user frustration after Anthropic recently restricted subscription limits for third-party tools.
- RunPod advertises cheaper, on-demand GPU access for AI/ML tasks, offering hardware like H100 and RTX 4090, billed per second with quick setup.
The Importance of Benchmarks
31 minutes ago
- Aha! Develop team found a sprint report taking 15 minutes to load due to a 1,000-day sprint, triggering a performance investigation.
- The performance bottleneck was identified in sprint progress parsing, affecting all sprint reports like retrospective and burndown.
- The original algorithm cloned data for each sprint day, including empty days, causing memory and speed issues.
- Hypothesis: memory usage scaled with days, making later days slower to clone as more records accumulated.
- Solution #1 introduced a new data structure (ParsedSprintProgressData) that stores empty days as null, reducing memory usage by up to 90%.
- Benchmarking revealed that memory improvements only led to a 20% runtime gain, disproving the initial hypothesis.
- Profiling identified two hot paths: lodash.cloneDeep() and moment.format(), which were major performance overheads.
- Solution #2 replaced cloneDeep() with explicit duplication and moment.format() with a custom date formatting function.
- Combining both approaches resulted in faster parsing, less memory usage, and minimal impact from empty days.
- The problematic report now parses in a fraction of a second, a three-order-of-magnitude improvement, though rendering still takes seconds.
Ascending into the Realm of Japanese Charts
3 hours ago
- The author discovers a 1920s Japanese statistical chart collection through online antiquarian markets, aided by AI for translation and research.
- AI tools help navigate and scale the search, uncovering digitized archives like the National Diet Library, transforming a simple purchase into a deep historical exploration.
- The collection reveals a coordinated public campaign in 1920s Japan to popularize statistical thinking, linked to events like the Great Kantō Earthquake and industrialization.
- Over 19 volumes are acquired through a logistical process involving Japanese dealers and a forwarding service, highlighting the challenges and successes of cross-border collecting.
- The essay argues that Western-centric graphic history overlooks significant Asian works due to language barriers, uneven digitization, and archival depth, emphasizing the value of curiosity and taste in uncovering hidden gems.
N-Day-Bench – Can LLMs find real vulnerabilities in real codebases?
an hour ago
- N-Day-Bench measures LLMs' ability to discover real-world vulnerabilities disclosed after their knowledge cut-off.
- The benchmark is adaptive, with monthly updates to test cases and model versions.
- All traces are publicly accessible, providing transparency into the evaluation process.
- Leading models include GPT-5.4, GLM-5.1, and Claude Opus-4.6, with GPT-5.4 achieving the highest average score of 83.93.
Meta creating AI version of Mark Zuckerberg so staff can talk to the boss
6 hours ago
- Meta is developing an AI version of Mark Zuckerberg trained on his mannerisms, tone, and public statements to help employees feel more connected.
- The AI character, using Zuckerberg's images and voice, is part of Meta's broader AI integration to enhance efficiency, lower costs, and accelerate work.
- Meta is investing billions in AI, including advanced models like Muse Spark, to remain competitive, despite facing legal issues over platform safety and addiction concerns.
Ukraine Tested Space-Launch Capabilities in Middle of War Twice
2 hours ago
- Ukrainian intelligence successfully conducted two rocket launches into space during the war, with altitudes of over 100 km and 204 km.
- The launches were combat missions under military intelligence leadership, using hypersonic missiles capable of striking up to 500 km.
- Ukraine also launched a rocket carrier from a transport aircraft at 8 km altitude, a first in Europe and record-setting.
- This air-launch technology could evolve into an 'airborne spaceport', enhancing range and efficiency for civilian and military use.
- Ukraine's UASAT LEO satellite constellation, with about 300 satellites, is set to launch its first satellite via SpaceX in October.
The Perils of an Over-Optimized Life
2 hours ago
- Practiced a morning routine called POWER for 15 years, but once created an unsustainable 'Ultimate Morning Routine' waking at 3am for intense study, prayer, and exercise.
- The over-ambitious routine failed immediately due to lack of sleep and mental strain, highlighting the importance of consistency over idealism.
- Observes a pattern among productivity-minded Christians of over-optimizing routines into impossibility, stressing that fruitfulness comes from simple habits, not machine-like precision.
- Suggests focusing on key habits like daily devotions, healthy living, planning, and curated information as the 80/20 for effective productivity, advising to avoid overcomplication.
(AMD) Build AI Agents That Run Locally
4 hours ago
- GAIA is an open-source framework for building AI agents that run locally on Python and C++.
- All processing is on-device with no cloud dependency, API keys, or data leaving the device.
- It supports AMD optimization with NPU and GPU acceleration on Ryzen AI.
- Features include an Agent UI for desktop chat, document Q&A (RAG), speech-to-speech, code generation, image generation, MCP integration, and agent routing.
- Installation is simple via npm or commands, with quick start guides for both Python and C++.
- Resources include SDK reference, GitHub source, component specs, glossary, and Discord community.
Stanford report highlights growing disconnect between AI insiders and everyone
2 hours ago
- AI experts and the public increasingly disagree on AI's impact, with experts more optimistic than the general public.
- Public anxiety about AI is growing, especially in the U.S., focusing on jobs, healthcare, and the economy, while experts worry more about AGI risks.
- Negative sentiment is led by Gen Z, with many young people becoming less hopeful and more angry despite high AI usage.
- Online reactions to events like attacks on AI leaders reflect public frustration, drawing comparisons to incidents involving corporate executives.
- Survey data shows a large gap: 56% of experts foresee positive AI impacts in 20 years, versus only 10% of Americans being more excited than concerned.
- Divergence is stark in specific areas: 84% of experts see AI benefiting healthcare, compared to 44% of the public; similar gaps exist for jobs and the economy.
- Americans distrust government regulation of AI, with only 31% trusting responsible oversight, the lowest among nations surveyed.
- Globally, perception of AI benefits slightly increased from 55% to 59%, but nervousness about AI also rose from 50% to 52%.
Civilian volunteers across Europe 3D print weapon parts for Ukraine
3 hours ago
- A volunteer network across Europe uses 3D printers to produce drone parts and bomb casings for the Ukrainian military.
- These operations, run from private spaces like garages and spare rooms, work around the clock to fill supply gaps when official channels fail.
- The printed items include mechanical safety switches, LED torches, antennas, and lightweight plastic casings that extend drone range.
- One network, involving about 400 volunteers, produced 100 tonnes of plastic items last year, sent directly to soldiers after verification via a restricted website.
- Volunteers, who are civilians, print only the non-explosive components; the military adds explosives and detonators to complete the bombs.
- Similar groups exist in the UK, with individuals like Etienne Paresys joining networks such as Print Army to contribute despite having no prior 3D printing experience.
- Experts note this grassroots, bottom-up mobilization is emblematic of modern Ukrainian society, enabling decentralized manufacturing that reduces supply chain demands.
New Mexico governor signs nation's first universal child care law
4 hours ago
- Governor Lujan Grisham signed a universal child care law, making New Mexico the first state in the U.S. to guarantee no-cost child care for families.
- A second law removed local zoning and HOA barriers to expand child care availability in the state, supporting the governor's 'cradle to career' education plan.
- Universal child care aims to help parents work, improve family quality of life, and boost the state's economy, removing the financial burden of child care costs.
- The program has already enrolled thousands of additional families and children, with tens of thousands expected to receive free care this year.
- Funding includes up to $700 million from the Early Childhood Education and Care Fund through 2031, with fiscal safeguards for sustainability.
- Investments have increased child care wages significantly and grown the number of professionals in the field, addressing workforce needs.
- New Mexico now offers both tuition-free college and universal child care, positioning it as a national model for other states to follow.
Zuckerberg Didn't Think He Was Robotic Enough So Now He's Using AI
3 hours ago
- Mark Zuckerberg's company Meta is developing an AI clone of him using his image, voice, mannerisms, and strategic thoughts for employee interactions.
- This initiative aims to help Meta catch up to competitors like Google and OpenAI in AI technology, despite challenges in scaling 3D AI avatars due to high computing power requirements.
- The project draws parallels to Meta's failed metaverse venture, a costly VR endeavor that didn't gain traction despite significant investment and company rebranding.
- Critics view this AI clone effort as another questionable move in the name of 'efficiency,' potentially impacting jobs, data center resources, and digital safety.
New Orleans's Car-Crash Conspiracy
5 hours ago
- Tractor-trailers are a major risk on U.S. highways, accounting for 10% of fatal accidents despite being 5% of vehicles.
- A staged accident fraud scheme in New Orleans involved intentional collisions with trucks to file lucrative lawsuits.
- Cornelius Garrison was a key 'slammer' who orchestrated crashes, working with lawyers like Vanessa Motta and Sean Alfortish.
- The fraud exploited high insurance limits on trucks, with lawyers and doctors profiting while passengers risked injury.
- Garrison cooperated with the FBI, but was murdered in 2020, revealing ties to Red Harris and others in the conspiracy.
- Vanessa Motta was convicted of fraud and related charges in 2025, while Alfortish faces murder conspiracy charges.
- The scheme highlights systemic issues in Louisiana's legal and insurance systems, with high auto-insurance rates and minimal reform.
Just Enough Chimera Linux
3 hours ago
- Chimera Linux is a community-built Linux distribution that uses musl as its C library, dinit for system initialization, and FreeBSD core tools for its userland.
- This guide details the installation of Chimera Linux with OpenZFS filesystem and ZFSBootMenu bootloader on a single disk using a two-partition layout (ESP and ZFS pool) with native encryption and zram for swap.
- Key steps include preparing the live environment, partitioning the disk, creating an encrypted ZFS pool, installing the base system, configuring the OS, setting up ZFSBootMenu, and enabling zram swap.
- Assumptions for the installation: x86_64 architecture with UEFI boot, disabled Secure Boot, wired network connection, and no hibernation support.
- Post-installation configuration covers setting root password, creating a user with superuser privileges, installing packages, configuring console font/keyboard, timezone, hostname, services, and ESP partition.
GitHub Stacked PRs
2 hours ago
- Stacked PRs allow arranging pull requests in an ordered stack to merge them all in one click.
- Each PR represents one focused layer of a change, reviewed independently and landed together.
- GitHub provides a UI for navigating between PRs in a stack and checking each layer's status.
- A powerful CLI (`gh stack`) facilitates creating stacks, performing rebases, pushing branches, creating PRs, and navigating layers.
- AI agents can be integrated to teach them how to work with stacks using `npx skills add github/gh-stack`.
- Stacked PRs address issues with large pull requests by breaking big changes into small, focused PRs.
- A stack consists of a chain of PRs where each targets the branch of the PR below it, ultimately landing on the main branch.
- GitHub supports stacks end-to-end with UI features like stack maps, branch protection enforcement, and CI runs.
- The CLI handles local workflow, while GitHub's PR UI provides context for reviewers.
- Merging can be done for all or part of the stack, with automatic rebasing of remaining PRs after a merge.
Picasso's Guernica (Gigapixel)
3 days ago
- Recent photographic technical studies on Guernica by the Reina Sofía Museum's restoration team have revealed important conclusions for researchers and restorers.
- Changes in Picasso's painting composition are documented by Dora Maar's photographs and now visible through infrared reflectography.
- Key findings include: the soldier's head was originally upside down and later rotated; hidden eyes in the bull's head; and alterations from the canvas's multiple mounting and unmounting.
- The artwork has suffered craquelados (cracks), fissures, and loss of material due to handling and rolling, with craquelados being more severe.
- Wax accumulations on the surface result from a 1957 restoration at MoMA using a wax-resin mixture applied from the reverse of the canvas.
- Residues from a 1974 vandalism act with red acrylic spray paint are still visible as micro-residues, though the original painting was protected by a varnish applied in 1962.
- A glossary details alterations like abrasion, craquelado, and repinte, as well as technical data including changes in composition and underlying drawing.
What We Learned Building a Rust Runtime for TypeScript
3 days ago
- Encore's TypeScript support involved creating a new runtime in Rust instead of extending the Go runtime or using TypeScript directly, to avoid reimplementing infrastructure for each new language and to overcome Node.js's single-threaded limitations.
- The Rust runtime handles all infrastructure concerns like HTTP lifecycle, database connections, pub/sub, and tracing, leaving TypeScript for business logic only, which improves performance by leveraging multi-threading.
- A prototype using Go as a sidecar with Node.js was discarded due to high latency from IPC overhead and operational complexity from managing multiple processes.
- The runtime is integrated into the Node.js process via Rust NAPI bindings, allowing direct interaction and avoiding serialization, with enhancements to handle JavaScript promises and cancellation guards.
- Application metadata (generated at compile time) and runtime configuration (generated at deploy time) are separated, enabling the same application to run in different environments without code changes.
- An embedded gateway using Pingora handles routing, auth, and more within the same process, allowing user-defined TypeScript auth handlers to execute directly without IPC.
- Pub/sub and object storage systems abstract cloud provider differences through trait objects, supporting NSQ, GCP, and AWS with consistent interfaces.
- A custom binary trace protocol with efficient serialization and time anchoring provides detailed, sampled tracing without overhead, though an OpenTelemetry adapter is planned.
- A TypeScript parser built on SWC extracts infrastructure declarations from code to generate metadata, enabling features like architecture diagrams and API docs.
- Key learnings include investing in structured error types, prioritizing OpenTelemetry support, using more snapshot testing, and exploring a unified runtime for Go and Rust.
- Benchmarks show Encore.ts outperforms frameworks like Express.js with significantly higher throughput and lower latency, especially with validation enabled.
An Introduction to Obsidian
2 days ago
- Obsidian is a local markdown-based note-taking app, ensuring file ownership and protection from enshittification.
- It offers powerful features like interlinking notes, extensive plugin support, and consistent functionality across desktop and mobile.
- Users are advised to explore Obsidian strategically, avoiding FOMO, and keeping their setup simple for effective knowledge work.
- Common uses include content creation, knowledge management, project tracking, and media organization with a preference for minimal plugins.
- Syncing is achievable via tools like Google Drive and GitHub, while features like graph view and canvas are used sparingly.
- While Obsidian is a primary tool, alternatives like Notion (collaboration) and Logseq (journaling) serve complementary roles.
Put your SSH keys in your TPM chip
a day ago
- SSH private keys can be stored in a TPM chip to enhance security, preventing extraction and keeping keys off the filesystem.
- TPMs are less secure than portable HSMs like Yubikey since they are device-bound and don't require physical presence.
- Install tpm2-tools and related software, then create a PKCS#11 token in a persistent directory like ~/.tpm2_pkcs11.
- Import an existing SSH key into the TPM by converting it to PEM format and using tpm2_ptool with a user PIN.
- Use the TPM SSH key by setting TPM2_PKCS11_SO environment variable and configuring SSH to use the PKCS#11 provider.
B-trees and database indexes (2024)
6 hours ago
- B-trees are foundational data structures in DBMS like MySQL and MongoDB, used for efficient data lookups via indexes.
- B-trees store ordered key/value pairs in nodes, with each node sized to match disk blocks for optimized persistent storage.
- B+trees, a variant used in MySQL, store values only in leaf nodes, have linked lists for in-order traversal, and allow more keys in inner nodes.
- MySQL's InnoDB engine stores table data in a B+tree with the primary key as the tree key; secondary indexes use additional B+trees.
- Choosing a sequential primary key (e.g., auto-incrementing integer) over random keys like UUIDs reduces I/O and improves insert/query performance.
- Primary key size matters: smaller keys (e.g., BIGINT vs. UUID) allow more keys per node, leading to shallower trees and faster lookups.
- InnoDB uses a buffer pool to cache pages in memory, reducing disk I/O, but minimizing page visits remains crucial for performance.

Hasty Briefsbeta

Claude Code may be burning your limits with invisible tokens

The Importance of Benchmarks

Ascending into the Realm of Japanese Charts

N-Day-Bench – Can LLMs find real vulnerabilities in real codebases?

Meta creating AI version of Mark Zuckerberg so staff can talk to the boss

Ukraine Tested Space-Launch Capabilities in Middle of War Twice

The Perils of an Over-Optimized Life

(AMD) Build AI Agents That Run Locally

Stanford report highlights growing disconnect between AI insiders and everyone

Civilian volunteers across Europe 3D print weapon parts for Ukraine

New Mexico governor signs nation's first universal child care law

Zuckerberg Didn't Think He Was Robotic Enough So Now He's Using AI

New Orleans's Car-Crash Conspiracy

Just Enough Chimera Linux

GitHub Stacked PRs

Picasso's Guernica (Gigapixel)

What We Learned Building a Rust Runtime for TypeScript

An Introduction to Obsidian

Put your SSH keys in your TPM chip

B-trees and database indexes (2024)

Claude Code may be burning your limits with invisible tokens

The Importance of Benchmarks

Ascending into the Realm of Japanese Charts

N-Day-Bench – Can LLMs find real vulnerabilities in real codebases?

Meta creating AI version of Mark Zuckerberg so staff can talk to the boss

Ukraine Tested Space-Launch Capabilities in Middle of War Twice

The Perils of an Over-Optimized Life

(AMD) Build AI Agents That Run Locally

Stanford report highlights growing disconnect between AI insiders and everyone

Civilian volunteers across Europe 3D print weapon parts for Ukraine

New Mexico governor signs nation's first universal child care law

Zuckerberg Didn't Think He Was Robotic Enough So Now He's Using AI

New Orleans's Car-Crash Conspiracy

Just Enough Chimera Linux

GitHub Stacked PRs

Picasso's Guernica (Gigapixel)

What We Learned Building a Rust Runtime for TypeScript

An Introduction to Obsidian

Put your SSH keys in your TPM chip

B-trees and database indexes (2024)