- #zig #perf
- Zig profiling on Apple Silicon | Bugsik
- mstange/samply
- a command line sampling CPU profiler which uses the Firefox profiler as its UI
- for macOS, Linux, and Windows
- verte-zerg/poop at kperf-macos
- a fork of poop to use kperf under Mac
- wolfpld/tracy: Frame profiler
- A real time, nanosecond resolution, remote telemetry, hybrid frame and sampling profiler for games and other applications
- C, C++, Lua, Python, Fortran, Rust, Zig, C#, OCaml, Odin, GPU
- Windows, Linux, Android, FreeBSD, WSL, OSX, iOS, QNX
- Generic Programming and anytype - Docs - Ziggit
- Using Zig allocator for C libraries (Alignment question)
- `Allocator.rawAlloc`/`.rawFree`
- `@alignOf(std.c.max_align_t)`
- ANDRVV/zprof: 🧮 Cross-allocator profiler for Zig
- battle-tested in raptodb/rapto: transposition-heuristic storage, low memory footprint and high-performance querying
- Writing memory efficient C structs (on HN)
- 10^19712 years to exhaust u65535
- u65535 is the integer type with the highest max value supported in Zig (from Zig doc)
- #physics
- Background independence
- #crypto
- Fintech dystopia (on HN)
- #perf
- Giving Benchmarks a Boat
- TPC-C is a benchmark for transactional databases that models an order system with goods spanning many warehouses
- Each warehouse comes with some set amount of data to be managed and some amount of query traffic that adheres to some specific distribution laid out in the spec
- they can account for
- what's the proportion of hot vs. cold data
- what's the ratio of transactions-per-minute to bytes stored
- what's the proportion of read-only queries, vs. simple, point writes, vs. complex read-write transactions
- what proportion of transactions go cross-shard
- The metric: "we sustained the workload at X warehouses with Y hardware at Z cost."
- #movie
- Nothing to watch – Experimental gallery visualizing 50k film posters
- #agent
- Show HN: A GitHub Action that quizzes you on a pull request (on HN) #aicr
> PR Quiz uses AI to generate a quiz from a pull request and blocks you from merging until the quiz is passed.
- Principles for production AI agents (on HN)
- context
- modern LLMs just need direct detailed context, no tricks, but clarity and lack of contradictions
- structuring context so that the system part is large and static and user one is small and dynamic works great
- provide the bare minimum of knowledge in the first place, and the option to fetch more context if needed via tools
- context compaction tools can help avoiding logs and other artifacts from the feedback loop to bloat the context
- tools
- good tools typically operate on a similar level of granularity, and have a limited number of strictly typed parameters
- idempotency is highly recommended to avoid state management issues
- LLMs are very likely to misuse your loopholes, and that’s why you don’t want to have any loopholes
- v.s. human API users are more capable of reading between the lines, can navigate complex docs and find workarounds
- designing an agent to write some DSL (domain-specific language) code with actions rather than calling tools one by one is a great idea
- feedback loop
- actor-critic approach: where an actor decides on actions and a critic evaluates them
- allow actors to be creative, and critics to be strict
- error recovery
- a chain of bad fixes is not fixable anymore - just discard and try again
- agents try alternatives because the explicitly requested tool call failed with crucial ingredients
- tell them to ask for these ingredients before trying alternatives
- baseline agent + logs + LLM analysis = insights for improvments
- Q&A: Combining Math and LLMs
- Raw math/algos and LLM are both powerful tools in your modeling toolkit
- If you’re missing either tool from your toolkit, then that will severely hamstring your modeling abilities and many problems will remain inaccessible to you.
- Show HN: Companies use AI to take your calls. I built AI to make them for you
- top comment on HN introduced a great use case for LLM-back reception
- local plumbing business
- understand and repeated back the client's request in a slightly different, more professional way for confirmation
- asked a few more smart follow-ups
- the plumber called back and jumped straight into solutions, pricing, and his availability
- prefer to talk to LLM, if the issue can be quickly triaged to the right human who actually understand the situation
- synchronous interaction by bot
- can perform first level troubleshooting, ask for clarification, begin to form a plan and get your buy-in
- v.s. fire-and-forget email form
- incomplete reports, missing information, people who have no idea what they're talking about
- Show HN: Terminal-Bench-RL: Training Long-Horizon Terminal Agents with RL (on HN)
- Custom scaffolding (system prompt and tools) using Qwen3-32B achieved 13.75- The author has built an RL system, but it has not been used for anything due to cost limitations.
- Stop selling “unlimited”, when you mean “until we change our minds”
- the playbook
- launch with generous/unlimited limits
- build user dependency
- add caps targeting "less than 5- frame as "sustainability" or "fairness"
- that "5- power users with deep workflow integration
- early adopters who took platform risks
- team influencers who drive organizational adoption
- $200/month Claude Max subscribers doing serious work
- Tokens are getting more expensive
- Claude Code weekly rate limits
- The Rise of Vibeinsecurity
- a story about how devs lost their jobs over vibe apps, and vibe apps are hacked by vibe hackers immediately after deployment
- this is also an ad for a conference called HackAIcon
- It's rude to show AI output to people
- whenever you propagate AI output, you're at risk of intentionally or unintentionally legitimizing it with your good name, providing it with a fake proof-of-thought
- Agentic Coding Things That Didn’t Work
- #sec #privacy
- Tea app leak worsens with second database exposing user chats
- #idea
- Why Try When Others Could Do Better?
- Even the smartest, highest-agency people in the world are severely bandwidth constrained and don’t get around to doing most of the things they have the potential to do.
- Compound that over and over again in some niche and you get hundreds, thousands, millions of miles ahead of the fastest runners who aren’t running down that niche.
- Pragmatism in Programming Proverbs
- many interesting quotes
- Objects should shut up (on HN)
- I'm also deeply annoyed by unnecessary alarms
- Don’t Let Architecture Astronauts Scare You #system
- I'm rebelling against the algorithm
- Maybe writing speed actually is a bottleneck for programming
- #rust #metaprogramming
- Advanced Rust macros with derive-deftly
- rules_derive: deriving using macro_rules