HNHacker News
TopNewBestAskShowJobs

nijave

2,500 karma · joined August 22, 2018

[ my public key: https://keybase.io/nickv; my proof: https://keybase.io/nickv/sigs/qMdHcggaKRYKr002ExXFme5MjZeOQbHt-PyAl_07HDk ]
submissionscomments
nijave··on google.com/goto: Google's anti-scraping update
Hmm that's actually pretty clever. They can serve each result page slightly different encrypted links and it should be obvious right away if it's a SERP bot (trying to grab a page of links) or a human that just picks a few here or there.

I wonder if this would also work on other sites getting hammered with bots. Allow each anonymous user 1 "real" page load then turn the rest into encrypted links that the web server can decrypt. If a session cookie with reputation exists, stop screwing with the links.

Kind of annoying but it'd allow tracking if the same agent/bot is churning through IPs/User Agents.

nijave··on Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
It's (one of) the best freely available?
nijave··on Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
This explains a lot about Sonnet 5.
nijave··on Automattic's board forces CEO Matt Mullenweg into leave of absence
He tried to extort them. Hosting providers have been selling Wordpress hosting for years without "giving back"

Does Wordpress "give back" to PHP and Linux?

nijave··on Asahi Linux on M3
Eh, they probably have enough server hardware/infrastructure and potentially custom accelerators and other hardware that staffing a team makes sense

Quick web search says they at least need some kernel developers for virtualized Linux on macOS

nijave··on Keep Our Servers Running
Are there on staff lawyers hiding in payroll or do they contract everything?
nijave··on Asahi Linux on M3
Which doesn't even boot on modern Apple hardware without all the closed source bits...
nijave··on Asahi Linux on M3
They could open source their software so others could utilise the code instead of needing reverse engineering
nijave··on Show HN: Kadō – open-source habit tracker, with non-binary habit score, for iOS
Anything similar for Android (or anyone want to burn some tokens cloning)?

I see the readme mentions it's inspired by Loop. Ideally something that also allows tracking not-necessarily-habit daily things (drank coffee at time y)

nijave··on Which tools do Claude, Codex and Cursor choose? We measured 17k runs to find out
Azure database??? In house bot protection??? In house search???

Some of these are absolutely wild. Surprised Strands didn't even get mentioned for agent frameworks.

nijave··on Which tools do Claude, Codex and Cursor choose? We measured 17k runs to find out
Actually, yes. Probably this https://github.com/anthropics/claude-code/issues/88041#issue...
nijave··on Reverse Engineering Unknown File Formats with ImHex
LLM and Google are suggesting it might be this https://learn.microsoft.com/en-us/openspecs/exchange_server_...

A lot of Microsoft formats are open standards... Created by and only used by Microsoft...

nijave··on Reverse Engineering Unknown File Formats with ImHex
A lot of older stuff just dumped structs in general
nijave··on Reverse Engineering Unknown File Formats with ImHex
Claude.ai, set to Opus xhigh, attach file, "Try to decode the file format of this save game from Name of Game. Do research if necessary"

If the file's too big, you'll probably need to download a harness/local app. If you go local route, make sure you have a backup in case the LLM breaks the file

The local route it can also try to poke at the binaries that read/write the file so it would have a bit more to work with going that route

nijave··on Whistleblower warns Postal Service mail ballot system has catastrophic problems
I think it's roughly the same in Ohio
nijave··on Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x
I use routatic/proxy (actually my own fork with various changes/fixes http://github.com/nijave/proxy/ -- fyi the fork is all vibes...)

Some gateways/proxies/hosts also offer multiple endpoints (z.ai has both Anthropic endpoint and Completions endpoint)

Edit: Here's the PoC I made for work although it assumes you're using Cloudflare Access (VPN) with a Cloudflare AI Gateway to hand wave the auth away (auth to the LLM API gets auto injected by Cloudflare Access derived from the client's VPN connection) https://gist.github.com/nijave/6f5e5793a992619808135ce85fa49...

nijave··on Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x
It would be interesting to try to control for system prompt although obviously there's some coupling to what the harness provides and the instructions the model receives on using it
nijave··on Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x
Agree with the exception of Haiku. Haiku is insanely fast if it works for your task
nijave··on Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x
>nobody is going to be doing 17 benchmark passes on a $50-100 task

Hopefully we'll see more of this as big companies try to optimize token usage where the cost of benchmarking is dwarfed by the potential savings across the org

nijave··on Show HN: FrontierHarness Eval – 9 harness, same model, cost per pass varies 17x
Afaik the reverse is also true. Some models are trained for specific harness setups so work better/more efficiently in that harness. If you give a model trained on a Read/Write/Bash tool setup a harness with only Bash where it needs to generate shell commands, I'd expect it to perform differently.

Related, there was a small debacle when Claude Code yeeted out a prompt change telling Opus to use Bash for everything. It was speculated it was an optimization to try to make it more efficient by encouraging it to chain a bunch of "tool use" commands into a giant shell command. However, they didn't communicate it clearly and it broke a bunch of setups that expected/hooked tool calls

There's also the (anecdotal?) tradeoff that dumb models tend to do better with smart tools and smart models tend to do better with dumb tools (smart tools -> easier to use but require more context usage for specs, dumb tools -> harder to use but more versatile and save context)

nijave··on We could save petabytes of cache storage with Zstandard and Pingora
That part is a little bit confusing.

I think they probably don't care about storage on the devices that do the compressing and are optimizing for quickly pushing hot content to edge locations. So the compression at the source saves bandwidth during the pushing to edge phase and allows the edges to hold more (reducing churn, further saving bandwidth back to the source).

Put a different way, they're trying to make cache evictions cheaper (less bandwidth to refill) and less likely (bigger cache on same disk size)

"Ignore cold assets" makes more sense with that framing

Although if that's the case, the CPU statement still is a bit confusing.

nijave··on We could save petabytes of cache storage with Zstandard and Pingora
Curious how this application scheme compares to filesystem + transport compression. You'd end up potentially compressing and decompressing more often but the higher software doesn't need to know what's happening and the compression happens in kernel space.

ie btrfs

You could also layer on out of band dedupe and probably push out cache updates with btrfs snapshots although maybe that ends too convoluted

nijave··on We could save petabytes of cache storage with Zstandard and Pingora
Idk but btrfs and zfs manage to pull it off

Seekable OCI (SOCI) uses an index so I imagine that's an option (real byte range a-b maps to compressed range x-y). Presumably you'd still need to read the header and some additional pieces

nijave··on AI Agents and the Refactoring That Never Happens
Vendor lock in
nijave··on AI Agents and the Refactoring That Never Happens
Yeah, you gotta make em prove their work so they don't hallucinate.

Try "build a script to trace callers" and "cite each file and line that calls the function"

Won't fix everything but usually helps in my experience

Although you still run into hard-to-find things especially in dynamic/duck typed languages or codebases with heavy use of reflection or code bases used as libraries

nijave··on AI Agents and the Refactoring That Never Happens
>A computer can hold far more in “working memory” than a human can

That's a bit of a straw man. State of the art agents are limited to ~3.8Mi (1M tokens). That's usually where I run into issues--an LLM can't possibly hold as much context as a human and it's more of an art than a science getting the most important things squeezed in. It's especially prudent for complex codebases/systems.

An agent only knows what it can see. It doesn't know oldCruftyFunction is still critical to Bob's Excel macro that generates financial reports and yanks the codebase in as a bastardised dependency. A lot of times agents give a fall sense of security by making it seem like something complicated and unsafe is actually safe.

nijave··on AI Agents and the Refactoring That Never Happens
Gotta get IT to slip superpowers or at least the TDD skill into MDM/enterprise controls
nijave··on AI Agents and the Refactoring That Never Happens
It's been fun reviving some abandoned personal/side projects and seeing them through. I have a bunch of random partially completed code that I never got around to finishing and it's fun seeing it work even if I have no use for it. Tons of "proof of concepts" left at "concept" that never reached "proof" ha
nijave··on Claude Session URL appended to commit messages and PR descriptions by default
I made a skill which is maybe why GPT is doing okay but Claude seems to straight up ignore it (it invokes/reads it but doesn't follow the directions)
nijave··on Claude Session URL appended to commit messages and PR descriptions by default
Eh, I don't want the tools I use to stamp their names on everything. It's also not uncommon for me to switch between GPT and Claude or different models in the harness if one of them isn't doing well. GPT-5.6 Terra has been doing a significantly better job at writing concise PR bodies and commit messages than Claude's usual multi paragraph word soup.

I don't need "Created by Claude Code 1.2.3, Opus 4.8 inside tmux inside ghostty on macOS ## on Mac Book Pro 2019 by git 123 in /Users/nijave/repo"

If you want session context, commit the session or prompt or add a git note. I think opaque URIs are the wrong approach unless they're to an issue tracker with more details. Proprietary URIs are not easily usable and interoperable with other tools. Commited information should still be valid and useful in 10 years.

I'm also not sure session is the right level of abstraction. I recently setup Claude Code with subscription delegation to opencode harness with z.ai coding plans and opencode go coding plan and it's been working fairly but there is no singular "session"

← PreviousPage 5 of 34Next →