Hacker News: pertymcpert

New comment by pertymcpert in "We replaced RAG with a virtual filesystem for our AI documentation assistant"

pertymcpert — Wed, 08 Apr 2026 16:46:24 +0000

Understood. I’m willing to defy guidelines and take the consequences. I still think it’s worth pointing out slop so people don’t waste their time talking to a machine.

New comment by pertymcpert in "We replaced RAG with a virtual filesystem for our AI documentation assistant"

pertymcpert — Wed, 08 Apr 2026 16:45:18 +0000

Each to their own. I appreciate you writing that comment yourself.

New comment by pertymcpert in "System Card: Claude Mythos Preview [pdf]"

pertymcpert — Wed, 08 Apr 2026 16:43:47 +0000

I’m beyond trying to convince people to take this technology seriously. You’ll learn for yourself.

New comment by pertymcpert in "System Card: Claude Mythos Preview [pdf]"

pertymcpert — Wed, 08 Apr 2026 02:41:10 +0000

If anything I’m seeing too much skepticism and not enough alarm. People burying their heads in the sand, fingers in their ears denying where this is all going. Unbelievable except it’s exactly what I expect from humans.

New comment by pertymcpert in "Project Glasswing: Securing critical software for the AI era"

pertymcpert — Wed, 08 Apr 2026 02:34:03 +0000

What evidence makes you say that? Do you have insider info?

New comment by pertymcpert in "Project Glasswing: Securing critical software for the AI era"

pertymcpert — Wed, 08 Apr 2026 02:31:14 +0000

This isn’t talking about compaction. This refers to performance as the model is loaded with 500k to 1m tokens.

New comment by pertymcpert in "Project Glasswing: Securing critical software for the AI era"

pertymcpert — Wed, 08 Apr 2026 02:29:34 +0000

Did you read the article?

New comment by pertymcpert in "We replaced RAG with a virtual filesystem for our AI documentation assistant"

pertymcpert — Sat, 04 Apr 2026 07:22:18 +0000

They do care about files. They also care about how you express yourself, your tone, all sorts of seemingly unimportant details.

New comment by pertymcpert in "Google releases Gemma 4 open models"

pertymcpert — Fri, 03 Apr 2026 05:26:28 +0000

I’d be very interested in trying it if you could spare the time to write up how to tune it well. If not thanks for the input anyway.

New comment by pertymcpert in "Google releases Gemma 4 open models"

pertymcpert — Fri, 03 Apr 2026 05:14:28 +0000

That’s how the model is trained to signal the end to its generation and to indicate its thinking.

New comment by pertymcpert in "$30B for laptops yielded a generation less cognitively capable than parents"

pertymcpert — Mon, 23 Feb 2026 06:07:55 +0000

There is a large body of research that shows it's not what you're saying it is FYI.

New comment by pertymcpert in "$30B for laptops yielded a generation less cognitively capable than parents"

pertymcpert — Mon, 23 Feb 2026 06:07:03 +0000

Did you read the article?

New comment by pertymcpert in "Aqua: A CLI message tool for AI agents"

pertymcpert — Mon, 23 Feb 2026 06:04:46 +0000

You need help.

New comment by pertymcpert in "Are compilers deterministic?"

pertymcpert — Sun, 22 Feb 2026 02:16:40 +0000

GCC and LLVM consider it a bug if the compiler is non-deterministic. If re-running the compiler generates different output because of things like address differences for example then it's something that needs to be fixed. So yes they are deterministic.

New comment by pertymcpert in "AVX2 is slower than SSE2-4.x under Windows ARM emulation"

pertymcpert — Wed, 18 Feb 2026 19:03:49 +0000

Those are extremely performance critical operations. A lot of people use their phone many times an hour.

New comment by pertymcpert in "AVX2 is slower than SSE2-4.x under Windows ARM emulation"

pertymcpert — Wed, 18 Feb 2026 19:02:43 +0000

You can definitely SVE vectors on the stack, there are special instructions to load and store with variable offsets. What you can't do is to put them into structs which need to have concretely sized types (i.e. subsequent element offset need to have a known byte offset).

New comment by pertymcpert in "Rivian R2: Electric Mid-Size SUV"

pertymcpert — Thu, 12 Feb 2026 23:58:51 +0000

My mistake, thanks.

New comment by pertymcpert in "GPT‑5.3‑Codex‑Spark"

pertymcpert — Thu, 12 Feb 2026 20:30:21 +0000

MCP is fine if your tool definition is small. If it's something like a sub-agent harness which is used very often, then in fact it's probably more context efficient because the tools are already loaded in context and the model doesn't have to spend a few turns deciding to load the skill, thinking about it and then invoking another tool/script to invoke the subagent.

New comment by pertymcpert in "GPT‑5.3‑Codex‑Spark"

pertymcpert — Thu, 12 Feb 2026 20:28:41 +0000

They're talking about "skills" which are not the same thing as tools. Most models haven't been trained on the open SKILL spec, and therefore aren't tuned to invoke them reliable when the need occurs.

New comment by pertymcpert in "GPT‑5.3‑Codex‑Spark"

pertymcpert — Thu, 12 Feb 2026 20:26:40 +0000

You can just do dynamic binning.