Hacker News: mmargenot

New comment by mmargenot in "Everyone is building LLM routers, we deprecated ours"

mmargenot — Fri, 31 Jul 2026 19:42:17 +0000

I find this routing problem to be opaque and I’m generally skeptical that the label people are trying to predict is meaningful.

If you really need more discrimination of the complexity of an input to get an efficient response, sft or rl tuning something for your harness would be more effective.

New comment by mmargenot in "Google Books (or similar) all book scans – $200k bounty (2025)"

mmargenot — Sat, 04 Jul 2026 20:44:38 +0000

How do you handle the more densely written pages in script ? I did a very similar exercise OCRing works from this exact collection, but I stuck with the English books for the first pass.

New comment by mmargenot in "Website streamed live directly from a model"

mmargenot — Thu, 23 Apr 2026 12:22:40 +0000

I love this approach, very Diamond Age. I uploaded a picture of my cat and learned a lot about coat genetics. I think it fixated on his coloration.

New comment by mmargenot in "Stanford report highlights growing disconnect between AI insiders and everyone"

mmargenot — Mon, 13 Apr 2026 21:46:45 +0000

AI applications that would help normal people in a significant way are pretty lacking, so I'm not surprised. So much conversation about AI products is cycles of "this tech will change everything" without material backup outside of coding agents.

New comment by mmargenot in "Components of a Coding Agent"

mmargenot — Sat, 04 Apr 2026 20:36:11 +0000

Here’s an explicit example of the above from today using the above dataset: https://x.com/kaiostephens/status/2040396678176362540?s=46

New comment by mmargenot in "Show HN: sllm – Split a GPU node with other developers, unlimited tokens"

mmargenot — Sat, 04 Apr 2026 16:55:35 +0000

This is a great idea! I saw a similar (inverse) idea the other day for pooling compute (https://github.com/michaelneale/mesh-llm). What are you doing for compute in the backend? Are you locked into a cohort from month to month?

New comment by mmargenot in "Components of a Coding Agent"

mmargenot — Sat, 04 Apr 2026 16:47:39 +0000

It is more common now to improve models in agentic systems "in the loop" with reinforcement learning. Anthropic is [very likely] doing this in the backend to systematically improve the performance of their models specifically with their tools. I've done this with Goose at Block with more classic post-training approaches because it was before RL really hit the mainstream as an approach for this.

If you want to look at some of the tooling and process for this, check out verifiers (https://github.com/PrimeIntellect-ai/verifiers), hermes (https://github.com/nousresearch/hermes-agent) and accompanying trace datasets (https://huggingface.co/datasets/kai-os/carnice-glm5-hermes-t...), and other open source tools and harnesses.

New comment by mmargenot in "Show HN: I built a frontpage for personal blogs"

mmargenot — Fri, 03 Apr 2026 13:26:47 +0000

Very cool! This was a good impetus to actually add RSS to my blog.

New comment by mmargenot in "I'm OK being left behind, thanks"

mmargenot — Fri, 20 Mar 2026 18:40:50 +0000

Especially when people pushing it are trying to capture your attention, it’s good to be deliberate about the tech that you introduce.

New comment by mmargenot in "Apple is fighting for TSMC capacity as Nvidia takes center stage"

mmargenot — Thu, 15 Jan 2026 23:42:23 +0000

Do you have a source for this? Most information I’ve seen around this (e.g. Acquired podcast, from the Costco side) claims strong positive relationships.

New comment by mmargenot in "Ask HN: How are you doing RAG locally?"

mmargenot — Thu, 15 Jan 2026 17:34:47 +0000

Cursor uses a vector index, some details here: https://cursor.com/docs/context/semantic-search

New comment by mmargenot in "Ask HN: How are you doing RAG locally?"

mmargenot — Thu, 15 Jan 2026 17:32:17 +0000

I made an obsidian extension that does semantic and hybrid (RRF with FTS) search with local models. I have done some knowledge graph and ontology experimentation around this, but nothing that I’d like to include yet.

This is specifically a “remembrance agent”, so it surfaces related atoms to what you’re writing rather than doing anything generative.

Extension: https://github.com/mmargenot/tezcat

Also available in community plugins.

New comment by mmargenot in "Ridley Scott's Prometheus and Alien: Covenant – Contemporary Horror of AI (2020)"

mmargenot — Sun, 12 Oct 2025 21:50:41 +0000

I had the good fortune of seeing Lawrence of Arabia in 70mm in a theater and then going to watch Prometheus within the same two week span. It gave me a much greater appreciation for the movie [Prometheus], and what it was trying to do.

New comment by mmargenot in "Mathematicians discover prime number pattern in fractal chaos"

mmargenot — Wed, 08 Oct 2025 14:03:33 +0000

The patterns associated with primes are inherent to the numbers themselves and not their representations. The numbers are the pattern.

New comment by mmargenot in "Devpush – Open-source and self-hostable alternative to Vercel, Render, Netlify"

mmargenot — Tue, 07 Oct 2025 13:53:55 +0000

I like the idea of self-hostability, but not having to think about the deployment of the frontend piece has been a huge accelerant for me, someone who typically thinks only of ML and backend components.

New comment by mmargenot in "Will I run Boston 2026?"

mmargenot — Tue, 16 Sep 2025 20:21:28 +0000

You can also run for a charity by raising sufficient money for a cause that sponsors you.

New comment by mmargenot in "Show HN: AI-powered web service combining FastAPI, Pydantic-AI, and MCP servers"

mmargenot — Mon, 15 Sep 2025 22:31:26 +0000

`outlines` (https://github.com/dottxt-ai/outlines) is very good and supported by vLLM as a backend structured output provider (https://docs.vllm.ai/en/v0.8.2/features/structured_outputs.h...) for both local and remote LLMs. vLLM is probably the best open source tooling for the inference side right now.

New comment by mmargenot in "We can’t circumvent the work needed to train our minds"

mmargenot — Wed, 10 Sep 2025 17:18:17 +0000

Executing on meaningful knowledge work also might require many different paths, depending on the context and the environment. To me it's more about the method of inquiry and how you begin than it is the specific content. Sure, more individual facts help to guide that inquiry, but at any given moment you're only truly going to be able to recall a subset of those.

New comment by mmargenot in "We can’t circumvent the work needed to train our minds"

mmargenot — Wed, 10 Sep 2025 17:14:01 +0000

It's weird to read this from zettelkasten.de, given that the method is precisely about cultivating such a graph of knowledge. "Knowing enough to begin" seems to me to be the express purpose of writing and maintaining a zettelkasten and other such tools.

New comment by mmargenot in "We can’t circumvent the work needed to train our minds"

mmargenot — Wed, 10 Sep 2025 16:01:13 +0000

> You have to remember EVERYTHING. Only then you can perform the cognitive tasks necessary to perform meaningful knowledge work.

You don't have to remember everything. You have to remember enough entry points and the shape of what follows, trained through experience and going through the process of thinking and writing, to reason your way through meaningful knowledge work.