Hacker News: fabmilo

New comment by fabmilo in "Scaling Karpathy's Autoresearch: What Happens When the Agent Gets a GPU Cluster"

fabmilo — Thu, 19 Mar 2026 19:01:10 +0000

I am fascinated by this example of using AI to improve AI. I won a small prize using this technique on helion kernels at a pytorch hackathon in SF.

The next step are: - give the agent the whole deep learning literature research and do tree search over the various ideas that have been proposed in the past. - have some distributed notepad that any of these agents can read and improve upon.

New comment by fabmilo in "Show HN: Hacker News archive (47M+ items, 11.6GB) as Parquet, updated every 5m"

fabmilo — Wed, 18 Mar 2026 19:47:52 +0000

Was thinking the same thing. probably once a day would be more than enough. if you really want a minute by minute probably a delta file from the previous day should be more than enough.

New comment by fabmilo in "Show HN: Better Hub – A better GitHub experience"

fabmilo — Thu, 26 Feb 2026 11:27:02 +0000

indeed. make a loom showing us why is better.

New comment by fabmilo in "How I've run major projects (2025)"

fabmilo — Mon, 09 Feb 2026 23:53:33 +0000

There is tons of good advice. This blog post can be easily turned into a skill for agents.

New comment by fabmilo in "Generative Modeling via Drifting"

fabmilo — Fri, 06 Feb 2026 23:41:41 +0000

New generative modeling using a single inference step

Generative Modeling via Drifting

fabmilo — Fri, 06 Feb 2026 23:41:41 +0000

Article URL: https://arxiv.org/abs/2602.04770

Comments URL: https://news.ycombinator.com/item?id=46919708

Points: 2

# Comments: 1

New comment by fabmilo in "The Waymo World Model"

fabmilo — Fri, 06 Feb 2026 19:33:26 +0000

Very impressive work from Waymo. The driving with a tornado in the horizon example kind of struck my imagination, many people actually panic in such scenarios. I wonder though the compute requirements to run these simulations and producing so many data points.

New comment by fabmilo in "Flux 2 Klein pure C inference"

fabmilo — Sun, 18 Jan 2026 22:20:36 +0000

because of the principle: you only understand what you can create. You think you know something until you have to re-create it from scratch.

New comment by fabmilo in "Ask HN: What Are You Working On? (Nov 2025)"

fabmilo — Mon, 10 Nov 2025 04:57:42 +0000

VAE for real time video generation, WAN 2.1 / Matrix Game 2.0

New comment by fabmilo in "Coral NPU: A full-stack platform for Edge AI"

fabmilo — Sun, 19 Oct 2025 06:39:58 +0000

How much would cost to produce these ?

New comment by fabmilo in "Show HN: FleetCode – Open-source UI for running multiple coding agents"

fabmilo — Wed, 08 Oct 2025 20:07:41 +0000

nice, didn't knew this tool either

New comment by fabmilo in "Claude Sonnet 4.5"

fabmilo — Mon, 29 Sep 2025 21:47:39 +0000

Yeah I totally agree, we need time to completion of each step and the number of steps, sizes of prompts, number of tools, ... and better visualization of each run and break down based on the difficulty of the task

New comment by fabmilo in "Ask HN: What are you working on? (September 2025)"

fabmilo — Mon, 29 Sep 2025 21:43:18 +0000

How does it work? is just a documentation specification like spec kit?

New comment by fabmilo in "Show HN: I built an MCP server using Cloudflare's code mode pattern"

fabmilo — Sun, 28 Sep 2025 18:08:24 +0000

I was just reflecting on this blog post after reading it this morning. What do you think on code mode after implementing it? At this point would not be better to just have a sandboxed api environment with customizable api/tools endpoints? basically an RL environment :)

New comment by fabmilo in "The AI coding trap"

fabmilo — Sun, 28 Sep 2025 16:04:34 +0000

one axis that is missing from the discussion is how fast they are improving. We need ~35 years to get a senior software engineer (from birth to education to experience). These things are not even 3.5 years old. I am very interested in this space, if you are too dm me on X:@fabmilo I am in SF.

New comment by fabmilo in "AI for Scientific Search"

fabmilo — Thu, 03 Jul 2025 19:49:40 +0000

I like zotero, I started vibe coding some integration for my workflow, the project is a bit clunky to build and iterate the development specially with gemini & claude. But I think that is the direction to take instead of reinvent from scratch something

New comment by fabmilo in "Show HN: Defuddle, an HTML-to-Markdown alternative to Readability"

fabmilo — Fri, 23 May 2025 01:16:06 +0000

reference to the library: https://trafilatura.readthedocs.io/en/latest/

for the curious: Trafilatura means "extrusion" in Italian.

| This method creates a porous surface that distinguishes pasta trafilata for its extraordinary way of holding the sauce. search maccheroni trafilati vs maccheroni lisci :)

(btw I think you meant trafilatura not trifatura)

New comment by fabmilo in "A Research Preview of Codex"

fabmilo — Fri, 16 May 2025 16:38:49 +0000

more excited about the rust impl than the typescript one.

New comment by fabmilo in "Intellect-2 Release: The First 32B Model Trained Through Globally Distributed RL"

fabmilo — Mon, 12 May 2025 03:33:12 +0000

The interesting delta here is that this proves that we can distribute the training and get a functioning model. The scaling factor is way bigger than datacenters

New comment by fabmilo in "Multi-Token Attention"

fabmilo — Thu, 03 Apr 2025 02:27:42 +0000

I read the paper and the results don't really convince me that is the case. But the problem still remains of being able to use information from different part of the model without squishing it to a single value with the softmax.