Hacker News: aabdi

New comment by aabdi in "Nvidia Cosmos 3"

aabdi — Mon, 01 Jun 2026 17:06:48 +0000

This is mostly a decompression, it’s fairly standard nowadays. The point is to get the data from the internal compressed version into the human usable version.

We can technically reason at pixel or char level encodings but it’s going to be much more expensive generally. Think of the overall technique as a way to get computer go faster.

You see it with Qwen talker, most multimodal projectors, etc

New comment by aabdi in "Nvidia Cosmos 3"

aabdi — Mon, 01 Jun 2026 14:05:11 +0000

SOTA open source model for image and vid generation. Beats all others but is too big to run on most people’s computers at 64b params.

Still impressive nonetheless given its artificially generated training sets.

Beats nano banana 1 but not yet competitive with 2 or seedance2, grok imagine,etc.

New comment by aabdi in "Why are large language models so terrible at video games?"

aabdi — Mon, 01 Jun 2026 11:53:53 +0000

Isn’t this more of a “we didn’t rl the model to do games so it can’t do it?”

Something like snake or tic-tac-toe is straightforward.

New comment by aabdi in "Dynamic Workflows in Claude Code"

aabdi — Thu, 28 May 2026 18:10:01 +0000

wrote something similar for my own use/work stuff; seems everything is converging towards similar ideas.

IMO, this style of workflow/agentics is how all SWE'll look like long term. Automate everything into a big pipe-y thing. How it's gonna be modelled is up in the air though. lots of different approaches:

mine: https://github.com/portpowered/you-agent-factory

https://github.com/ComposioHQ/agent-orchestrator

https://github.com/gastownhall/gastown

https://github.com/openai/symphony

Show HN: Infinite you – multi agent workflow system

aabdi — Thu, 07 May 2026 16:21:04 +0000

Seems lots of folks are building workflow orchestrators these days. This is mine~. Would like your guys feedback!

Mostly supports codex, customizable graph orchestration so it supports all weird flows you can think of (ralph loops, spawn worktrees, write review). Bootstraps itself nowadays (i.e. the last ~900 commits).

supports script and agent based nodes in the graph. i.e. you can paply post condition checks via script nodes or reviewers.

Completion is marked when a node in the graph moves a token from processing to failed/complete.

Comments URL: https://news.ycombinator.com/item?id=48051265

Points: 2

# Comments: 0

New comment by aabdi in "Show HN: I built a sub-500ms latency voice agent from scratch"

aabdi — Tue, 03 Mar 2026 08:38:15 +0000

it was too hard~, they all tried real hard and the models just kept failing. The models only got good enough -1.5 years ago~.

I mean its deployed now (Alexa+/gemini). but its expensive as hell. and also kinda useless. Claude cowork/clawbot form factors are better.

Wrong form factor/use case really. People really wanna buy stuff using clawbot.

New comment by aabdi in "Steve Jobs' cabinet"

aabdi — Fri, 25 Jul 2025 21:07:26 +0000

is this the right way to think about LLMs?

For example, the syscalls for linux are never changing IMO. The cost is unbounded to change, even with AI.

Should your APIs be treated any different?

At the very least, the LLMs work better with better APIs and data models, which yet accelerates the solving of problems.

New comment by aabdi in "“Normal” engineers are the key to great teams"

aabdi — Thu, 13 Mar 2025 21:35:10 +0000

it's sorta like, why doesn't everyone just kill themselves you know?

sometimes, you just find fun in things and it's cool. other times, it's like what other other thing you gonna do? fish or hang with people or do drugs or dance? software's a hobby really. sometimes its more fun.

but really it's all preference.