Robert Bach

PhD candidate at UCF | Quantitative Finance

[Robert Bach @robertbach_]

@draftandfocus Hardly. I only make new notes when I really have to.

[Robert Bach @robertbach_]

I handed my Obsidian vault to three coding agents: Claude Code, Codex, and Hermes Agent on a local model.

They capture sources, keep the index, answer from my notes, and lint the vault. I approve every note that becomes permanent.

Open-sourced the setup. Thread:

[Robert Bach @robertbach_]

@RubeBuilds I think Claude for now 🤔

[Robert Bach @robertbach_]

@harleyfoote_ Maybe a systematic index script could be better 🤔

[Robert Bach @robertbach_]

@harleyfoote_ Thank you for the suggestions however I believe the skill is quite robust against hallucinations. After ai create my notes, the agents won't touch my index until you run /wiki skill. It only read the frontmatter so there is no chance it make stuff up for the index.

[Robert Bach @robertbach_]

@thanghvu Thanks brother!

[Robert Bach @robertbach_]

@bjcatar What model you use to power Hermes? I am using Qwen 3.8 27B. It's good for simple tasks but I find for heavy task like /lint and /wiki, the quality is drop significantly

[Robert Bach @robertbach_]

@Shadowfetch Thank you for the kind words. I hope it makes sense for people with the set Zettelkasten setup

[Robert Bach @robertbach_]

@thanghvu Open source my External Brain setup!

[Robert Bach @robertbach_]

6. One command, and it never overwrites your files:

curl -fsSL https://t.co/wubc94Y9M7 | bash -s -- ~/external-brain

Repo: https://t.co/mcCEWy5sNU
Full write-up: https://t.co/LKv4sUbWwB

[Robert Bach @robertbach_]

5. A frontier model and a 27B local model share the vault: a mkdir write lock, git rules for the second machine, and a smaller role for the small model.

Skills are written in "80% ASD-STE100", another Karpathy tip. Short imperative steps work far better on the small model.

[Waseem @waseemhnyc]

this is wild... all open source? are you kidding me

[Sophie Wang @SophieLWang]

I wrote a blog post detailing the writing approach I took while working on my recent paper: https://t.co/0aUhDLOnYD

I describe the behind the scenes for how this paper came to be, why and how I storyboarded figures before running experiments, and some figure design tips! :)

[Robert Bach @robertbach_]

@s6juncheng @grok this looks fishy, is this malicious?

[Peter Chen @PeterLauLukCh]

Accepted to #NeurIPS2026, we reshaped Monte-Carlo Tree Search 🙉

Introducing 2FFS, a new tree search algorithm that combines multi-fidelity bandits to resolve the fundamental trade-off: should we use cheaper, approximated evaluations, or expansive but accurate samplings?

More: