frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: I built a web tool to see and edit what an AI thinks before it answers

https://lucid.earthpilot.ai
16•ada1981•5h ago
I run a small AI lab and playground and got super excited about Anthropics paper "Verbalizable Representations Form a Global Workspace in Language Models" (https://transformer-circuits.pub/2026/workspace/index.html)

It talks about how they use a tool they call a Jacobian Lens to view inside the middle layers of LLM while it's working before it commits to a word (token).

I wanted to see if I could get a version of this running on the open models and to my surprise it worked! I ran some experiments with it and build a public facing free tool anyone can use with your own prompts.

Ask the model to describe a symbol of "three curving lines of water" and you can watch "ocean", "sea", and "surf" light up a few layers deeper before it settles on "waves".

You can also edit the internal state. Insert "fire" into the middle layer of the ocean prompt and the answer shifts to something about heat.

For fun / curiosity sake, I also developed way to let the model read its own inner workspace and then decide to suppress or amplify a concept, and run the prompt again.

Interesting finding from running it across models. J-lens beats a plain logit lens on some architectures and does nothing on others, and it isn't about size. A 0.5B Qwen reads better than a 2.8B Pythia. Every Pythia I tried gained basically nothing; the Llama and Qwen models gained a lot. https://lucid.earthpilot.ai/research

This is a 48 hour old project based on emerging research and built on a small model, a small probe set on rented GPUs - but I found it genuinely exciting. The code is open.

I also included a page context "Docent" AI agent you can chat with about whatever you see to help understand what is going on.

Happy to have folks poke around and break it.

I imagine the applications for allowing models to self-reflect / edit internal states can be useful for alignment, confidence, bias detection, etc. and this tool lets you play with the early stages of that.

Comments

ada1981•47m ago
also from our lab is Personality Bench: https://persona.earthpilot.ai/ which runs every frontier model through personality tests.
krackers•30m ago
>J-lens beats a plain logit lens on some architectures and does nothing on others, and it isn't about size

The paper talked about this, the jacobian matrix corrects for the shift in basis from initial to final layer compared to logit lens which assumes that the residual remains in the same basis across layers. Maybe the latter is in fact true for some models/architectures so the J-lens doesn't do anything extra?

Show HN: Getting GLM 5.2 running on my slow computer

https://github.com/JustVugg/colibri
190•vforno•14h ago•42 comments

Show HN: 18 Words

https://18words.com/
755•pompomsheep•9h ago•273 comments

Show HN: Rubiks Cube Solver

https://speedcube.com.br/
8•wozzp•1h ago•0 comments

Show HN: I mapped 8.5M research papers into an interactive atlas

https://tomesphere.com/atlas
54•leonickson•20h ago•21 comments

Show HN: I built a web tool to see and edit what an AI thinks before it answers

https://lucid.earthpilot.ai
16•ada1981•5h ago•2 comments

Show HN: RandoFont – A browser for Google Fonts

https://randofont.alesh.com
14•aleshh•1h ago•2 comments

Show HN: Analog Watch

https://analog.watch
85•ezekg•8h ago•79 comments

Show HN: Use Any Agent as an Orchestrator

https://github.com/backnotprop/orchestrator
2•ramoz•15m ago•0 comments

Show HN: Sighthound - open-source vulnerability scanner for source code

https://github.com/Corgea/Sighthound
10•asadeddin•7h ago•0 comments

Show HN: Reverse-engineering web apps into agent tools

22•pancomplex•6h ago•4 comments

Show HN: Reviving my 2001 college band with AI

https://www.fadingmaize.com
7•jacobgraf•2h ago•1 comments

Show HN: Frugon – Find which LLM calls a cheaper model could handle (local, MIT)

https://github.com/Rodiun/frugon
27•jarodrh•2d ago•8 comments

Show HN: Abralo – Free, easy way to run several Claude Code agents in one window

https://abralo.com/
17•cwbuilds•1d ago•12 comments

Show HN: Ilya's 30 papers, explained in audio

https://listendock.com/30-papers
2•janpmz•31m ago•0 comments

Show HN: Pylon Sync, an agent-first full-stack realtime framework

https://www.pylonsync.com
7•ericc59•5h ago•0 comments

Show HN: Interactive SQLite3 Query Inspection

https://github.com/yeet-src/sqlitefeed
3•zasc•49m ago•0 comments

Show HN: FableCut – A browser video editor AI agents can drive (zero deps)

https://github.com/ronak-create/FableCut
82•ronak_parmar•9h ago•54 comments

Show HN: Policy enforcement for Claude Code, Cursor, and Codex

https://kastra.ai/
8•carlosjimenez1•7h ago•0 comments

Show HN: Devthropology – Better Insights for GitHub Repos

https://devthropology.com/demo
28•dpc94•5h ago•9 comments

Show HN: LastShelf – an emergency map of your family's documents bills& contacts

https://www.lastshelf.ai/
39•sbrown12•7h ago•24 comments

Show HN: FormGrid – Forms that can look like CLIs, pixel art, GeoCities and more

https://www.formgrid.com/
9•bjoernm•7h ago•2 comments

Show HN: Silo Terminal – a dark terminal profile inspired by Silo

https://github.com/onehorizonai/silo-terminal
3•gusnewman•2h ago•0 comments

Show HN: EVconomics – EV vs. gas cost-of-ownership calculator with live prices

https://calc.evconomics.org
2•DanCharge•3h ago•0 comments

Show HN: Rubber Duck

https://rubberduck.greg.technology/
8•gregsadetsky•8h ago•2 comments

Show HN: Onboard-CLI–a fast developer tool built in Go uses AST and LLM

https://onboard-cli.vercel.app/
2•yr_animesh•3h ago•1 comments

Show HN: Microsoft releases Flint, a visualization language for AI agents

https://microsoft.github.io/flint-chart/#/
337•chenglong-hn•1d ago•133 comments

Show HN: Codex Explorer, a local session manager for Codex CLI

https://github.com/Tobiadefami/codex-explorer
4•tobiadefami•4h ago•0 comments

Show HN: Follow London Trains in 3D

https://ride.nexttrain.london/
151•mgranados•5d ago•60 comments

Show HN: A zero-dependency auditor for decision records in AI-touched systems

https://github.com/forgedculture/legibility-field-kit
5•forgesignals•8h ago•0 comments

Show HN: Yamanote.fun – A complete soundscape for Tokyo's Yamanote line

https://www.yamanote.fun/
229•madebymagnolia•2d ago•50 comments