frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

I found 257 forgotten online accounts – built a tool to find yours

https://ghostsweep.com
1•Komieter•34s ago•1 comments

Show HN: Install agent skills from many sources using one command

https://github.com/xlab/agent-skills-project
1•Xlab•3m ago•0 comments

ICE Is Going on a Surveillance Shopping Spree

https://www.eff.org/deeplinks/2026/01/ice-going-surveillance-shopping-spree
3•BeetleB•4m ago•0 comments

Only known copy of Unix V4 found in storage

https://www.sltrib.com/news/education/2026/01/07/university-utah-finds-tape-with/
1•gusplus•6m ago•0 comments

Show HN: I made Python library for composition-first AI programming

https://github.com/prostomarkeloff/funcai
1•notmarkeloff•7m ago•0 comments

Show HN: I made Python library for high-order combinators (retry, timeout, etc.)

https://github.com/prostomarkeloff/combinators.py
1•notmarkeloff•7m ago•0 comments

Nanochat Miniseries v1

https://github.com/karpathy/nanochat/discussions/420
2•mfiguiere•7m ago•0 comments

Samsung Galaxy Z TriFold hands-on

https://mashable.com/article/samsung-galaxy-z-trifold-hands-on-ces-2026
1•kristianp•15m ago•0 comments

Show HN: Workflow automation builder for saving website data

https://www.addtosheets.com/automations/
4•siegers•19m ago•0 comments

AI hampered productivity of software developers,m

https://fortune.com/article/does-ai-increase-workplace-productivity-experiment-software-developer...
2•msolujic•20m ago•1 comments

Claude-Code v2.1.0

https://github.com/anthropics/claude-code/commit/870624fc1581a70590e382f263e2972b3f1e56f5
1•handfuloflight•21m ago•0 comments

I built a Million Dollar Homepage for text, and it's chaotic

https://www.themillionlines.com
1•lolzenom•23m ago•2 comments

Show HN: Kerns – A Continuous Research Workspace

https://www.kerns.ai/
3•kanodiaayush•24m ago•0 comments

FOSS book: programming essentials with Guile Scheme

https://www.draketo.de/software/programming-scheme
1•ArneBab•25m ago•1 comments

Show HN: LLM-powered What If text gen for fun

1•techbuilder4242•27m ago•0 comments

Fidji Simo: ChatGPT Health and what AI can do for a broken system

https://fidjisimo.substack.com/p/chatgpt-health
1•nadis•27m ago•1 comments

Minneapolis Mayor Blasts Kristi Noem's BS ICE Shoots Kills Woman in Minnesota

https://www.thedailypoliticususa.com/p/minneapolis-mayor-blasts-kristi-noems
24•Jacquie11•27m ago•10 comments

There Is a Sickness Eating Away at American Democracy

https://www.nytimes.com/2026/01/06/opinion/trump-jan-6-jefferson-davis.html
4•whack•29m ago•0 comments

How being on LinkedIn feels like now

https://plonkedin.vercel.app/
4•falloutx•29m ago•0 comments

Cursor: Dynamic Context Discovery

https://cursor.com/blog/dynamic-context-discovery
2•leerob•30m ago•0 comments

The Films of 2025:Q4

https://scottsumner.substack.com/p/the-films-of-2025q4
1•paulpauper•32m ago•0 comments

Teenygrad

https://github.com/tinygrad/teenygrad
1•jxmorris12•32m ago•0 comments

Persuasion of Humans Is the Bottleneck

https://erikschiskin.substack.com/p/persuasion-of-humans-is-the-bottleneck
1•paulpauper•32m ago•1 comments

CKSyncEngine Questions and Answers

https://christianselig.com/2026/01/cksyncengine/
1•chmaynard•33m ago•0 comments

The Philosophy of Solvej Balle

https://endsdontjustifythemeans.com/p/the-philosophy-of-solvej-balle
1•paulpauper•34m ago•0 comments

CheckMyLLM – A real-time "status board" for LLM reliability

https://checkmyllm.com/
2•valpine8•34m ago•2 comments

Longbeard: Catholic Social Teaching and AI

https://www.longbeard.com/blog/catholic-social-teaching-and-ai
1•admaiorem•34m ago•1 comments

Recipe for a great startup dev team

https://renaissance.kelsus.com/p/recipe-for-a-great-startup-dev-team
1•nadis•36m ago•0 comments

Show HN: AI Swarm v3 – Self-host your own headless AI agents

https://ai-swarm.dev
1•plurb-unus•38m ago•0 comments

HP Reveals Keyboard Computer with Ryzen AI Chip

https://www.hp.com/us-en/desktops/business/eliteboard.html
2•tonymet•38m ago•1 comments
Open in hackernews

Show HN: Symbolic Circuit Distillation: prove program to LLM circuit equivalence

https://github.com/neelsomani/symbolic-circuit-distillation
16•nsomani•1d ago
Hi HN, I've been exploring various applications of formal methods to ML/interpretability and I've been hoping to get more eyes on the approach.

I have been working on a small interpretability project I call Symbolic Circuit Distillation. The goal is to take a tiny neuron-level circuit (like the ones in OpenAI's "Sparse Circuits" work) and automatically recover a concise Python program that implements the same algorithm, along with a bounded formal proof that the two are equivalent on a finite token domain.

Roughly, the pipeline is:

1. Start from a pruned circuit graph for a specific behavior (e.g. quote closing or bracket depth) extracted from a transformer. 2. Treat the circuit as an executable function and train a tiny ReLU network ("surrogate") that exactly matches the circuit on all inputs in a bounded domain (typically sequences of length 5–10 over a small token alphabet). 3. Search over a constrained DSL of common transformer motifs (counters, toggles, threshold detectors, small state machines) to synthesize candidate Python programs. 4. Use SMT-based bounded equivalence checking to either: - Prove that a candidate program and the surrogate agree on all inputs in the domain, or - Produce a counterexample input that rules the program out.

If the solver finds a proof, you get a small, human-readable Python function plus a machine-checkable guarantee that it matches the original circuit on that bounded domain.

Why I built this

Mechanistic interpretability has gotten pretty good at extracting "small crisp circuits" from large models, but turning those graphs into clean, human-readable algorithms is still very manual. My goal here is to automate that last step: go from "here is a sparse circuit" to "here is a verified algorithm that explains what it does", without hand-holding.

What works today

- Tasks: quote closing and bracket-depth detection from the OpenAI circuit_sparsity repo. - Exact surrogate fitting on a finite token domain. - DSL templates for simple counters, toggles, and small state machines. - SMT-based bounded equivalence between: sparse circuit -> ReLU surrogate -> Python program in the DSL.

Limitations and open questions

- The guarantees are bounded: equivalence is only proven on a finite token domain (short sequences and a small vocabulary). - Currently focused on very small circuits. Scaling to larger circuits and longer contexts is open engineering and research work. - The DSL is hand-designed around a few motifs. I am not yet learning the DSL itself or doing anything very clever in the search.

What I would love feedback on

- Are the problem framing and guarantees interesting to people working on mechanistic interpretability or formal methods? - Suggestions for next benchmarks: which circuits or behaviors would you want to see distilled next? - Feedback on the DSL design, search strategy, and SMT setup.

Happy to answer questions about implementation details, the SMT encoding, integration with OpenAI's Sparse Circuits repo, or anything else.

Comments

aappleby•1d ago
No examples in the readme?
nsomani•20h ago
There are two examples provided - quote matching and bracket closing.