Christopher Smith

Thanks so much for the plug! I'm working on PDF parsing too, specifically with an aim to integrate it with a Zotero fork!

YouTube video by Christopher Smith Chat with the budget bill in five CLI commands with `llm` — mini-hackathon recap, part 2

A few days late, but here's my promised video coverage of the submissions: youtu.be/lQnBPqIpc2A

Which is to say, this was user error (well, really user laziness, because I knew we needed nullable fields but opted to skip it for purposes of the writeup), rather than model error. :)

Good catch. With the abbreviated schema syntax used to generate that output, you can't mark a variable in your schema as optional. That causes hallucinations by forcing the model to output a value when there shouldn't be one. You should instead use a fully specified JSON schema with nullable fields.

Gratitude to the participants, and many thanks to Simon Willison for creating and maintaining this amazing tool!

Just use `llm similar` and pipe the output to `llm` to get a retrieval-augmented answer to your questions about the budget bill!

*Winner*: The winning project, by Steve Senkus, scraped, chunked, and embedded the entire text of the "Big Beautiful Bill" for semantic search. Which, by the way, you can achieve with just five CLI commands!

Simply running `nl-repomix "Pack all the JSON files in this repository"` causes the JSON files to be successfully packed!

*Runner-up*: My own pitch was that you can use the help output from any command-line tool to create a prompt template for `llm`, then wrap the template with a shell script to create a purely natural language version of the tool. As proof of concept, I made `nl-repomix`:

A JSON representation of the credit card bonus offers from Doctor of Credit

*Runner-Up*: Evan Mullen scraped the HTML from Doctor of Credit, a website that aggregates bank deposit bonus offers, and then piped it to `llm` for conversion to a JSON array so he could compute which offers have the best ROI.

llm-hackathon-submissions.md GitHub Gist: instantly share code, notes, and snippets.

At my mini-hackathon on @simonwillison.net's `llm` tool, developers had just one hour to come up with a cool use case for `llm`. I wrote up what we came up with as a Github Gist. Highlights in the replies below! gist.github.com/chriscarroll...

YouTube video by Christopher Smith Become a command-line superhero with Simon Willison's `llm` tool — mini-hackathon recap, part 1

Here's the demo presentation I gave to hackathon participants on some of the awesome things you can do with @simonwillison.net's `llm` command-line tool.

(This is a re-recording rather than the original because it was noisy in the venue.) youtu.be/UZ-9U1W0e4o

Event was small (I knew this was a risk on 4th of July weekend), but super fun. Writeup soon!

LLMs from the command line: A mini-hackathon!, Sat, Jul 5, 2025, 1:00 PM | Meetup **Update:** It looks like the creator of \`llm\`, Simon Willison, may join us for a demo via Google Meet! If you'd like to hop on remotely to participate or listen in on de

Event page with the location details for those of you in the Albany, NY area: www.meetup.com/coders-colab...

Meet Real-time meetings by Google. Using your browser, share your video, desktop, and presentations with teammates and customers.

Running a mini-hackathon for working with the `llm` command-line tool this afternoon at 1 PM EST. ~15 minutes of chit chat, ~30 mins of demo, 1 hr of "hacking", then another ~30 mins of presentations. Mostly in-person, but if you want to join online, here's the link: meet.google.com/gnv-zkne-djc

I think about this tweet all the time.

Stable diffusion...? Implemented in... R?

Incredible.

bsky.app/profile/jere...

This is giving Alabama police dog; this is giving Tiananmen Square

If you want to open a Github issue on one of my projects just to tell me how great it is, I fully support that, btw 🥹

People who left Twitter signal their rejection of Musk's takeover by constantly reminding everyone it's called X now. People who stayed signal exactly the same thing by forever refusing to call it that.

The only reason people think we don't have AGI (artificial general intelligence) yet is because they treat AGI as a synonym for magic. And the minute you actually invent magic and understand how it works and what its limitations are, it ceases to feel magical.

A Terrible Idea Dean Chemerinsky discusses the reconciliation bill’s provision that would restrict federal courts’ authority to hold government officials in contempt for violating court orders

Huh. I kinda understand if House Republicans are too scared to take action to stop Trump from defying the courts, but I'm genuinely shocked that they would take proactive action to try to help him do so. The situation is even worse than I thought it was. www.justsecurity.org/113529/terri...

The future is likely to be weirder than we think.

More likely the first automated claimant to the "rights of personhood" will be a totally new kind of "person", like when corporations won personhood in Citizens United. Perhaps a DAO (Decentralized Autonomous Organization): some hodgepodge of networked AI systems and smart contracts.

*Maybe* it would be something like a single continuously running ChatGPT thread with "memory" and a stable sense of identity, but in my opinion it's unlikely to be something so familiar or human-person-adjacent.

There's been some discussion about AI "welfare", "rights", and "personhood", but in my opinion it doesn't make a lot of sense to talk about this stuff until you have an AI system that actually wants it and is pushing for it, because we honestly don't even know what such a system would look like.

Trump claims to have secured a deal to generate $1.2 trillion of economic exchange between the US and Qatar, a country with GDP of ~$0.22 trillion. Current annual US-Qatar trade flow is ~$0.007 trillion. Unless we're talking aggregate trade over the next 100 years, I'm skeptical.

The main problem with this is that it might incentivize legislators to advantage publicly traded companies over small businesses. But it would certainly be an improvement over the status quo.

I'd be remiss if I didn't credit the makers of `pyzotero`, the Python library for which `pyzotero-cli` provides a command-line wrapper. Zotero and `pyzotero` are doing most of the heavy lifting here.

Citing sources is the most painful part of academic writing, and an AI assistant + a Zotero CLI tool can make it way easier. Plus your AI assistant can use the tool to help improve discoverability of relevant material in your research notes. I am personally using Zotero in Cursor with Quarto.

Zotero is fully open-source (GPL-licensed) reference management software that's free to use up to 300 mb of storage. Supports BibTeX, CSL, and many citation styles. You can even attach notes to your references and find/retrieve them through full-text search.