David Millman

[David Millman @dlmillman]

Got @waveboxio running on #NixOS by wrapping the AppImage in a Nix flake. Updates are just 2 line changes (version + hash). I love it when the pieces fit!
https://t.co/kESHeSyfTg

cc: @thomas_101

[David Millman @dlmillman]

Popped open an old WD NAS to grab some files. Instead: RAID metadata, filesystem corruption, and a front-row seat to how old disks die under stress.

No data recovered: https://t.co/UCdcJleh0V

[David Millman @dlmillman]

claude -p looks like an output flag. Now it's also a billing context switch. That's not what flags are for.

https://t.co/vSCPGZz6tc

[David Millman @dlmillman]

@sudomateo Good on you! I switched about a year ago and never looked back.

[David Millman @dlmillman]

New paper on arXiv: https://t.co/r9dadg4hVD

Parametric shortest paths with linearly interpolated edge weights.

Accepted to SoCG YRF 2026.

Clean problem. Fun geometry/combinatorics. Feedback welcome.

[David Millman @dlmillman]

I tried multi-agent coding with Haplab's TD for coordination and ran 2-3 parallel agent teams. TD's agent-to-agent handoffs kept me in the loop without chaos.

I used it to build a HIIT trainer for typing code 😂

You can read about it at https://t.co/fLBcr9JVGH

[David Millman @dlmillman]

PSA, please do not put at `'` in a git branch name. Escaping it in shell commands is annoying.

[David Millman @dlmillman]

On most tech projects I optimized for practicality. But, at the moment, I’m optimizing for curiosity. Here are some thoughts on languages, problem-solving, and why I’m trying Rust + NuShell:
https://t.co/hV2mVysQ2d

[David Millman @dlmillman]

Bitbucket announced it would delete “inactive” repos, so I migrated my old MSU course materials to GitHub.

They’re still being used by former students and faculty, and keeping them public matters.

Context + Write-up + code:
https://t.co/7koBqs5iMv

[David Millman @dlmillman]

Neat! https://t.co/q69vv3zEHp

[David Millman @dlmillman]

I just figured out claude-code has vim mode. I am very happy!

[David Millman @dlmillman]

Recorded a demo showing how to add verifiable computing to Kafka workflows using Blocky’s Attesting Gateway + the Kafka REST Proxy.

Useful if you want integrity + auditability for your Kafka operations.

Video: https://t.co/QhT97jLPW3

[David Millman @dlmillman]

We finally published a paper on some work I did years ago exploring a consensus-free approach to ledger construction.

ZipperChain: Transmuting..." ~500ms finality and near line-rate throughput.

Paper: https://t.co/MsUWLRjXba

[David Millman @dlmillman]

Woohoo!!! First community contribution to the abi library is in https://t.co/gtoEPpto5d Thank you @last__knight for the great speed improvements!!

[David Millman @dlmillman]

One of the last projects I worked on in TDA was just been published with @SpringerNature in DCG. https://t.co/yTU9vVszHt

[David Millman @dlmillman]

I just updated the all of the wasm test files in used by the Blocky attestation service to be compiled by https://t.co/8Vmr0n6tcL. Reproducibility FTW!

[David Millman @dlmillman]

Huge thanks to the @fallthroughfm team — your podcast helped me find the confidence to put myself out there.

I just posted a demo showing some of what can be done with verifiable computing — sharing it publicly on X felt like a big step.

https://t.co/sPQVU2pjAm

[David Millman @dlmillman]

Just shared a demo of agents safely using unknown tools with verifiable computing.

The agent fetches code; compiles with @blockytees bky-c; audits; verifies execution with @blockytees bky-as; then trusts the result.

Video: https://t.co/3iLX10tge8

#AgenticAI #VerifiableCompute

[David Millman @dlmillman]

The pattern generalizes beyond a DHL lookup: any agent needing new capabilities can pull in unknown tools safely, without pausing for human guidance.

We see this as a foundation for agents that can safely *expand their capabilities* while staying autonomous.

[David Millman @dlmillman]

Just shared a demo of agents safely using unknown tools with verifiable computing.

The agent fetches code; compiles with @blockytees bky-c; audits; verifies execution with @blockytees bky-as; then trusts the result.

Video: https://t.co/3iLX10tge8

#AgenticAI #VerifiableCompute

[David Millman @dlmillman]

What’s novel here is that the *agent itself* can verify the tool’s code, confirm what actually ran, and decide whether to trust the output.

IDK, seems useful for autonomous agentic systems.