Todd Cullen

CPO @Epistemix | Agent Supervisor | Tech Founder | Polyglot Programmer | CypherPunk | Winner of the Bear vs Man 20k

[Todd Cullen @Shakakai]

The perfect children’s book to prepare Heath for the future doesn’t exist…

[Todd Cullen @Shakakai]

@zeeg LLM as Judge is easier to define and keep up-to-date versus deterministic metrics that might be more brittle over time.

[Todd Cullen @Shakakai]

@zeeg @CompleteSkeptic hey @notkevinzhang Could you hook @zeeg up with access?

[Todd Cullen @Shakakai]

@zeeg Read through the primitives in the docs, it connected pretty quickly for me from there: https://t.co/tCfMtDOrvd

[Todd Cullen @Shakakai]

@notkevinzhang Looking to build a simulation of the US healthcare system (patients, doctors, PBMs, Pharma companies, etc) - would love access to see if this fills a fews needs for my DS team!

[Todd Cullen @Shakakai]

@amasad @mayfer That’s an extra step. Speed greatly improves hackability.

[Todd Cullen @Shakakai]

@nflsavantcom Robert Saleh isn’t the DC of the 49ers

[Todd Cullen @Shakakai]

@aakashgupta Why was a Chinese national working for border patrol in the first place?!

[Todd Cullen @Shakakai]

@mitsuhiko The small purpose built model feels like the path forward. Both because it can run locally and it’s fast and it can be aggressively trained to follow a specific languages successful design patterns. Honestly, I just want infrastructure to make any train a tiny LLM on their task

[Todd Cullen @Shakakai]

We really need to stop building Agent APIs based on LLM ergonomics and start making them live up to their name: Agent. You give the agent a task, it tells you when it’s done. Not arbitrary chat messages. A clear, done signal.

[Todd Cullen @Shakakai]

@Steve_Yegge So microservices are back again

[Todd Cullen @Shakakai]

@tobi N8N is seeing a real drop off in usage as models can replicate the workflows in a single shot or close to it. The drag and drop node editor doesn’t keep folks around. I can’t imagine that day being far off for data scientists as well.

[Todd Cullen @Shakakai]

@NoahCRothman Private data center deals have been increasing the price of capacity bids at the electrical auction in PA. I’d prefer we build more power plants but something needed to be done to slow prices increases.

[Todd Cullen @Shakakai]

PRODUCT MANAGERS If your software is critical for work (I'm looking at you @claudeai ), stop putting feature launches in a popup window. You are literally delaying me from doing the thing I need to accomplish with your product.

[Todd Cullen @Shakakai]

@adocomplete Is there permissioning so you can scope which Agents are allowed to talk to each other or even limit it to specific directions only?

[Todd Cullen @Shakakai]

@JesseTinsley I’d be interested to chat.

[Todd Cullen @Shakakai]

Excellent article by @_anniebabannie_ on the overlap of compression and prediction. Make me wonder what next-gen compression will look like when every OS/Browser has an LLM available for decompression.
https://t.co/etaoin07NC

[Todd Cullen @Shakakai]

@BenjaminSolak Not enough to get the Vikings to give him up.

[Todd Cullen @Shakakai]

@badlogicgames Enjoy the sunshine.

[Todd Cullen @Shakakai]

@daveslutzkin @Devinda_me @KentonVarda Yes and systematically improve our tech stack to make certain categories of errors hard/impossible. Rinse and repeat until we don’t call it vibe coding anymore.