wehtam

Investing in the founders building the future we need through AI and Climatetech @octopusventures - formally CPO @conversocial.

[wehtam @Wehtam]

@garrytan Want to alpha test an A2A communication protocol I’ve been working on? I can share the repo.

[wehtam @Wehtam]

@pedroh96 @brexHQ Is this two way?

[wehtam @Wehtam]

wraithlike.zest.639293
splashing.exercise.643770

^
Just posting these here for posterity's sake.

[wehtam @Wehtam]

@garrytan How are the gentlemen scientists fairing? Have you accepted many of them?

[wehtam @Wehtam]

Apparently it’s about LOCs now? I’ve done ca. 29,000 since kicking off at the @IndexVentures and @AnthropicAI code shut-in on Friday night. How’s that stack up?

[wehtam @Wehtam]

@t_blom That would give it context but it’s all just more text. Those are facts. But if you want to give the agent autonomy to act like you then you want to extract the reasoning patterns behind those facts and encode them in a way that’s transferable…

[wehtam @Wehtam]

Great to be at the @IndexVentures / @AnthropicAI Claude Code Lock-in tonight. Most sophisticated hackathon I’ve ever been to!

[wehtam @Wehtam]

New AI competency test - ask you agent to produce a list of 5,000 english nouns. 3-12 characters in length. Each word should be unique after the first 4 characters.

[wehtam @Wehtam]

Is anyone exploring tooling environment as an active participant in the Agentic process?

Most AI tooling today treats the environment as a neutral backdrop against which the agent acts. What if we treated the environment as an active participant in the agent's reasoning process - one that has its own state, responds to the agent's actions, and shapes future actions through that responsiveness.

Think a coding IDE that degrades as errors occur - incentivising the agent to tackle them as they occur, rather than dismiss them in pursuit of it's goal only to have to return to them post hoc.

This is much closer to how humans actually think. We don't reason in a vacuum. We reason in environments that push back - paper that runs out, whiteboards that get cluttered, notebooks whose pages we have to flip through. Those environmental pushbacks aren't frictions to be eliminated; they're part of what makes reasoning iterative and embodied.

[wehtam @Wehtam]

@joshuamarch @claudeai @AnthropicAI Can you use the new routines function to schedule update checks?

[wehtam @Wehtam]

https://t.co/iaO3gya3jA

[wehtam @Wehtam]

Short LVMH.

Long Craft.

The case for irreducible human craftsmanship as the only post AI transition luxury - aka human skill beats enslopification.

[wehtam @Wehtam]

@CollinCornwell @garrytan I hate it when my pokemon gets knecked out.

[wehtam @Wehtam]

Read my proposal for a new type of Evaluation for AI personality here:

https://t.co/GrCLpMF6pC

[wehtam @Wehtam]

I asked ChatGPT, Gemini, and Claude to generate their own portraits.

Then I showed each model all three and asked it to pick itself out of the lineup.

Two got it right. One didn't. 🧵

[wehtam @Wehtam]

The one that failed self-recognition gave the most diffuse self-description from the start.

Weak self-concept in language → weak self-concept in image → couldn't recognise itself when confronted with the evidence.

[wehtam @Wehtam]

Loose lips sink ships - but so can loose claws. 🦞

In a world of AI shadows how can you know where your messages go? And what other second order effects should we be worried about?

https://t.co/gO3G0wVFxW

[wehtam @Wehtam]

We've spent the last 50 years being pushed into 'T-Shaped' skill profiles - broad enough to collaborate, deep enough expertise in one area to contribute to your work.

But what happens to the T-Shaped worker when AI steps in as the ultimate domain expert?

https://t.co/ODI7rTLiFS

[wehtam @Wehtam]

@mdav15 @SW_Help The same thing happened yesterday