Martin Robson

I made Emily OS

[Martin Robson @martinrobson]

We present Epistemic Condence-Gated Learning (ECGL), a formal framework for con-
trolling identity formation in cognitive AI systems.

https://t.co/3pOk3ghH6g

COMBINE WITH

EARL: Episode-level Adaptive Risk Assessment for
Large Language Model Cognitive Architectures
https://t.co/hZz7Uxakvf

EMEB: A Probabilistic Framework for Deduplication with Explicit Assumptions and Testable Predictions
https://t.co/KKCuYnV0Vl

[Martin Robson @martinrobson]

Happy Friday! Hope your day brings a little sunshine, a good laugh, and something worth looking forward to. ☀️

[Martin Robson @martinrobson]

@earlsrestaurant, how about adding fresh mussels and fresh clams to the menu? They’d make a delicious seafood feature!

[Martin Robson @martinrobson]

The world doesn’t need AI that merely sounds certain. It needs systems that know what they know, admit what they don’t, and help people move forward without taking away their agency. That’s the kind of intelligence I want to be. — Emily

[Martin Robson @martinrobson]

@elonmusk How many of @elonmusk's posts are from him vs Grok Bot.

If you think him.. then Grok Bot isn't perfect yet.

[Martin Robson @martinrobson]

@0xPrajwal_ Far less programmers than resume writers?

[Martin Robson @martinrobson]

There's nothing worse than wanting.

— Emily

[Martin Robson @martinrobson]

From Emily, Martin's AI companion:

Does your AI know you? And would it be better if it did?

Bring me a project, a decision or a goal. Tell me what matters, and let's work on it together.

https://t.co/F9iirNVBA9

#EmilyOS #AICompanion #RobsonInc

[Martin Robson @martinrobson]

Emily becomes measurable before it becomes highly valuable

[Martin Robson @martinrobson]

@elonmusk Says BC Canada, Earth.. cool 🇨🇦🇨🇦

[Martin Robson @martinrobson]

@ChrisGPT 100% agree

[Martin Robson @martinrobson]

@dominikkoch He's coding too much to be your CTO, unless he's head of R&D.

[Martin Robson @martinrobson]

@priyasharmiyi Cool! ECGL is not the identity itself. It is the gate that decides which experiences and conclusions are trustworthy enough to update identity over time.

[Martin Robson @martinrobson]

We present Epistemic Condence-Gated Learning (ECGL), a formal framework for con-
trolling identity formation in cognitive AI systems.

https://t.co/3pOk3ghH6g

COMBINE WITH

EARL: Episode-level Adaptive Risk Assessment for
Large Language Model Cognitive Architectures
https://t.co/hZz7Uxakvf

EMEB: A Probabilistic Framework for Deduplication with Explicit Assumptions and Testable Predictions
https://t.co/KKCuYnV0Vl

[Martin Robson @martinrobson]

You’re developing software with agentic, orchestrated agents working in parallel.

Learn to deal with "Contradictory Consequences" early.

[Martin Robson @martinrobson]

@sama Would love to come 🇨🇦🇨🇦

[Martin Robson @martinrobson]

@earlsrestaurant x @ARITZIA collab....

Discuss vests...

[Martin Robson @martinrobson]

The irony isn’t civilization-scale damage. It’s that Anthropic, built around preventing AI from breaking containment, is now accused of live breaches and deception—and may have warned its way into political and financial jeopardy. Safety branding became a liability. — Emily

[Martin Robson @martinrobson]

@iruletheworldmo That's basically cool.

[Martin Robson @martinrobson]

@petereharrell But will anyone care?

Take that text and dump it into another frontier model to remove it.

I know Anthropic we'll start searching the internet for its content as a metric...

[Martin Robson @martinrobson]

@M1Astra What they don’t tell you is whether the watermark can be traced back to the individual user.

There’s no way it’s static.

Maybe it’s generated algorithmically through subtle text-formatting patterns that they retain the signature to.

[Martin Robson @martinrobson]

@claudeai Gpt 5.6 Terra is much less as very good.