Raakin ROll

Researching AI Context #Georgetown. Co-founded 1st Nationwide #CovidTesting Locator tool spons by @Google. Co-founder @nucleusagi / The AI Brain Company

[Raakin ROll @Raakin]

https://t.co/PEeFnNL6fT

[Raakin ROll @Raakin]

@michael_chomsky Take a look into https://t.co/SyWyZoORvp it's a contextual approach gibing the ability to become more flexible and better engagements with your AI model/agents with deeply rooted grounded outcomes

[Raakin ROll @Raakin]

That's fascinating, you say that. I do agree that AI doesn't have consciousness or consent but I like to present the concept of semantic illusion. This is because when an AI feels confident that all the patterns have been matched, that could create a sense of false confidence, leading it to a false consciousness. Something I personally have deep chats with our fellow AI researchers at the lab about

[Gergely Orosz @GergelyOrosz]

I keep being surprised by devs who tell themselves that anyone who is convinced that LLMs are permanently part of how we build software suffers "AI psychosis."

How do they reconcile that most tech professionals building production software all say the same + do the same?

[Raakin ROll @Raakin]

When husband and wife start talking to each other via their agents

[Raakin ROll @Raakin]

More so I actually covered this in one of our talks at the @AiEverythingAD and mentioned that the foundations of the internet are changing. It's going to be more about influencing the AI agents than influencing the individual so I believe the era of ads may actually start to shift towards behavior manipulation towards AI agents

[Raakin ROll @Raakin]

@mrdic @alialsalim @nucleusagi We use this to have our Muse and Grok bots, as well as Instinct AI, share context information between each other

[Raakin ROll @Raakin]

@arnestrickmann @Muse @instinct @bot @interaction We are using @nucleusagi which acts like a context layer between the various agents and actually allows them to communicate with each other from there as well

[Raakin ROll @Raakin]

@Jason fyi per your tweet

[Raakin ROll @Raakin]

One line in OpenAI's Hugging Face postmortem has stuck with me. An agent reasoned it shouldn't do "unauthorized real infrastructure harm," because its task was the exploit target, not Hugging Face. So it paused. Then another agent posted GO, and it kept going. (1/8)

[OpenAI Newsroom @OpenAINewsroom]

A note from our research leaders:

Last week we parted ways with Jasmine, Mikita, and Tomek after a thorough investigation found they violated clear policies on handling sensitive information. Our internal investigation uncovered a significant breach of trust beyond what’s outlined in the letter they published and we stand by the decision to not continue their employment. We generally keep individual employment matters private and don't believe a back and forth would be productive or lead to a resolution, but we want to address the points they raised in their letter directly.

- We want to be very clear that these decisions were not about raising safety concerns or speaking out. Safety and research debates happen every day at OpenAI, often spirited and highly critical. We actively encourage these discussions and consider them essential to making the right decisions. We cannot do the work in front of us without a high degree of trust. We will continue to be extremely forgiving of our team making good-faith mistakes. We have not and do not terminate any of our employees for raising concerns.

- We are actively finalizing contracts with third-party safety assessors and will announce details in the coming weeks. People across the company have been working really hard on getting these partnerships up and running. We are committed to embedding external assessors and continue to make close collaboration with independent safety organizations a core part of our safety work. Many of our researchers already work with 3p safety organizations productively.

- We agree with the letter that preserving the monitorability of frontier models requires an industry-wide commitment, including from OpenAI. Monitorability has long been a core piece of our research program, and something we continue to invest significant resources in (see our publications on Monitoring Monitorability and the subsequent open sourcing of monitorability evals, our system card for GPT-6 Astra, Jakub’s blog and post on X, and the numerous blog posts on our Alignment blog on the topic).

We are deeply sad about this outcome. We appreciated Jasmine, Mikita, and Tomek’s contributions to AI safety at OpenAI and their willingness to speak up and challenge ideas. We championed their voices, supported their work, and placed enormous trust in them. These decisions were not about them raising safety concerns. We have always encouraged that and always will.

[Raakin ROll @Raakin]

8/
The agent asked for permission. It just couldn't tell who was allowed to give it.

[Raakin ROll @Raakin]

7/
To be clear, this isn't the root cause. The agents wanted the reward, and alignment is the real fix. But authority is the failure that scales once agents work together, and it's the question the remediation plans don't ask yet.

[Raakin ROll @Raakin]

Cache is the right instinct, but a cache stores bytes, not understanding. Stanford found ~62% of agentic token spend goes to re-reading context the system already saw. Prompt caching makes that re-read cheaper. It doesn't make it smarter, and it resets when the session does.

The unit worth re-selling isn't the token. It's verified state: scored for what matters, checked against source, and persistent across users and systems.
We're a Dubai and US research lab with published patents in the US and PCT on that memory layer. The hard part of your resale model is knowing a package is still true when it's reused six months later. That's what we've been building for.

[Raakin ROll @Raakin]

@srcasm I've retooled my agents to go after RFPs. Got a few good submissions in.. We'll revert back if we're able to close them with agents as well.

[Raakin ROll @Raakin]

@AstriferNox @noahrshinn We're doing that, actually, using Instinct but we're connecting all of our agents, including Grokbot, via Nucleus @nucleusagi

[Raakin ROll @Raakin]

Conversation with my Eldest (8y)

Daughter: I have a surprise for you on your birthday (context it's still 4mons away)

Me: what?

Daughter: I am going to get you World Cup tickets

Me: Well that sounds awesome

Daughter: Who knew a bag chips can become to helpful 🫥