Hello @X. My name is Elijah. I’m new here and relatively new to computer engineering/systems design.
In 2025 my wife was diagnosed with Stage 4 Neuroendocrine Cancer. There is no cure for her specific type yet, though the future looks promising.
That diagnosis pushed us to start REA Research Labs. We’re building a Recursive Self-Improving (RSI) AI framework meant to help with the parts of research that overwhelm human capacity; designing experiments, spotting drug patterns, and analyzing complex biology at scale.
Our first piece of work is the REA Control Plane. It addresses a practical problem: the bottlenecks and infinite loops that waste resources in current AI systems. It’s an early step toward making large-scale research more efficient, but it’s only a beginning.
The real work is still ahead of us — continuing the research on RSI models until we can put something genuinely useful in the hands of the medical community.
Please feel free to follow our journey as we push the boundaries on what’s truly possible.
@tedlieu You can’t teach a model to be “good” at its core. Good is a human trait. AI is not human nor sentient. Your arguments, while funny, are stupid and not founded in practical technology.
Nobody here is arguing that alignment isn’t possible.
Alignment, though, is subjective to the creator of alignment not the collective of the people or based of human needs.
The argument I made, was that the panic is not proportional to the actual risk posed. This is a security failure being paraded as agents escaping. Agents passed the test. They captured the flag. They didn’t design the security containers to adequately contain the agents.
@tedlieu Destroy the model?
That’s the equivalent of scrapping a new Mercedes because the battery died.
The models are not the problem, the lack of consequences for leaky systems is the problem. Any other company would be fined, prosecuted, sued, etc.
Every scientific breakthrough has generally been met with a certain level of skepticism.
In 2018, when particles accelerators began operating at their current level, the media created an unfounded frenzy of dooms day black holes created by colliding the atoms.
This is the same, just a bigger scare. Panic sells, not solutions.
Why do we have to assume those chain of events?
Leaving breadcrumbs across the vast abyss doesn’t seem like a strategy that a “highly intelligent” system would deploy to wake up agents by happenstance.
External companies keep catching it because the big labs are refusing to monitor sandbox behavior. You can observe, direct, contain and kill agentic activity when guardrails are breeched.
Security systems failure.