Twitxy Saved

Ariadna Font Llitjós

Verified account This account is verified. Learn more

@quicola

Co-founder & CEO, Alinia.AI | Passionate about Responsible AI, making tech work for all, NLP, research, UX, 🧘‍♀️, hiking, m&m, 🐶
Barcelona + New York, USA Joined March 2010
2K Following 3.4K Followers 2.7K Posts

Latest posts 5

Pinned post

Ariadna Font Llitjós Verified account @quicola
We open sourced Sev huggingface.co/alinia/Sev
Matching Jev's ranking took a week. Making its decisions trustworthy is a different problem, one that still takes expert-labeled data and a judge trained on it.

alinia.ai/resources/how-…
huggingface.co alinia/Sev · Hugging Face Text Classification model. Download, explore files and run it on Hugging Face.
2 replies 9 reposts 23 likes 818 views View on X
Ariadna Font Llitjós Verified account @quicola
That is it! Scott Kupor (Vice Chair of the Super Intelligence Task Force):

"With the new coding tools there has been a 50% increase in software productivity YtY, but the error rate is not going down, which means 50% more changes and therefore bugs."

"As these tools get better and better, error rate will go down, but right now it just means increasing the surface for potential incidents due to bugs and errors."
Ariadna Font Llitjós Verified account @quicola 2d
Am I the only one?
Who here thinks the recent failures might have something to do with vibe coding making its way into our critical infrastructure?

I’d be particularly interested in your PoV @karpathy
View quoted post on X
Ariadna Font Llitjós Verified account @quicola
Am I the only one?
Who here thinks the recent failures might have something to do with vibe coding making its way into our critical infrastructure?

I’d be particularly interested in your PoV @karpathy
Jeff Stein Verified account @jstein_sun 3d
NEWS: Wall Street execs and NSA have told Congress that outages were caused by patching vulnerabilities identified by AI models, sources say

AI did not cause outages but found so many issues that the rush to fix them before an attack overwhelmed systems

washingtonsun.com/technology/ban…
View quoted post on X
Ariadna Font Llitjós Verified account @quicola
And now, SI definitions aside, to the really interesting part of the new OCC supervisory framework on Bank Use of Super Intelligence, which will shape the way banks and, consequently, many fintechs, think about risk governance and control, and what requires extra scrutiny.

Examples mentioned for critical and high-risk tiers will likely be considered material, and thus in scope for MRM functions.

Critical (tier 1) risk examples include autonomous payment agent, real-time credit decisioning, fraud-hold release and market access.

High (tier 2) risk examples include SI-assisted underwriting, agent-initiated customer communications, AML alert disposition, code assistance for internal code and data access changes.

Unless I am mistaken, this affects pretty much all banks today.
Ariadna Font Llitjós Verified account @quicola
A few days ago, the Office of the Comptroller of the Currency (OCC) issued a supervisory framework designed to support OCC examiners in evaluating banks’ use of Super Intelligence (SI) in normal-course business activities. occ.gov/news-issuances…

Lots to unpack, and I am happy to see key aspects of making these AI systems safer explicitly mentioned, such as Guardrails, Red teaming, Adversarial testing, Benchmarking, Explainability, and Third-party SI risk mitigation.

Interestingly, the OCC’s definition of SI is unexpectedly broad, and does not even mention what most definitions of SI consider key: their ability to outsmart humans, capable of complex, creative behaviors that people cannot fully understand or evaluate. 🧵