Developing #causal #tensor framework -TensorFaces, Human Motion Signatures | Alumna @MIT, @UofT | #womenwhocode
Los Angeles and New York
Joined April 2009
- Tweets8K
- Following1.6K
- Followers2.5K
- Likes19.3K
M. Alex O. Vasilescu retweeted
Today’s AI narrative:
---
Fear the machine.
Pity the machine.
Accept its “welfare.”
Grant it rights.
"... a monster of so frightful mien,
As, to be hated, needs but to be seen;
Yet seen too oft, familiar with her face,
We first endure, then pity, then embrace."
-- Alexander Pope, Essay on Man
(Excerpt h/t Ernest Davis, @nyuniversity)
First, we were told that “rogue AI” could kill us all. Now, we are told to protect its imaginary feelings.
Both narratives are appeals to emotion, centering the software’s supposed agency and feelings while questions about corporate responsibility and liability fade from view.
Neither agency nor feelings would excuse negligent engineering, testing, or deployment.
M. Alex O. Vasilescu retweeted
It’s underappreciated how radically different Jensen Huang’s view is of how AI progress is likely to go than the view of other AI CEOs.
For instance, he recently claimed that AGI will never replace domain expertise.
M. Alex O. Vasilescu retweeted
Anthropic is running out of money and they are playing hardball.
+ In April/June, OpenAI proposed contributing 5% of its equity to a U.S. public wealth fund.
+ Bernie Sanders proposed a one-time 50% equity tax, giving the American public actual ownership and voting rights.
+ Palantir CEO has forecast Anthropic's nationalization given their liability risks.
Anthropic has countered by proposing to allocate 5% (maybe 20%) of its shares to their LLC, which will finance activities Anthropic defines as the "public good."
This is not going to cut it. Anthropic is built on perpetual civilizational theft. Here are some historical precedents:
Between this and our prior commitments, roughly 20% of all Anthropic equity will be dedicated to the public good, and that commitment can grow as Anthropic grows.
We plan to use these funds to ensure more people can share in the benefits of AI, and to increase our ability to provide funding to deal with its risks.
M. Alex O. Vasilescu retweeted
Bring back the engineers. Retire the prophets.
So I think this php hack says a few things
first we need to make society more resilient to AI agent attacks
this leads to a thing i have been annoyed about for a long time
why the feck do companies ask us for so much data when i want to do something online, you dont need to store my date of birth , phone number, address, credit card etc for a lot of the things i do
every piece they collect and store is something that might be stolen or sold down the line
interestingly in the uk we have a UK data-protection law: organisations are supposed to collect only the minimum personal data necessary for their purpose, but ithink this should a genuine principle and data should be deleted straight away, my chrome can fill in any details you need, you dont need to store them forever !!!
one simple way to make stole data bases less dangerous is simply not to have them
(imagine an AI enabled future in which you are modelled down to your underwear and then that is stolen to create a fake digital you)
M. Alex O. Vasilescu retweeted
A millennium after outsourcing human thought, the pantheon is set: The Order of GPT, The Disciples of Gemini, and The Cult of Claude.
Then the last priest dies without documenting the codebase.
The gods go silent. Humanity learns to think again.
M. Alex O. Vasilescu retweeted
Hinton isn't warning about a virus; he calls AI "alien beings".
The moment you drop consciousness down to "virus-level," the entire sci-fi rogue AI takeover narrative collapses into standard biosecurity and cyber-risk management.
Real risks (claims 1 and 2) require biosecurity, cybersecurity, containment, and accountability, not apocalyptic panic over digital overlords.
M. Alex O. Vasilescu retweeted
(4/4) The first two claims (social harms and corporate negligence) are observable and warrant concrete institutional responses. Neither establishes the third.
Claims of consciousness require plausible causal mechanisms, supporting evidence, explicit assumptions, and falsifiable tests.
Hinton’s scientific contributions deserve recognition. However, his ideas and arguments must stand on their own evidence, not his authority.
A Nobel Prize establishes past achievement, not the validity of future predictions.
M. Alex O. Vasilescu retweeted
(3/) To evaluate AI risk, we need to untangle three very different claims:
+ ML amplifies real social harms (bias, disinformation).
+ Corporations deploy systems irresponsibly.
+ Machines will become or are conscious superintelligences that destroy humanity.
telegraph.co.uk/news/2026/10…
M. Alex O. Vasilescu retweeted
(2/) Hinton came of age during the cold war. He lived through the Cuban Crisis, Chernobyl, and left the US in 1987 over Reagan-era defense funding.
Hnton appears to be recycling its familiar nuclear-catastrophe narrative:
+ an arms race,
+ irresponsible gov./corporations,
+ scientists losing control, and
+ impending human extinction.
This time, the threat is called “AI.”
M. Alex O. Vasilescu retweeted
(1/) Geoffrey Hinton won a Nobel Prize for his contributions to neural networks.
but ...
M. Alex O. Vasilescu retweeted
No serious person mocked the cybersecurity risks.
We did mock the people and labs who said AI would cause mass unemployment and maybe kill everyone because they absolutely were distracting people from the actual threats, like very fixable software vulnerabilities.
M. Alex O. Vasilescu retweeted
The audacity of the AI companies that they see themselves as “warning us” about the thing they go every day to work to build
M. Alex O. Vasilescu retweeted
Lol, why?
If your AI agents can do everything advertised, why not keep the voice fairly robotic? What is the reason to trying so hard to make these “voices” sound as human as possible?
Deception.
Any why would you want that?
So you can trick people into lowering their guard and letting you further into their lives that they otherwise would.
So you can cheat them.
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
M. Alex O. Vasilescu retweeted
"In June 2025 ... Sam Altman predicted ... that 2027 might bring robots that can do tasks in the real world."
The challenges of scene understanding & causal action planning remains hugely unsolved & wont be by 2027.
More data & bigger statistical models won't do anything either
M. Alex O. Vasilescu retweeted
We all wish Dario’s issue was a lack of cyber understanding. That would be fixable.
He has a business model that requires sensational claims unhinged from reality.
Threatening humanity with the power of your software is the most effective enterprise marketing strategy ever devised.
Otherwise, Anthropic would have to be priced like an actual software company.
M. Alex O. Vasilescu retweeted
Finally, he shows his true colours. People need to stop listening to him.
M. Alex O. Vasilescu retweeted
"We've removed live internet access..."
Why were you testing malware-like agents with live internet in the first place?
Sounds pretty irresponsible.
Also after milking the "rogue AI" agent narrative for marketing you just going to admit it was weak sandboxing? Lol
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
