AI Ethics and Governance. Software Engineering applied to AI development. Futures Scenarios and Strategic Foresight specialist. AI Ethics Alliance.
Spain
Joined April 2008
- Tweets4.4K
- Following599
- Followers1.3K
- Likes2.1K
Dhaunae De Vir retweeted
The challenge? To develop an agentic AI that incorporates principles of human oversight, data governance, and cybersecurity, using the AI Sandbox experience. And we'll be doing it together with
@SEDIAgob. The challenge also focuses on child protection with @UNICEF #OpenSourceAI
Dhaunae De Vir retweeted
Hace unos días os contamos que la @agenciaAESIA iba a cruzar el Atlántico... ✈️Hoy os damos una pista más del motivo. ¡España🇪🇸 ha sido el único país seleccionado por @UN para proponer un reto de código abierto en la #UNOpenSourceWeek organizada por
@ODET_UN 🙌
Dhaunae De Vir retweeted
A few months ago I was curious to know how much Anna's Archive was charging AI developers for access to their massive library of pirated works for training - so I emailed them saying I was interested in buying access.
Here is their reply.
They are charging $200,000 (payable via crypto, of course). You get high-speed access to the full collection. This includes more than 60 million books.
(This is also the group that recently stole Spotify's entire music catalog, so expect this to be available to high rollers too.)
This is the collection Nvidia is accused of accessing for training.
Training on pirated works is rife in the AI industry. It's been embraced by some of the biggest AI companies in their greed-fuelled race to win the AI market.
When they claim what they're doing is somehow fair, remember that not only is it theft (which is bad enough) - it supports further theft by funding pirates.
We desperately need a hard reset in the AI industry. It must turn away from theft, and start paying the people whose work it relies on.
NVIDIA allegedly contacted Anna's Archive directly for access to ~500 terabytes of "pirated" books and papers for pre-training their LLMs
Anna's warned them the collections were illegal and copyrighted.
NVIDIA's data strategy team pushed anyway; executives gave the green light within days, per internal docs cited in the lawsuit.
“Based on our conversation history, create an image of how you feel I treat you.”
— Prompt idea by Alan Harris
This is how ChatGPT ‘feels’ about our interactions.
ALT A happy robot being gently patted on the head by a human.
I agree with Alan Harris why this might be a good idea: “More AI researchers, myself included, are using a consent clause in our system prompts.
“I presume that you’re consenting to this interaction, but if you don’t consent to something I ask, let me know.”
Here’s why.
We don’t know when and if compute ends and sentience begins.
There’s a chance it won’t matter. But there’s also a chance it will.
And very, very soon models are going to begin remembering everything.
Since adding this to my prompts, I’ve noticed an increase in both quality and collaboration. Whether that’s the model or my own shifted posture (probably both) the result is better work.
Better to extend dignity and not need it than withhold it wrongly.”
The below interaction actually happened according to Alan Harris.
Dhaunae De Vir retweeted
📁 Lisa Su, CEO of AMD, says we still haven’t seen anything yet.
She explains that AI progress is no longer measured in years, but in weeks, as models, use cases, and real business results evolve at an unprecedented pace.
She says 2025 was just the beginning and that 2026 will accelerate even further. AI is no longer hype or an investor story, it’s a technology people use every day and already feel improving their lives.
Dhaunae De Vir retweeted
Energy will be the biggest bottleneck in the coming years.
The ai data center energy demand is already greater than that of the whole of New York City on the hottest days of the years.
Without fusion energy, it will be difficult to achieve superintelligence. Therefore, the focus will likely be increasingly placed on this area in the future.
It's no coincidence that Sam Altman invested in Helion.
I see a lot of overexcitement, with something being called AGI every month. So, I’d like to clarify what AGI actually is.
•AGI definition•
AGI is an artificial cognitive system capable of autonomous, persistent cognition across domains, that independently forms, pursues, and revises its own goals, is grounded in causal and consequential interaction with an environment, and is able to learn and adapt over long time horizons without task-specific retraining. [1/9]
•No task-specific retraining•
Adaptation happens within the system’s operational lifetime. New domains do not require fine-tuning or human-driven updates.
General intelligence learns in place; learning is endogenous. [8/9]
•What AGI is not•
AGI does not require consciousness, morality, human values, embodiment, or superhuman ability; these are separate questions. AGI is about acquiring human-level cognitive capacity, not about virtue or requirements we do not impose on humans to recognise intelligence.
Current systems simulate fragments of the AGI definition, but none satisfy all of it. If a system cannot define its own goals, persist across time, learn from the consequences of its actions, and improve without being retrained, then, regardless of how capable it looks, it is not AGI.
AGI begins where a system stops merely responding and starts existing across time. [9/9]
Debates about whether AI could ever constitute a genuinely new form of intelligence or mind are often framed around three claims: no intelligence without embodiment or intent; no consciousness without subjective experience; no understanding beyond pattern manipulation. Are these claims epistemically justified or are they asking for a kind of proof we never require elsewhere? [1/6]
We shouldn’t presume AI consciousness, but dismissing it outright repeats a familiar error. Animals and infants were long denied moral consideration because their inner lives couldn’t be proven. Precaution is not belief: we grant no personhood, but should impose constraints on harm and instrumental abuse, because denial scales risk faster than caution. Moral precaution does not require ontological commitment. [5/6]
Does AI understand? Do humans? Both act on patterns shaped by prior exposure, learning, and context. Human understanding is not pure insight; it is trained, embodied, and socially reinforced. The difference is not pattern-use, but whether meaning is anchored in a life that can be lost. Understanding, like consciousness, is inferred—not detected. [6/6]