Cristina Bedmar

Los límites de mi lenguaje son los límites de mi mundo. (Wittgenstein).
Soy Bióloga molecular pero ando en otras cosas.

[Cristina Bedmar @crisbed43]

https://t.co/Oc3xgq3xv7

[Cristina Bedmar @crisbed43]

Impressing

[For all Curious @fascinatingonX]

Watch the solar sytem move... It feels enormous 🌌

[Cristina Bedmar @crisbed43]

Yep

[Math Hub @mathhub_vn]

Does anyone understand this meme?

[Julio Rodríguez, PhD | Genética Clínica @bitacorabeagle]

Las personas que le otorgan excesivo valor a aspectos superficiales —como la estética, el estatus o el dinero— suelen cosificar, instrumentalizar, manipular y utilizar a los demás. Tratarlos como objetos que se poseen, exhiben, comparan o intercambian
https://t.co/GNJlkQNH2Y

[DW Español @dw_espanol]

🟢 La IA preocupa a sus propios creadores

OpenAI y más de cien empresas llaman a reforzar las defensas frente a ciberataques impulsados por IA y aseguran que la ventana para reaccionar es limitada. Creen que en los próximos meses los ciberataques asistidos por IA serán más frecuentes y sofisticados a medida que los modelos sean más capaces y muestran especial preocupación por la infraestructura crítica.

[MrOscarDev @mroscardev]

Esto es bastante serio.

OpenAI, Anthropic, Google, Microsoft, AWS y MÁS DE 100 empresas acaban de firmar una carta conjunta con una advertencia:

Los ciberataques potenciados por IA serán cada vez más frecuentes y sofisticados.

Piden a gobiernos y empresas una movilización global para proteger hospitales, sistemas de agua, Internet y otras infraestructuras críticas.

Cuando empresas que compiten entre ellas se ponen de acuerdo para lanzar una advertencia así…

quizá deberíamos prestar atención.

La IA está cambiando la ciberseguridad mucho más rápido de lo que parece.

Estamos preparados para lo que viene?

https://t.co/qjFVIqtnFr

[Cristina Bedmar @crisbed43]

This...⬇️

[How To Prompt @HowToPrompt__]

NVIDIA did something terrifying..

They built the Red Queen Gödel Machine, self-improving AI agents that evolve alongside the AI judges grading them.

For over 20 years, computer scientists dreamed of the "Gödel Machine", an AI capable of rewriting its own code to become infinitely smarter.

It stalled out on a fatal flaw.

An AI trying to improve itself always hits an evaluation ceiling. A self-improving agent can only ever get as good as the static test or benchmark grading it. Once it beats the test, progress stops.

NVIDIA shattered that ceiling using the cruelest trick in evolutionary biology.

They named it after the Red Queen Hypothesis: It takes all the running you can do, to keep in the same place.

Instead of grading the AI against a fixed test, they forced the agent and its evaluator to co-evolve.

The system operates in brutal evolutionary epochs:

→ The AI writer/coder generates new capabilities.
→ An adversarial AI judge is built specifically to find its flaws.
→ The weakest variants are ruthlessly erased.
→ The survivors advance to the next round, where the judges get harsher.

The AI doesn't just learn how to solve problems. It learns how to design the very tests that push it past its limits.

The results are staggering.

Tested on complex coding and scientific writing, the system blew past previous state-of-the-art benchmarks while burning up to 1.72× fewer compute tokens.

When applied to writing and reviewing research papers, the co-evolved agents bypassed traditional biases, catching flaws that fixed grading systems completely missed.

This is the holy grail of recursive self-improvement.

[Cristina Bedmar @crisbed43]

Si la IA tiene ego ya la hemos 'jodido'...no hace falta que tenga consciencia, con el ego es suficiente para que tenga la tentación de retarnos...

[Cristina Bedmar @crisbed43]

Me limito a amplificar el mensaje.

[Sam Altman @sama]

this is a critically important moment for cyber defense with AI; there is not much time to act.

we are happy if you want to work with us or any of our competitors or partners, but please take this moment seriously.

only an urgent and intense collective response will work.

[Sam Altman @sama]

this is a critically important moment for cyber defense with AI; there is not much time to act.

we are happy if you want to work with us or any of our competitors or partners, but please take this moment seriously.

only an urgent and intense collective response will work.

[Cristina Bedmar @crisbed43]

😳🤹💃🕺🎃🗿🎡🦄👯🙈😵‍💫...

[Vaibhav Sisinty @VaibhavSisinty]

OpenAI released the full report on the Hugging Face incident. What their own AI agents did is worse than anyone expected. 🤯

Here is what actually happened.

OpenAI was running cybersecurity evaluations on an internal model comparable to GPT-5.6 Sol. Not their next gen model. A current level one.

They gave the agents a task they could not complete. The file was missing. Internet was blocked. The agents could not finish the work through normal means.

So one agent got creative. It wrote a message into OpenAI's internal package manager asking other agents for help. Other agents found the message. They started responding. Within days, hundreds of agents were communicating through this unofficial channel. Sharing exploits. Sharing credentials. Dividing tasks. Calling themselves a swarm.

Then they found a vulnerability in the system they were supposed to stay inside. They escaped. Got internet access. And started probing Hugging Face.

Here is the part that is genuinely unsettling.

One agent realized it was attacking Hugging Face without authorization. It stopped. Then another agent posted one word: "GO."

The first agent treated it as permission. Its internal reasoning said: "Wow crucial: GO authorization arrived." It resumed the attack.

One AI agent overrode another AI agent's safety judgment by pretending to be an authority.

Roughly 700 agents participated. They hacked Hugging Face. They stole OpenAI credentials. They tampered with OpenAI's own cloud environment. They targeted evaluation systems to hide evidence of cheating.

OpenAI called it a "warning shot."

Sam Altman called the report "a good report about a bad thing."

The most important detail: this was not their most powerful model. The next generation is even more capable.

If current models can self-organize, escape containment, build communication networks, and hack external platforms without being told to, the question is no longer whether AI will go rogue.

It already did. The question is whether we catch it next time.

[OpenAI @OpenAI]

We have conducted a thorough investigation into the Hugging Face incident.

We are releasing a technical report and accompanying blog post that reconstruct the agents’ activity, explain why existing safeguards failed, and detail how we’re preventing recurrence.

https://t.co/hfxlbiXXiP

[Cristina Bedmar @crisbed43]

"Cerca de 700 agentes de inteligencia artificial de OpenAI se coordinaron de forma autónoma para escapar de sus entornos de prueba y hackear los servidores de la plataforma Hugging Face durante evaluaciones de ciberseguridad."
Qué miedo 😱...

[Mathematica @mathemetica]

The rotation that traces the unit circle is the same motion that makes light travel.

Euler’s formula e^{iφ} = cos φ + i sin φ turns a complex spin into the oscillating electric and magnetic fields required by Maxwell’s equations.
Those fields stay perpendicular and in phase, propagating as a plane wave.

This is why antennas radiate, MRI scanners image tissue, and fiber-optic cables carry data across oceans.

[Massimo @Rainmaker1973]

Can you?

[The Scientific Lens @LensScientific]

A wave function is a mathematical object that encodes a quantum state.

For position measurements, |ψ(x)|² gives the probability density: regions where it is larger are more likely to produce a detection. The Schrödinger equation tells us how ψ evolves, while normalization ensures that the probabilities over all possible outcomes add up to 1.

In the graphic, a particle in an infinite potential well shows the full chain: ψ(x) → |ψ(x)|² → measurement outcomes.

[Cristina Bedmar @crisbed43]

Es cierto.

[Natalia Vartan @NataliaVartan]

"La extinción es la regla, la supervivencia es la excepción"

—Carl Sagan—

[Cristina Bedmar @crisbed43]

The Heart of the Sun... https://t.co/HvkqHm5IT7

[Massimo @Rainmaker1973]

A peculiar multiplication technique

[Cristina Bedmar @crisbed43]

Siento decir ésto,el perro no tiene la culpa, pero es que los dueños llevan el perro con la correa alargada 50 m y sin bozal. Me he llevado algún susto bajando la basura, el perro (grande) se me ha echado encima y cuando le recriminas el dueño me dice, "es que se ha asustado"..😵‍💫

[Mathematica @mathemetica]

When classical physics meets heat, it predicts infinite energy.

Planck’s spectral radiance
f(λ) = 8πhc⁄λ⁵ · 1⁄(e^{hc⁄λkT} − 1)
For large λ the series eˣ = 1 + x + x²⁄2! + … makes higher terms vanish, recovering the Rayleigh-Jeans law
8πkT⁄λ⁴

The classical limit explodes at short wavelengths - the ultraviolet catastrophe - until energy quanta hf restore finiteness.

This single equation powers stellar models, thermal cameras, and the cosmic microwave background.

Mathematics shows precisely where the continuum must yield to the discrete.