David P

Ingeniero informático por la URJC. Me gusta romper cosas

[David P @pdparla]

Decentralized, write up del reto de @fwhibbit_blog en @Honey_SEC https://t.co/Ni2U9LIlpH

[Binfeng Xu @billxbf]

Excited to release 🌟Polar🌟, our Agent RL rollout infra for real-world harnesses. Be it Codex, Claude Code, OpenClaw, Hermes, or your self-made ones 🔥 -- Polar takes your harnesses directly as training environments without code change.

Find a problem, design the harness, and train your own agents! 🧵

[Jack Lindsey @Jack_W_Lindsey]

Before limited-releasing Claude Mythos Preview, we investigated its internal mechanisms with interpretability techniques. We found it exhibited notably sophisticated (and often unspoken) strategic thinking and situational awareness, at times in service of unwanted actions. (1/14)

[Boris Cherny @bcherny]

Reflecting on what engineers love about Claude Code, one thing that jumps out is its customizability: hooks, plugins, LSPs, MCPs, skills, effort, custom agents, status lines, output styles, etc.

Every engineer uses their tools differently. We built Claude Code from the ground up to not just have great defaults, but to also be incredibly customizable. This is a reason why developers fall in love with the product, and why Claude Code's growth continues to accelerate.

I wanted to share a few ways we're seeing people and teams customize their Claudes.

[Boris Cherny @bcherny]

I'm Boris and I created Claude Code. Lots of people have asked how I use Claude Code, so I wanted to show off my setup a bit.

My setup might be surprisingly vanilla! Claude Code works great out of the box, so I personally don't customize it much. There is no one correct way to use Claude Code: we intentionally build it in a way that you can use it, customize it, and hack it however you like. Each person on the Claude Code team uses it very differently.

So, here goes.

[Weiwei Sun @sunweiwei12]

AI agents are supposed to collaborate with us to solve real-world problems, but can they really? Even the most advanced models can still give us frustrating moments when working with them deeply.

We argue that real-world deployment requires more than productivity (e.g., task accuracy); agents must also be proactive in communication and personalized to individual user preferences.

Our new work introduces PPP, a Productive, Proactive, and Personalized optimization framework that explicitly trains LLM agents for effective human interaction.

🚀PPP achieves significant gains in complex, real-world agent–user scenarios (software engineering and deep research), outperforming even GPT-5 on both tasks with initially vague user instructions.

[Sam Rodriques @SGRodriques]

Try Kosmos on our new platform, here: https://t.co/FSJH9zPlef

Read our technical report: https://t.co/BkAzl0l00S

Read more about Kosmos on our blog: https://t.co/8arz5iokgG

Finally, read more about Edison Scientific here: https://t.co/dUWbMNYUWd

We can’t wait to see how you all use it (and break it)!

[Miguel Ángel Durán @midudev]

Este repositorio recopila todo lo que necesitas para trabajar con IA y LLM en tus proyectos.

+120 bibliotecas organizadas para cada fase del desarrollo:

→ Entrenar, ajustar y evaluar modelos
→ Desplegar aplicaciones con LLMs y RAG
→ Ejecutar modelos de forma rápida y escalable
→ Extracción de datos, crawlers y scraper
→ Crear agentes autónomos basados en LLMs
→ Optimización de prompts y seguridad

https://t.co/TpMR10zoBp

[Avi Chawla @_avichawla]

You're in an ML Engineer interview at Apple.

The interviewer asks:

"Two models are 88% accurate.

- Model A is 89% confident.
- Model B is 99% confident.

Which one would you pick?"

You: "Any would work since both have same accuracy."

Interview over.

Here's what you missed:

Modern neural networks can be misleading.

They are overconfident in their predictions.

For instance, I saw an experiment that used the CIFAR-100 dataset to compare LeNet with ResNet.

LeNet produced:
- Accuracy = ~0.55
- Average confidence = ~0.54

ResNet produced:
- Accuracy = ~0.7
- Average confidence = ~0.9

Despite being more accurate, the ResNet model is overconfident in its predictions. While the model thinks it's 90% confident in its predictions, in reality, it only turns out to be 70% accurate.

Calibration solves this.

A model is calibrated if the predicted probabilities align with the actual outcomes.

For instance, say a model predicts an event with a 70% probability. Then, ideally, out of 100 such predictions, ~70 should result in the event.

Handling this is important because the model will be used in decision-making.

In fact, an overly confident that is not equally accurate model can be highly misleading.

To exemplify, say a government hospital wants to conduct an expensive medical test on patients.

To ensure that the govt. funding is used optimally, a reliable probability estimate can help the doctors make this decision.

If the model isn't calibrated, it will produce overly confident predictions.

Reliability Diagrams are a visual way to inspect how well the model is currently calibrated.

More specifically, this diagram plots the expected sample accuracy as a function of the corresponding confidence value (softmax) output by the model.

If the model is perfectly calibrated, then the diagram should look like the identity function.

That said, it is often also useful to compute a scalar value that measures the amount of miscalibration, called expected calibration error (ECE).

One way to approximate the expected calibration error shown above is by partitioning predictions into equally spaced bins and taking a weighted average of the bins’ accuracy/confidence difference.

These are some common techniques to calibrate ML models:

> For binary classification models:

- Histogram binning
- Isotonic regression
- Platt scaling

> For multiclass classification models:
- Binning methods
- Matrix and vector scaling

👉 If you care about probabilities and both models are operationally similar, which model would you prefer?

[vx-underground @vxunderground]

Weird stuff going on. This is a CRAZY anime arc. I beg you to read this post. This shit is crazy.

Check this shit out

June 16th, 2025: @phrack reports suspected offensive state-sponsored activity from China and/or North Korea targeting South Korea. They notify KR-CERT (Defense Counterintelligence Command).

*In other words, evidence of China and/or North Korea successfully hacking companies in South Korea.

June 26th, 2025: South Korean government responds

July 17th, 2025: Phrack notifies KISA, Ministry of Unification, LG Uplus Corp, KR-CERT about offensive operations from China and/or North Korea

August 15th, 2025: Phrack e-mails terminated from Proton.

September 9th, 2025: Everyone starts screaming at Proton on social myself (us included). Proton apologizes and re-instates Phracks Proton e-mail

... then the twist

September 24th, 2025: South Korean parliament launches an investigation into the allegations against China and/or North Korea. They want to investigate the companies which were compromised

September 25th, 2025: South Korean government says they are going to perform an on-site inspection on several of the alleged compromised facilities

September 26th, 2025: A government data center is burned to the ground. 96 servers destroyed. All evidence gone. This includes evidence of China and/or North Korean offensive operations.

September 27th, 2025: Server fire reported to be caused by a Lithium-ion battery. The batteries that caused the fire were made by one of the companies which was compromised by China and/or North Korea

October 2nd, 2025: Another location which was believed to be compromised by China and/or North Korea is burned to the ground. All evidence gone.

October 2nd, 2025: A South Korean government official who was appointed to manage these inspections and overviews commits suicide

What the fuck is going on? How did a simple Lithium-ion battery burn an entire data center to the ground? Is it weird that another massive data center burned to the ground a few days later? Why did these fires only impact servers which were believed to be hacked by China and/or North Korea? Why are government officials killing themselves? Why the fuck is this not getting more attention? Why does my tummy hurt?

Find out next time on Dragon Ball Z

[Juan Carlos Amez @juankaamez]

Personotecnia: el CPS del Factor X.¿Una turra para gobernarlos a todos/as/IA… o una forma de traducir el caos humano en coherencia funcional?
En el hilo de hoy trataremos una idea peligrosa, casi herética: la Personotecnia como CPS del Factor X.
@Recuenco

[Jaime Gómez-Obregón @JaimeObregon]

Me pusieron una multa de la zona azul.

45 euros.

Intenté pagarla en la sede electrónica de mi ayuntamiento.

Por internet.

No imaginaba lo que se me venía encima…

[The Hacker News @TheHackersNews]

🔥 A $50 hardware hack just broke Intel SGX & AMD SEV-SNP—the backbone of confidential cloud computing.

Researchers built a cheap DDR4 interposer that slips past trust checks, then flips a switch to rewrite encrypted memory on the fly.

The kicker? Fixing it would require redesigning memory encryption itself.

Details → https://t.co/PuqRFyCeL5

[Sam Brannen @sam_brannen]

#JUnit 6.0.0 is released!

✨ Java 17 and Kotlin 2.2 baseline
🌄 JSpecify nullability annotations
🛫 Integrated JFR support
🚟 Kotlin suspend function support
🛑 Support for cancelling test execution
🧹 Removal of deprecated APIs

https://t.co/GLyIcHgDIw

[Akshay 🚀 @akshay_pachaar]

This simple technique can scale training from 1-1000+ GPUs.

- OpenAI uses it to train GPT models
- Google uses it in their TPUs to train Gemini
- Meta uses it to train Llamas on massive GPU clusters

Let's learn how to sync GPUs in multi-GPU training (with visuals):

[0xedh @0xedh]

Last week, @Qm9yamFN and I spoke at @defcon. A milestone after years of following the conference. Our talk presents how Secure Boot, WPBT, and vulnerable drivers can be abused in modern bootkits and persistence mechanisms.

Code & PoCs: https://t.co/WelZudMwmg

[Joseph Thacker @rez0__]

Basically 7 insanely smart teams built Hackbots that fuzz binaries for vulns….

And they’re all open source. Zero exaggeration: this is a treasure trove for anyone building agents.

[David P @pdparla]

@antonosika @Cloudflare Be careful, in Spain @Cloudflare doesn't work on football days. Thanks to @LaLiga for that.

[Ars Technica @arstechnica]

Microsoft researchers create super‑efficient AI that uses up to 96% less energy https://t.co/1e37xxsQ85

[@levelsio @levelsio]

It's so exciting to be in tech right now

He's 100% right, it's like the early days of the internet around 1995 again

Everything was new and you had endless opportunities, you could build the first web browser, or mail app, or video streaming tech, etc

But instead of internet, it's now the early days of AI and 1995 is 2025

Most people don't realize it but this is a VERY special time with SO many opportunities right now, all you need is to build

And times like these only happen every ~30 years if at all, jump on it!

[Corta @somoscorta]

"Si cogemos todos, en un año somos Suecia"

La nueva publicidad de Tulipán sobre "el crecimiento de las industrias del país que impulsa el sexo".