· ai
Why “Smart” AI Still Makes Dumb Decisions
Intelligence without constraints is just speed When an AI system makes a bad decision, we usually blame the model. But most of the time, the model did exactly...
-
- · ai
I Trained Probes to Catch AI Models Sandbagging
TL;DR: I extracted “sandbagging directions” from three open‑weight models and trained linear probes that detect sandbagging intent with 90‑96 % accuracy. The mo...
- · ai
Detecting Adversarial Samples from Artifacts
Overview Many AI systems can be fooled by tiny, almost invisible edits to images that cause them to give incorrect answers. Researchers have discovered a simpl...
- · ai
Dario Amodei - resigns from openai & built AI safety
Dario Amodei: From OpenAI Researcher to Anthropic CEO An in‑depth, human‑first look at the researcher who walked out of OpenAI and helped build Anthropic — why...
- · ai
On Evaluating Adversarial Robustness
Why some AI defenses fail — a simple look at testing and safety People build systems that learn from data, but small tricky changes can make them fail. Researc...
- · ai
Scaling Language Models: Methods, Analysis & Insights from Training Gopher
Researchers built a very large language system called Gopher to see what happens when computers read lots and lots of writing. As the models grew in scale, they...
- · ai
Beyond the Imitation Game: Quantifying and extrapolating the capabilities oflanguage models
Summary - Researchers assembled BIG-bench, a collection of 204 tasks created by many contributors to evaluate current and future language model capabilities. -...
- · ai
Autonomously navigating the real world: lessons from the PG&E outage
Article URL: https://waymo.com/blog/2025/12/autonomously-navigating-the-real-world Comments URL: https://news.ycombinator.com/item?id=46371730 Points: 15 Commen...
- · ai
Understanding Vibe Proving
How to make LLMs reason with verifiable, step-by-step logic Part 1 The post Understanding Vibe Proving appeared first on Towards Data Science....
- · ai
What is MLSecOps?
What is MLSecOps? MLSecOps is a framework that integrates security practices throughout the entire machine learning lifecycle, much like DevSecOps does for sof...
- · ai
Continuously hardening ChatGPT Atlas against prompt injection
OpenAI is strengthening ChatGPT Atlas against prompt injection attacks using automated red teaming trained with reinforcement learning. This proactive discover-...
- · ai
Why AI safety should be enforced structurally, not trained in
Most current AI safety work assumes an unsafe system and tries to train better behavior into it. - We add more data. - We add more constraints. - We add more fi...