Understanding and preventing misalignment generalization Recent work has shown that a language model trained to produce insecure computer code can become broadly “misaligned.” This surprising effect is called “emergent misalignment.” We studied why this happens. Through this
SAFETY
-
Superintelligence: Supervised or Reinforcement Learning Path
By
–
# Superintelligence from first principles some have said that superintelligence is "an engineering problem"; that we know how to build it now. but there are only two ways to learn: supervised and reinforcement learning. which one will get us there? i wrote about this:
-
Evaluating Post-hoc Explanations: Benchmarking XAI Methods for Governance
By
–
Learn: ✅ Why post-hoc explanations often fall short without evaluation ✅ How to benchmark SHAP, LIME, DLBacktrace, & more across data modalities ✅ Integrating xai_evals into governance, audit & compliance workflows 💬 Stay for a live Q&A with Pratinav
-
Human Reasoning Beyond Token Prediction: Mental Models
By
–
Human reasoning is not based on auto-regressive discrete symbol (token) prediction.
It is based on the manipulation of mental models in continuous representations spaces.
It is based on *searching* for a set of manipulations of this model to arrive at a particular result. -
Fumbling humanity at the brink of superintelligence
By
–
Would be the ultimate shame if we fumbled this whole humanity thing this close to achieving superintelligence.
-
LLM Weaknesses: Safe Integration and Mitigation Techniques
By
–
These weaknesses are crucial to understand so you can know where you can build with and use LLM tools safely and appropriately in your workflows and what techniques you can use to address these issues.
-
Gemini 2.5: Flash Lite, Pro Deep Think, Pro GA
By
–
Gemini 2.5 Flash Lite?
Gemini 2.5 Pro Deep Think?
Gemini 2.5 Pro GA? The next leap -
Power Concentration in AI: Rebutting Eugenic Ideologies
By
–
We have the equivalent of the scientologists running the field of AI with untold amounts of money and power and we're the ones who have to do point by point rebuttals of their eugenicists dreams.
-
Medical Policy Enforcement: Implicit Rules Shape Healthcare Professional Conduct
By
–
For example, when X was “we will yank the license of, destroy the reputation of, and if possible prosecute any medical professional who delivers a dose out of order”, which was *definitely policy until it wasn’t* but which *shaped actions later, too.*
-
Security Risks of Local LLM Agents vs Web-Based Interfaces
By
–
I should clarify that the risk is highest if you're running local LLM agents (e.g. Cursor, Claude Code, etc.). If you're just talking to an LLM on a website (e.g. ChatGPT), the risk is much lower *unless* you start turning on Connectors. For example I just saw ChatGPT is adding