No part of AI is even close to being solved. This includes vision, language, reasoning, and agents. Massive datasets + benchmarks = illusion of intelligence. For example:
SAFETY
-
Reward Functions in Intelligent Agency Are Not Arbitrary
By
–
In the context of intelligent agency, it's unsatisfying to treat reward functions as given or arbitrary. Reward functions are chosen within a mind, and they are instrumental to evolutionary success: if you pick the wrong reward function, you won't stick around pic.twitter.com/cAqJeLBRrU
— Joscha Bach (@Plinz) 25 mars 2026In the context of intelligent agency, it's unsatisfying to treat reward functions as given or arbitrary. Reward functions are chosen within a mind, and they are instrumental to evolutionary success: if you pick the wrong reward function, you won't stick around
-
AI Toy Safety Standards Needed to Protect Young Children
By
–
Report calls for #AI toy safety standards to protect young children
by @Cambridge_Uni @TechXplore_com Learn more: https://
bit.ly/41miWxW #ArtificialIntelligence #MachineLearning #ML -
Distributing AGI Benefits: A Path Beyond Technical Achievement
By
–
path to AGI is futile without distributing the benefits along the way
-
ARC-AGI-3 Benchmark Results and Saturation Timeline
By
–
ARC-AGI-3 benchmark is here It took from $2 to $9k for frontier models to complete the task at 0.2-0.3% acheivemnt. How soon would you expect it to get saturated?
-
Critique of AI Safety Group’s Strategy Against AGI Risks
By
–
ninhuem: nada portuguesas: TU TENTO JA FRANCESINHAAAAA??????!!!!!!!!! PVVVV
-
AI Scientist: Critical examination of autonomous research tool
By
–
AI Scientist, an autonomous research tool, first released in 2024, has now undergone peer review, highlighting its strengths and limitations go.nature.com/4t8x9uo [Translated from EN to English]
-
Frontier AI Achievement: Low Probability Assessment and Reasoning
By
–
The topics I brought up were more controversial than I thought, but it's reaching the right people and the recent Composer 2 tech report released since is encouraging too. My underlying opinion has not changed though. Here's my reasoning: – Low chance of achieving frontier:
-
AI-Powered Fraud Threats Outpace Organizational Response Capabilities
By
–
Fraudsters don’t have governance committees or budget cycles. They just act. New research by @TheACFE + SAS asks whether organizations can move fast enough to keep up with #deepfakes & other AI-charged #fraud threats. Spoiler: most can’t yet.
-
OpenAI Explains Model Spec and AI Model Behavior
By
–
The more AI can do, the more we need to ask what it should and shouldn't do. OpenAI researcher @w01fe joins host @AndrewMayne to explore the Model Spec, the public framework that defines how models are intended to behave. They break down how it works in practice, from the chain of command that resolves conflicting instructions to the way it evolves over time through real-world use, feedback, and new model capabilities. [Translated from EN to English]