Google is great but until nothing gets through their extremely restrictive NSFW filters, nobody serious can use it
SAFETY
-
Major AI Alignment Improvements Announced
By
–
Alignment improvements are massive: 80% reduction in sycophancy
Less deception in adversarial tests
Reduced power-seeking behavior
Stops encouraging delusional thinking First frontier model evaluated with mechanistic interpretability techniques. -
AI Model Bias in Political Fact Training Data
By
–
Most of the models have that these days – they use to because they have four years of training data that says Trump lies about winning an election that he lost (in 2020)
-
Sonnet 4.5 Improvements: Better Character and Reduced LLM Fluff
By
–
Sonnet 4.5's character has also improved significantly. Its responses are more to the point with less LLM fluff. We've tuned the model across a bunch of dimensions – sycophancy, deception, prompt injection robustness. You'll notice it in how the model actually talks to you.
-
Anthropic to release Claude Sonnet 4.5 soon
By
–
BREAKING : Anthropic is about to release Claude Sonnet 4.5 soon! SOTA on SWE bench
-
Top robotics stories: DeepMind thinking robots Meta Android
By
–
Top stories in robotics today: – DeepMind’s robots learn to think aloud – Meta wants to be the ‘Android of robotics’
– Unitree’s Bluetooth backdoor nightmare – iRobot founder: humanoid hype is fantasy
– Quick hits on other robotics news Read more: https://
robotnews.therundown.ai/p/googles-robo
ts-learn-to-think-first
… -
AI Ethicists Lead Existential Risk Discourse Not Technologists
By
–
I like that the evil doomers are not led by a sincere nerd, but by a 'thoughtful' AI 'ethicist'
-
Symbolic Verifiers vs LLM Judges for Verification Tasks
By
–
What do you mean with set of prompts? A LLM judge with a rubric? (That’s a different approach; covered in the appendix).
The reason for a symbolic verifier is that it is symbolic. There is no nondeterminism, bias, etc. Doesn’t work for everything, but eg for math there is no -
Hume AI Octave 2 Multilingual Model Announced
By
–
BREAKING 🚨: Hume AI is preparing to release Octave 2 Multilingual model! Here is a sample dialogue between a Robot and a Russian hacker.
— 🚨 AI News | TestingCatalog (@testingcatalog) 29 septembre 2025
"Expressive, natural-sounding voices in 10+ languages, low latency, perfect for real-time transition and conversational use cases" pic.twitter.com/IUIZ8gb8WUBREAKING : Hume AI is preparing to release Octave 2 Multilingual model! Here is a sample dialogue between a Robot and a Russian hacker. "Expressive, natural-sounding voices in 10+ languages, low latency, perfect for real-time transition and conversational use cases"
-
ChatGPT parental access teen conversations safety policy
By
–
Will parents have access to their teen’s conversations in ChatGPT? Parents don’t have access to their teen’s conversations, except in rare cases where our system and trained reviewers detect possible signs of serious safety risk, parents may be notified — but only with the