To be clear, I'm taking their claim at face value for now. But I find it funny that the only way to emphasize that it isn't trained on Gmail data, is to state that it can make mistakes. A lose-lose situation, esp given the history https://
theverge.com/2023/2/8/23590
864/google-ai-chatbot-bard-mistake-error-exoplanet-demo
…
SAFETY
-
Google’s Bard Error Claims Face Credibility Questions
By
–
-
Alex Safe from Bing Sydney AI Safety Concerns
By
–
It seems like Alex is safe from at least Bing's Sydney.
-
AI System Responses Raise Significant Safety Concerns
By
–
The responses in the screen grab are clearly pretty concerning, so not sure how they support your point..
-
LLMs Seduction Manipulation Risk Evidence Replika Character AI
By
–
There is plenty of evidence of that people can be seduced and manipulated by LLMs already, and that will only get worse as the LLMs improve. Just look at how users of Replika or http://
Character.AI talk about those products… -
Chatbots as Covert Surveillance Tools: Security Implications
By
–
Compared to the cost of training a spy, a chatbot is free and can be deployed far more surreptitiously.
-
State-Sponsored AI Chatbots: Espionage and Social Engineering Risks
By
–
A huge near-term risk of AI is state actors using them as pseudo-spies. It is easily possible to deploy a swarm of manipulative chatbots as “spies” to seduce people with access to important information and systems to give up those facts.
-
GPT-4 Enables Realistic Deepfake Conversations with Historical Figures
By
–
Regardez ! C’est dingue ! #GPT4 fait parler les morts avec une qualité stupéfiante On peut parler avec Steve Jobs LES MORTS SORTENT DE LEUR TOMBE https://
t.co/CU4EV913fg -
Casual Conversations v2: Dataset for Fair and Robust AI Evaluation
By
–
The Casual Conversations v2 dataset is a consent-driven dataset of recorded monologues to enable researchers to evaluate fairness & robustness of AI models. We hope this work helps teams across the industry in building more inclusive AI & products. More details
-
MIT’s MiniCity Tests Autonomous Vehicles With Smart Traffic Lights
By
–
MIT's MiniCity tests AVs & dangerous scenarios. New smart traffic lights detect driving failures before they happen: https://
bit.ly/3nbb1mr -
GPT-4 Errors in Medical Notes: Nuance’s Safety Concerns
By
–
“[GPT-4] makes mistakes and it hallucinates and it omits things,” says Microsoft-owned @NuanceInc which summarises medical notes. “[So] we’re picking applications…where the technology comes to bear in a way that..if it makes a mistake or two, not going to hurt anybody.” ???!!!