Key finding: AI models can "transcend" their training data, playing chess at a level significantly higher than the players they learned from. This challenges the notion that AI is limited by the quality of its training data
SAFETY
-
Superintelligence Control Delusions Criticized
By
–
He’s deluded about both how far we are from superintelligence and how to control it.
-
Ethical Concerns About Robot Mistreatment in AI Development
By
–
Impresionante habilidades! Pero… ¿no se nos está yendo de las manos lo de maltratar robots?pic.twitter.com/OFWxF7Dab4
— Carlos Santana (@DotCSV) 20 juin 2024Impressive skills! But… aren't we getting a bit carried away with mistreating robots?
-
Verifying Answers from Supposedly Brilliant Alignment Strategies
By
–
I mean, how do you verify the answers of something that is supposedly giving you brilliant alignment strategies?
-
Good Engineering Neglected: Safety Compromised in AI Development
By
–
A lot (most) of my Medium feed is chaff, but this one gets at why good engineering, from airplanes to AI apps, is not respected, and instead we get lower safety and [my words] even more load on the end user to accommodate AAS (artificial asinine intelligence) thrust upon us.
-
Anthropic announces Claude 3.5 Haiku and Opus releases
By
–
To complete the Claude 3.5 model family, we'll be releasing Claude 3.5 Haiku and Claude 3.5 Opus later this year. In addition, we're developing new modalities and features for businesses, alongside rigorous safety testing. Read more:
-
From Intelligence to Super Intelligence: Beyond Task Automation
By
–
Intelligence: what is this, what is that, write this, order that, book this, cancel that Super Intelligence: make me happy, solve my problems, take care of me
-
AI Control: Humanity Collective vs AI Vigilantes
By
–
If AI exceeding human intelligence is dangerous, who would you rather have controlling it: the collective intelligence of all humanity or a bunch of AI vigilantes?
-
What Does Safe Superintelligence Mean to You?
By
–
What do you think of when you read "Safe Superintelligence"?
