"scaling sparse autoencoders has been a major priority of the Anthropic interpretability team, and we're pleased to report extracting high-quality features from Claude 3 Sonnet, 1 Anthropic's medium-sized production model.
We find a diversity of highly abstract features. They both respond to and behaviorally cause abstract behaviors. Examples of features we find include features for famous people, features for countries and cities, and features tracking type signatures in code. Many features are multilingual (responding to the same concept across languages) and multimodal (responding to the same concept in both text and images), as well as encompassing both abstract and concrete instantiations of the same idea (such as code with security vulnerabilities, and abstract discussion of security vulnerabilities)."
We find a diversity of highly abstract features. They both respond to and behaviorally cause abstract behaviors. Examples of features we find include features for famous people, features for countries and cities, and features tracking type signatures in code. Many features are multilingual (responding to the same concept across languages) and multimodal (responding to the same concept in both text and images), as well as encompassing both abstract and concrete instantiations of the same idea (such as code with security vulnerabilities, and abstract discussion of security vulnerabilities)."
"...a feature that activates when Claude reads a scam email (this presumably supports the model’s ability to recognize such emails and warn you not to respond to them). Normally, if one asks Claude to generate a scam email, it will refuse to do so. But when we ask the same question with the feature artificially activated sufficiently strongly, this overcomes Claude's harmlessness training and it responds by drafting a scam email."
This review article discusses the concept of entropy in statistical physics and its role as both a tool for inference and a measure of time irreversibility. It highlights the developments in stochastic thermodynamics and the principle of maximum caliber, emphasizing the importance of cross-talk among researchers in disparate fields.
- Life's evolution on Earth began with single-celled organisms and evolved into complex life forms through environmental factors and extinction events, culminating in the emergence of mammals and ultimately our own species.
- The article highlights the significance of our species, Homo sapiens, within the grand scheme of things, occurring within the last 300,000 years.
- Despite the limited time our species exists compared to the vastness of cosmic time, it encourages us to consider the potential for future life forms
- The article highlights the significance of our species, Homo sapiens, within the grand scheme of things, occurring within the last 300,000 years.
- Despite the limited time our species exists compared to the vastness of cosmic time, it encourages us to consider the potential for future life forms
New study on mice decision-making reveals that choice is not a singular moment but a reflection of the brain’s preexisting state.
The research, using Buridan’s Assay, suggests that the mice’s brain constantly broadcasts its goal, even before options are available, with patterns of neuron activity predicting choice.
Hunger and thirst don’t directly drive behavior; instead, they modulate the brain’s goal-setting, with an element of randomness causing switches between needs, ensuring both are met over tim
The research, using Buridan’s Assay, suggests that the mice’s brain constantly broadcasts its goal, even before options are available, with patterns of neuron activity predicting choice.
Hunger and thirst don’t directly drive behavior; instead, they modulate the brain’s goal-setting, with an element of randomness causing switches between needs, ensuring both are met over tim
