American company Goodfire recently launched the Silico platform for public use, aiming to use machine-understandable techniques to analyze the internal workings of artificial intelligence. The platform integrates various tools, including mapping model activation patterns, tracking weight changes before and after training, and modifying specific weights to observe the impact on output. It also features an AI agent that helps users independently develop experimental plans. Goodfire has also launched a new funding program, offering free Silico access worth $1 million to academic and non-profit researchers in the field of explainability. Eric Ho, founder and CEO of Goodfire, said that viewing models as black boxes is a human choice, and using these tools can reveal the internal structure of models and the reasons for their behavior. Currently, this technology has been applied in the medical field, for example, to help British company Prima Mente analyze the predictive logic of its Pleiades epigenomic base model in detecting Alzheimer’s disease.
Goodfire recently launched the Silico platform for the public, aiming to analyze the internal workings of AI through mechanically interpretable technologies. The company also launched a new funding program, offering free Silico access worth $1 million to academic and non-profit researchers specializing in interpretability. Silico integrates various tools, including mapping model activation patterns, tracking weight changes before and after training, and modifying specific weights to observe the impact on output. It also includes an AI agent that helps users independently develop experimental plans. Eric Ho, founder and CEO of Goodfire, said that viewing models as black boxes is a human choice, and using these tools can reveal the internal structure of models and the reasons for their behavior. Currently, this technology has been applied in the medical field, for example, helping Prima Mente in the UK to analyze the predictive logic of its Pleiades epigenomic base model in detecting Alzheimer’s disease.