what’s the difference between a single neural network and multiple neural networks? are BERTs multiple neural networks working together in concert, if you count the encoder and decoder separately? the different layers of a MoE? certainly VLMs count.
I don’t think so. Think about the training process. All the nodes in those neural networks were trained on sets of data together. While certain sets of nodes may be more heavily responsible for the word, “pineapple”, for example, those subsets cannot function outside the context of the whole network.
Now, what if you had an LLM that hallucinates, and you trained a second neural network whose only job was to fact-check the output of the first neural network? Those are 2 neural networks trained on separate data, for separate purposes that work in tandem to provide accurate responses.
what’s the difference between a single neural network and multiple neural networks? are BERTs multiple neural networks working together in concert, if you count the encoder and decoder separately? the different layers of a MoE? certainly VLMs count.
I don’t think so. Think about the training process. All the nodes in those neural networks were trained on sets of data together. While certain sets of nodes may be more heavily responsible for the word, “pineapple”, for example, those subsets cannot function outside the context of the whole network.
Now, what if you had an LLM that hallucinates, and you trained a second neural network whose only job was to fact-check the output of the first neural network? Those are 2 neural networks trained on separate data, for separate purposes that work in tandem to provide accurate responses.