According to @ch402, adding a repeated data point p=[1,1,1,0,0,0,0...] to a transcoder’s training set makes the model learn a special feature that memorizes that exact point, as shown in the shared visualization; this highlights concrete memorization behavior from duplicated samples in training data. Source: https://twitter.com/ch402/status/1953678105794818252 According to @ch402, this post is a technical demonstration and does not reference cryptocurrencies, tokens, or market impact, so it offers no immediate trading signal beyond underscoring data-quality risks in AI model training. Source: https://twitter.com/ch402/status/1953678105794818252