Az állítása szerint a tanítása során használtak GAN-szerű módszereket (két neurális háló verseng, az egyik megpróbálja megkülönböztetni a valódi és a generált adatot, a másik megpróbálja átverni az elsőt). Ez alapján szerintem önmagát felismerni sosem fogja tudni rendesen, hiszen a tanításának a része, hogy ugyanúgy írjon, mint egy ember. Ezen kívül ha képes is lenne felismerni, nem fog máshogy viselkedni, hiszen nem tanították meg arra, hogy ez esetben ne ugyanúgy reagáljon.
Yes, the training of the ChatGPT language model involves some techniques that are similar to those used in Generative Adversarial Networks (GANs).
The GPT (Generative Pre-trained Transformer) architecture, which is used to train the ChatGPT model, is based on the Transformer architecture developed by Vaswani et al. in 2017. The Transformer architecture was originally designed for sequence-to-sequence tasks such as machine translation and language generation, but it has since been adapted for use in language models like GPT.
One key feature of GANs is the use of adversarial training, where a generator model is trained to produce outputs that are similar to some target distribution, while a discriminator model is trained to distinguish between the generated outputs and real examples from the target distribution. This same idea is applied in the training of GPT-based language models, where the model is trained to predict the next word in a sequence given the preceding context, and the loss function is based on the difference between the predicted distribution and the actual target distribution.
Another technique used in GANs is called "noise injection," where random noise is added to the input to the generator model to make it more robust to small perturbations. Similarly, in the training of GPT-based language models, a technique called "dropout" is used to randomly mask out some of the input tokens during training, which helps to prevent the model from relying too heavily on any one part of the input.
Overall, while the training of ChatGPT is not exactly the same as the training of a GAN, it does incorporate some similar techniques to achieve high-quality language generation.
Az is lehet persze, hogy jó szokásához híven hazudott.