A Message for AI Developers: Recently, I needed to update some libraries in Python, and in the process, TensorFlow stopped working through the GPU.

I am currently tackling a task in NLP, where I teach a computer to recognise various types of insults, slurs, coarse expressions, etc. It's about 70 MB of text for training. To help the computer understand this a bit better, I trained it on 2 million Czech words from Wikipedia and other sources – word embedding. In short, that's quite a lot of information, and one epoch of training took… two and a half hours on the CPU. Ten epochs then felt like an eternity – an entire day. So, I decided to get the GPU up and running again, and do you know how long one epoch takes? Just 6 seconds! I can definitely get used to a 1500x speedup.

I have a dream. To acquire a GPU with Tensor cores – the RTX 2080Ti graphics card (or at least a cheaper GeForce RTX 2070). The card is set to be released on 20 September – you can find the GPU reviews here: https://www.alza.cz/nvidia-turing-recenze-a-testy