Last summer, Hochreiter presented his XLSTM idea to 25 industry representatives at a hotel in Zug am Arlberg at the invitation of the Industrial AI Podcast and Hannover Messe. The group was impressed, as the familiar LLM models are too slow and computationally intensive for industry. However, none of the participants wanted to issue a blank cheque at the time. Hochreiter called for over 100 million euros in the parlour. But a Pierer representative was also part of the round. The company scours the world for innovations for its products, but also for production. This time they probably found what they were looking for at home, at the Rote Wand.

XLSTM is set to become the European alternative to the Transformer architecture that makes Large Language Models such as ChatGPT so powerful. The LSTM algorithm is widely used in industry today - many small and large industrial companies use the algorithm to optimise their production. "XLSTM will become even more powerful," predict insiders who have already been able to take a look at the architecture. The community is waiting for the first paper - it will come, but the focus in Linz is currently on a product and a patent procedure. XLSTM is LSTM plus exponential gating improved by vectorisation, some observers whisper. XLSTM is an autoregressive approach, can abstract and Hochreiter wants to keep it in Europe. "XLSTM is faster, we need less memory and has a linear runtime." What's more, XLSTM can always predict the next word better. This means simpler coding and logic. The technology from Linz also has a memory part and a semantic part. Thanks to this combination, XLSTM can draw better conclusions, explains Hochreiter in an interview.

Find out how they want to scale, when there will be details and the background in the Industrial AI Podcast.