Hacker News Reader

That's something that Microsoft research wrote two decades ago. And those results were well known in the NLP community.

Example: https://www.microsoft.com/en-us/research/wp-content/uploads/... (Michele Banko published a few similar papers on that topic)

View on HN

There was no hugely scalable approach before transformers, RNNs, the previous SOTA, were notoriously bad at scaling.

Yes, we needed clever ideas from scientists to make them scale. In fact, we still need clever ideas to make them scale because the current architectures still have all sorts of problems with length and efficiency.

Reply on news.ycombinator.com