Ask HN: When do you think LLM capacity will reach its ceiling?

When do you think { compiler | http | von Neumann architecture } capability will reach its ceiling?

Unless replaced by something far more capable, refinements and incremental improvements will continue to be applied to LLMs. They will become less expensive and more specialized. Machine learning and neural networks are part of the software stack. LLMs will continue to grow and improve for the foreseeable future.

AI is the latest downstream consequence of the 15 order of magnitude increase in global digital compute since 1946. If compute increases into the foreseeable future; so too will the capability of AI.

IMO we might not hit a hard ceiling on capacity, but we will likely hit a wall with data quality

As human-generated nutritional data runs dry, LLMs will end up eating their own garbage ?

Yeah, pretty much. Once the good human data runs out and models start training on each other’s output, quality drops fast.

Even worse, it makes it difficult to distinguish between real and fake data.

Sorry,I mean capability limits.