2084: IBM Research Letter and Hallucinations

Short post today, but would recommend reading the IBM research letter

I was reading the IBM research letter for 2022, and it was fascinating. I didn’t know that 2nm transistor widths had already been achieved in 2021, and were being worked on to be manufactured at scale - this could potentially increase battery life of cellphones by 45-75% - or that there was significant work being done on 1 nm transistors.

There’s also a lot on the future of AI hardware and software, and a quite interesting part on a hyper optimized computing systems for ML workloads, IBM z16, that could run models 75% more efficiently than x86. Well worth reading the various parts, even if it doesn’t sound very objective. It’s something to make you excited for what the future could bring.

There’s also an interesting paper on detecting hallucinations in machine learning models which was quite interesting. If you could combine this with the previous paper on editing facts, you could probably rather quickly converge to a rather accurate model. Soon, there might be no more complaints about the inaccuracy of models.