RSS Amplifier

Dimka's thoughts · Feb 28, 2024

The next big AI breakthrough

0
Sign in to vote or save

Dmitri Zdorov · Dimka's thoughts

Artificial Intelligence has been around for many years. However, usually the moment it becomes available to ordinary consumers, it is no longer called AI, and it becomes algorithms, or even just services. And these are all manifestations of so-called narrow AI. Over the last year or so there has been a special buzz about AI based on LLM - Large Language Models, and although everything there is not so trivial, the main idea is to statistically analyze a very large amount of text and find various patterns there, and this allows us to make good predictions from all the text (language) sucked into the model that a certain person would statistically say next.

Here's my small prediction. The next round in development and accordingly in usefulness in application will be based on LSM - Large Scientific Models. Although it may be called differently. But the essence is that instead of just text that people have written for the base will take various kinds of scientific data, such as measurements from experiments, parameters that are known about everything that we have already found out, laws and formulas and so on. For example, they tested a new wing shape in a wind tunnel, got some huge data set from there, click-click, and added it to the LSM. All temperature measurements all over the world, all read data from all telescopes, gas chromatographs and spectrometers, chemical compositions of all known materials and substances, recipes for solutions and descriptions of technological processes, videos of experiments, genomes, tables and calculations, traffic statistics and bird migration pattern etc. etc.

Then the AI will be able to produce not just a summary from text or a poem about what you thought up to ask, not a piece of program code, but something similar to new scientific and technological discoveries. The benefits of this will be tremendous, unless of course it falls into the wrong hands. Perhaps mere mortals will not have access to such things at first. And it's hard to get data there, and the data is disproportionately more than texts. But gradually, at first in some narrow directions and separate industries it will appear and make an incredible breakthrough. And though it will not be AGI yet, it will look like it way more than LLM.

No posts

Read the original on dimka.substack.com

Comments

Nothing yet. Say the first thing.

    Sign in to join the conversation.