Against Entropy

Solving the Holy Grail of Internet Problem: Combating the Informational Entropy & Establishing a Single Source of Truth

Abstract:

When you search for something on a search engine, like Google, you end up getting links. Not answers. It doesn't even qualify to be called a search engine, at its very best it is, but a sorting engine.

The biggest problem with current search engines is that they are not conversational, contextual or coherent.

With Against Entropy, I aim to solve that: Come up with LLM-powered search engines that are contextual, conversational, and coherent. And I believe that with the current pace of AI research, I can create an all-knowing being (LLMs/LMs for now, AGI soon).

The Problem:

Try searching for something like "How many people died of Leukemia in 2017?" and all you will find are links and links. Not Answers. You went there to search for answers, not links.

LLMs can do that. LLMs are amazing at it. Try asking the same question to GPT3 and be amazed. Not only you can follow up on it, but also chat with it as an all-knowing being and maybe a friend. But there is only one problem - it is not factually correct!

The Solution is simple: build a single SSoT (Single Source of Truth) layer that can interface between the existing body of information on the web and LLMs/LMs.

Build:

I am currently working on a few different algorithms & solutions, but they are very nascent. I am trying to build upon them. If you would like to learn more about the solutions, feel free to contact me through the details given below.

Even though I am focusing a lot on LLMs and SSoT layer, that's not the core problem that I am working on the main problem is Making "search" less bad. I am happy with a 95% accurate search with a good search UX rather than a completely accurate yet entropic search.

Future:

BCIs are gonna take over soon. Just a few weeks ago, we have achieved the ability to reconstruct language from non-invasive brain recordings. That is absolutely Nuts! AI Vision is already Solid. AR is still budding but I am hopeful about it. With all of them coming together - we can imagine a search engine embedded (non-invasively) into Brain! something very similar to Jarvis! The only missing piece for them to work is an SSoT.

Contact:

I am so pumped up to work on this. If this sounds exciting to you and you would like to discuss/fund/collaborate/enquire feel free to reach me out on @VirajCz on twitter or email me at virajcz@ucla.edu