Internet: The stupid AGI problem

Yash Bonde (~yash86)


AGI has always been a human vision and dream. Rapid progress in the domain of narrow AI over the past decade has given rise to a new hope. However there is no benchmark for AGI and it's more philosophical debate than engineering one. If you pick a random person from earth and ask about population of EU?, without access to internet it's going to very hard. With an internet though anything that answers it is automatically smarter than majority of population.

I propose igym a simple interface to develop agents for the internet that can help solve problems and not debate about viability of AGI. I go over the basics and then explain how to build an AI that can solve complex tasks and finally end with a demonstration.

Talk Outline

0-5 mins: Introduction

5-10 mins: Current methods

10-20 mins: My Framework for Solving AGI

20-25 mins: igym, RNN-sim and other code in that direction

25-30 mins: closing remarks and questions.


A brief of what Language models are and an idea of what intelligence is.

Content URLs:

igym repo

Speaker Info:

Yash Bonde is an independent AI researcher working on RL, analysis of NNs and problems where progress has been very slow. Professionally I consult NPAW on AI related projects such building an advanced natural language interface. I am also part of where I work on R&D, making educational material, optimising models for CPUs and overall product development.

Speaker Links:

My Personal blog and Github. More work can be found on NimbleboxAI Github.

Section: Data Science, Machine Learning and AI
Type: Talks
Target Audience: Intermediate
Last Updated: