Experiments in AI
Experiments with NLP and GPT-3 Podcast
Should India do core foundational AI research?
0:00
Current time: 0:00 / Total time: -5:32
-5:32

Should India do core foundational AI research?

Has India lost the foundational AI war?

The general feel from the Meta and NVIDIA events seem to be:

1.India has lost the foundation war.

2.Let's build use cases.

3.Lets maybe look at SLMs.

4.There is no research.

Personally, I disagree. I think we need to double down on research. Especially for global south.

The current approach seems to be

  1. Huge compute

  2. Lots of data

We can say, India does not have both or say both are costly.

Or we can say, is there a better way to do AI.

We might not find the answer, but shouldn’t we even try :)

Just as a reminder, core research is certainly happening in India.

On data collection, fine tuning and new models.

Example1: Telugu ASR.

మన ఇండియాలో రీసర్చ్ జరుగుతుంది.అదేంటంటే, డబ్బుతో కూడిన రీసర్చ్ కాదు. Please do try it out and share feedback.

https://speech-kws.ozonetel.com/asr_demo/

Example2: Semantic chunking for RAG

A new semantic chunking approach to RAG

·
September 24, 2024
A new semantic chunking approach to RAG

As we saw in my last blog post, there is a shape for stories.

Example 3: Building new embeddings for sentences.

Building a new embedding(bit vector) for sentences

·
February 9, 2023
Building a new embedding(bit vector) for sentences

One clear use case that is coming up from LLMs is document QnA. We have been experimenting with QnA for the last 2 years.

Discussion about this episode