AI training for Java developer

[KrishnaSri]

I’m a Senior Java Developer looking to transition into AI/ML, and I’m not sure where to start. Given my strong software engineering background, can anyone recommend a good online AI/ML trainer or program that focuses on practical, industry-relevant learning?

I’d really appreciate any suggestions or personal experiences. Thanks!

[Teluguredu]

 

I have been knowing about gen A.I in weekends for fun to know how it works from base.

First you need to learn about basic neural networks 

Forward pass ,hidden layers and back propagation,just to get a feel and intuition of everything works underneath

And then a little pytorch to learn to train your own models 

And then about transformers  and then RAG and then how to use langchain to train language models.

A.I is a broad field though ,ML is very math heavy and need to learn statistical learning but you don't need ML for fine tuning language models and programming.

You can start with andrej karpathy(co founder of open A.I and tesla director of A.I department ) zero to hero videos and watch one every night before sleeping.he builds gpt from ground up.

[Teluguredu]

17 minutes ago, KrishnaSri said:

I’m a Senior Java Developer looking to transition into AI/ML, and I’m not sure where to start. Given my strong software engineering background, can anyone recommend a good online AI/ML trainer or program that focuses on practical, industry-relevant learning?

I’d really appreciate any suggestions or personal experiences. Thanks!

Also ML is more math related than software.

[saravamnene]

My honest suggestion dont get into AI model building and agents

the online videos you see are so so basic and hardly helps you to crack interviews. 

Stick to software engineer and try to add AI to your workflows

Many people I saw in my friends circle ask me for AI recommendations.

I give a road map and they oh so just python, and RAG is enough

 I week of programming and understanding the math they give up

Read any arxiv paper and see if you can get a sense of it

AI Core is only for PHD's and not for youtube learning masters guys. Period

[Teluguredu]

17 minutes ago, saravamnene said:

My honest suggestion dont get into AI model building and agents

the online videos you see are so so basic and hardly helps you to crack interviews. 

Stick to software engineer and try to add AI to your workflows

Many people I saw in my friends circle ask me for AI recommendations.

I give a road map and they oh so just python, and RAG is enough

 I week of programming and understanding the math they give up

Read any arxiv paper and see if you can get a sense of it

AI Core is only for PHD's and not for youtube learning masters guys. Period

Not really A.I is a very big field.

To become an engineer you don't need phd but to become researcher you do.

I don't recommend getting phd tbh because you will be confined to that field and can do nothing else it's almost like a superspeciality doctor.

Only math knowledge you need is matrix multiplications ,probability and derivatives for backpropagation  ,rest of the things you can pick up while reading.

[saravamnene]

35 minutes ago, Teluguredu said:

Not really A.I is a very big field.

To become an engineer you don't need phd but to become researcher you do.

I don't recommend getting phd tbh because you will be confined to that field and can do nothing else it's almost like a superspeciality doctor.

Only math knowledge you need is matrix multiplications ,probability and derivatives for backpropagation  ,rest of the things you can pick up while reading.

ok build an LLM from scratch and tell me about the math and probability

[Teluguredu]

55 minutes ago, saravamnene said:

ok build an LLM from scratch and tell me about the math and probability

It's simple to build a language model just take a paragraph and keep building matrices for n-gram table 

For example 

If there are 36 words then build a 36*36 table that can predict next word after each word 36*36*36 to predict next word after 2 words 36*36*36*36 to predict next word after 3 words ..... Probably it will get accurate after 4-5 tables like that 

,normalise the counts into probabilities and use multinomial function to sample output from input doesn't even need a neural network lol.