Today I Learned this basic LLM concept from a conversation with my DS friend

LORY
5 min readJan 1, 2024

Talk with boss

“Could you take a look at LLM and find some use cases for our business to generate some revenue?” My boss said.

“Well, but I only have a very general knowledge about it,” I said.

“Okay please spend some time diving into it, then explain to me how it works, send an invite cc XYZ, and let’s discuss,” he said.

“Okay, let me try”. I said.

Asking my DS friend

“Hey bro could you explain to me in very very layman’s terms, how LLM works” I said to my data science friend.

“Sure, one analogy could be related to the job you are doing as part-time — interviewing”. He said.

“I was surprised, how?” I asked.

“Input is a CV plus a 60–90-minute interview process, so what is output, or how do you know if the person is shortlisted?” he asked me.

“Depends on the company requirement, most of them I will need to feedback: coding skill, design, and soft-skills,” I said.

“Okay now If I asked you to score a candidate, 1–10. you will write something like 9/10, 8/10, 7/10 soft-skills right?” he asked.

“Is it something like this?” I asked.

“Yes, similar, but done by machine learning model — that’s called the embedding process”

Embedding

“So in LLM there is some process similar to ‘Interview process’, convert the input into vectors?” I asked.

“Yes, that is a very important process, word embedding,” he said.

“So how does this work?” I asked.

“Well, I am not sure which one you are asking. there are different models, from word2vec, bert with attention mechanism then transformer, and then GPT, all different, It’s like when you study some framework,1.0, and 2.0 and keep evaluating”.

“Let’s start simple, the word2vec”, I said.

“It’s doing something like the ‘interview process’, the goal is to get below, “plot” the input into a vector space, and score it at each dimension. If we plot, may look like this. As…

--

--

LORY

A channel which focusing on developer growth and self improvement