Ask HN: How do you explain LLMs to those *still* unfamiliar?

[AlwaysNewb23]

(Or AI in general in some cases)

[Leftium]

"AI" has unfortunately become synonymous with "LLM," even though LLM's are only a subset of AI.

"Autocomplete" seems to be the simplest way of explaining it is just fancy pattern recognition.

Slightly longer:

> Traditionally programming meant explicitly telling computers how to do something. 'AI' just gets the inputs and outputs, and figures out the 'how' own its own.

---

I share these explainer videos when people ask me to explain AI to them:

- https://youtu.be/-4Oso9-9KTQ

- https://youtu.be/Sqa8Zo2XWc4

- https://youtu.be/ro130m-f_yk

- https://youtu.be/MWHN6ojlVXI

[cpach]

Stephen Wolfram has a good explanation: https://writings.stephenwolfram.com/2023/02/what-is-chatgpt-...

[JoeAltmaier]

Here's how CHATGPT describes itself:

Large Language Models (LLMs) are advanced artificial intelligence systems designed to understand and generate human language. Trained on vast amounts of text data, these models utilize deep learning techniques, particularly transformer architectures, to predict and produce coherent and contextually relevant text. They can perform a variety of tasks, including language translation, summarization, question answering, and content creation, by leveraging their ability to recognize patterns and relationships in language. LLMs, such as OpenAI's GPT-4, have become integral in both research and practical applications, demonstrating impressive capabilities in natural language processing and understanding.

[andrei_says_]

A word generator based on statistical probabilities. Has no understanding of the real world or logic. Everything is hallucinated, some word combinations coincide with reality but that’s a random occurrence.

[sk11001]

> Everything is hallucinated, some word combinations coincide with reality but that’s a random occurrence.

This isn't accurate, these models are trained and tuned to be correct, it's not a random occurrence.

[andrei_says_]

How is correctness guaranteed? The model has no understanding, just a statistical analysis.

Or are we speaking of higher probability for correctness?

[Davidbrcz]

I was on that side of the argument as well, but it seems that models can have emerging model states of a given situation

https://arxiv.org/abs/2210.13382

[gardenhedge]

Sentence generator that adds words based on learned probabilities of a trained model. When the training data is big enough and the model is tuned enough, it can be seen as AI.

[ai4u]

It’s worth clarifying: LLM is a game of probability, and artificial intelligence is a big topic of discussion. First, we need to figure out what the teacher should teach, and then figure out what content the students have already mastered.

[JojoFatsani]

Autocorrect but much bigger

[al_borland]

This is highly dependent on the person. Who are they to me? What is their technical background? Why do I think they need to know what an LLM is?

[sabrina_ramonov]

large-scale deep transformers to model next token probability distribution conditioned on context by optimizing loss function

[jjaison13]

it uses what it learns from a bunch of text that humans have written to decide how to best answer a question

[feeela]

A stochastic parrot.

[muzani]

your brain is a stochastic parrot

[nothercastle]

It’s a glorified auto correct

[julianbenjamin]

Google except easier

[ElectricBoogie]

"A stupid computer is trained to appear intelligent by predicting the next word."

[alexshendi]

Wait until the hype is over?

[tripleo1]

They have been working on other things for decades. Look there.