> Markdown version of [/events/world-congress-2024/sessions/397-intro-to-llms-and](https://www.wearedevelopers.com/events/world-congress-2024/sessions/397-intro-to-llms-and). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Intro to LLMs and recent developments


- **Date:** Friday, Jul 19, 2024
- **Time:** 14:20–14:50 (30 min)
- **Room:** STAGE 6 (120)
- **Event:** World Congress 2024

## Description

ChatGPT is used almost synonymously with LLMs. But that's just the tip of the iceberg – the development is extremely dynamic.

This presentation gives an overview of the architecture and the recent development of different models. We will show how models can run on your own computer (with or without a GPU) and with free cloud services.

A lot of effort is spent in researching optimizations like quantization which leads to speedups and reduced RAM usage.Sparse mixture of experts models (SMoE) are a new architectural approach which power the most recent powerful models. Combining different LLM approaches leads to interesting business opportunities which have not yet been fully realized.

In the outlook we venture a prediction about future directions in LLMs.


## Speaker

### [Christian Winkler](https://www.wearedevelopers.com/@christian-winkler)

CEO of datanizing, research professor at TH Nürnberg


## Related talks at this congress

- [Using LLMs in your Product](https://www.wearedevelopers.com/events/world-congress-2024/sessions/50-using-llms-in-your) — Daniel Töws
- [Efficient deployment and inference of GPU-accelerated LLMs​](https://www.wearedevelopers.com/events/world-congress-2024/sessions/73-efficient-deployment) — Adolf Hohl
- [Why do LLMs hallucinate?](https://www.wearedevelopers.com/events/world-congress-2024/sessions/151-why-do-llms) — ofermend
- [LLMs security – common pitfalls and their prevention.](https://www.wearedevelopers.com/events/world-congress-2024/sessions/152-llms-security-common) — Michał Mikołajczak