AI - AI Dynamics

→ View original post on X — @jeffdean

Congratulations on Princeton Admission Future Research

By

@jeffdean

–

13 July 2023 23h44

Congrats! Princeton is lucky to have you, and I'm looking forward to your future work!

→ View original post on X — @googleai

Symbol Tuning: Improving In-Context Learning with Arbitrary Symbols

By

@googleai

–

13 July 2023 23h40

Today on the blog, read all about symbol tuning, a method that fine-tunes models on tasks where natural language labels are replaced with arbitrary symbols. Learn how symbol tuning can improve in-context learning for benefits across various settings →
https://
goo.gle/3rt3r8U

Mode Collapse in Large Language Models: Book Output Example

By

–

13 July 2023 23h27

Here’s actual example of mode collapse in an LLM, where “book” is repeatedly outputted.

Mode Collapse in LLMs: Understanding Model Degradation

By

–

13 July 2023 23h21

This is an example of mode collapse—which it turns out happens in LLMs as well (and we discuss in the paper)!

AI Models Data Generation Distribution Guarantees

By

–

13 July 2023 23h09

Despite growing interest in using these models to generate data, there are currently no guarantees about the resulting distribution. For example, if I want a normal distribution, it would look like this:

Getting Started with ChatGPT: A Guide to Large Language Models

By

–

13 July 2023 23h09

Side note: If you’ve never you’ve never interacted with a large language model, try playing with ChatGPT: https://
chat.openai.com

LLMs Generate Synthetic Data: New Research Released

By

–

13 July 2023 23h09

As part of my takeover, I’ll (
@aspenkhopkins
) be covering some recent work. We released this project today on large language models’ (LLMs) ability to generate synthetic data