Daniel Paleka's Newsletter

Obvious ways RL can fail

Reinforcement learning works for some things and not for others. Why?

Nov 15 • 

Daniel Paleka

2

The two types of LLM preferences

The standard approach to measure values or preferences of LLMs is to:

Nov 10 • 

Daniel Paleka

13

October 2025 AI safety news: Adaptive attacks, Tokenization, Impossible tasks

These days, I imagine it is rough for researchers working on LLM defenses.

Nov 7 • 

Daniel Paleka

3

You are going to get priced out of the best AI coding tools

The best AI tools will become far more expensive. Andy Warhol famously said:

Nov 5 • 

Daniel Paleka

14

7

A/B testing could lead LLMs to retain users instead of helping them

OpenAI’s updates of GPT-4o in April 2025 famously induced absurd levels of sycophancy: the model would agree with everything users would say, no matter…

Nov 2 • 

Daniel Paleka

11

8

July 2025

Memetic optimization #1: brainrot

I. I don’t use social media except for X, and even there I peruse only the Following tab.

Jul 10 • 

Daniel Paleka

9

1

May 2025

March-April 2025 safety news: Antidistillation, Cultural alignment, Dark patterns

Happy NeurIPS deadline to all those who celebrate!

May 16 • 

Daniel Paleka

7

1

March 2025

GPT-4o draws itself as a consistent type of guy

When asked to draw itself as a person, the ChatGPT Create Image feature introduced on March 25, 2025, consistently portrays itself as a white male in…

Mar 31 • 

Daniel Paleka

22

12

2

January-February 2025 safety news: Emergent misalignment, SAE sanity checks, Utility engineering

Some papers I’ve learned something from recently, or where I have takes.

Mar 9 • 

Daniel Paleka

7

1

January 2025

You should delay engineering-heavy research in light of R&D automation

tl;dr: LLMs rapidly improving at software engineering and math means lots of projects are better off as Google Docs until your AI agent intern can…

Jan 6 • 

Daniel Paleka

12

1

October 2024

September/October 2024 safety news: Jailbreaks on robots, Breaking unlearning, Forecasting evals

Better version of the Twitter newsletter.

Oct 31, 2024 • 

Daniel Paleka

7

August 2024

July/August 2024 safety news: Tamper resistance, Fluent jailbreaks, Scaling limits

Better version of the Twitter newsletter.

Aug 31, 2024 • 

Daniel Paleka

5