Peter Dolan
astrocatcommander.bsky.social
Peter Dolan
@astrocatcommander.bsky.social
AI Research Engineer working on AI Safety and Alignment | formerly OpenAI, Waymo, DeepMind, Google. Father, photographer, Zen practitioner.
Today I called customer support and after navigating a deep phone tree I was finally connected to… an AI agent. It was totally unhelpful and insisted I needed to go look up the answer to its question in my manual.

Eventually I convinced it to connect me to a human.
April 16, 2025 at 12:35 AM
import os
from openai import OpenAI

client = OpenAI()

def sort(values: list[int]) -> list[int]:
response = client.responses.create(
model="gpt-4o",
input=f”Please sort {values}. Answer with comma separated values.")
return [int(i) for i in response.split(“,”)]
April 14, 2025 at 1:44 AM
Building on our novel `min` algorithm:

import random

def sort(values: list):
sorted = []
while values:
i = random.choice(range(len(values)))
vmin = values[i]
if all(vmin <= v for v in values):
sorted.append(vmin)
values.pop(i)

values.extend(sorted)
April 9, 2025 at 7:43 PM
import random
numbers = [5, 3, 9, 2, 8]

min = random.choice(numbers)
while not all(min <= n for n in numbers):
min = random.choice(numbers)
print(min)
April 2, 2025 at 4:36 PM
def iseven(n: int) -> bool:
if n == 2:
return True
elif n == 1:
return False
else:
return iseven(n-2)
March 31, 2025 at 6:23 PM
Very glad to see this work published, and congrats to Melody! This significantly develops some ideas I was pursuing at OpenAI - how does LLM reasoning enable more sophisticated and accurate understanding of human culture and social norms?

arxiv.org/abs/2412.16339
Deliberative Alignment: Reasoning Enables Safer Language Models
As large-scale language models increasingly impact safety-critical domains, ensuring their reliable adherence to well-defined principles remains a fundamental challenge. We introduce Deliberative Alig...
arxiv.org
March 20, 2025 at 4:56 PM
In adorable news, my 2yo daughter wishes me happy birthday 2-5 times every day ☺️
March 20, 2025 at 1:00 AM
Reposted by Peter Dolan
In an exclusive for Nature, I report on a paper that AI folks will probably debate for a LONG time.

The key finding: the time horizon of tasks AI can handle is doubling fast. Extrapolating the trend: AIs will be able to handle 1-month tasks by 2029. 🧵

x.com/METR_Evals/...
March 19, 2025 at 8:03 PM
I’ve been thinking about LLMs and cultural diversity. We could evaluate its understanding through some benchmarks, but could we go further and develop an artificial anthropologist? One that could tell us new things about ourselves from its training, or design new studies?
March 20, 2025 at 12:44 AM
@tmeato.bsky.social is there anywhere I can buy a copy of "Billie Eilish - Everything I Wanted (Demented Soul Imp5 Afro Mix)"? I love it on SoundCloud, and hope I can send you some $ for an offline copy
March 19, 2025 at 3:38 AM
I realized that my instant pot has a “warm” setting that you can set the temperature on, and it holds it pretty accurately - within one or two degrees at least. So, I got a small pond pump to agitate the water and am using it for sous-vide cooking. It works great!
November 26, 2024 at 10:33 PM
Thinking about making a fact-checking bot to trawl social media. Would be interesting to consider cases where there are conflicting sources, may be a good application of multi-agent debate methods. I developed some of these at OpenAI, found they made substantial improvements over a single agent.
November 20, 2024 at 10:13 PM
Just shared my experience, tips, and habits for effectively co-working with AI!: medium.com/@peterjdolan...

I’m eager for feedback and to hear your experiences and tips too!
September 15, 2023 at 11:33 PM