Pradyumna Prasad on pradyuprasad.com

(Image source here) In light of the unsanctioned AI compromise of OpenAI infrastructure, the hacking of Hugging Face, and other misalignment and security incidents, both OpenAI and Anthropic have allowed independent investigations of their models’ actions.

A benchmark for measuring reward hacking in LLM agents doing ML research.

No matter what you’re doing, from building a civilization on Mars to getting a summer internship, you will have to ask people for help.

A popular view of the economics profession, as one with oscillating, or perhaps no consensus, is summarized by Harry Truman’s supposed quip that he wanted a one handed economist, because they kept saying “on one hand …but on the other…”.

Evaluated Elicit’s systematic review workflow against 994 Cochrane reviews.

Benchmarked Elicit’s paper search against Google Scholar, Semantic Scholar, OpenAlex, and Consensus on BioASQ.

I started Bretton Goods as an endeavour in understanding the causes of the variations in the wealth of nations.

In September of this year, the Land Transport Authority said that three new autonomous bus services would start plying passengers in Punggol.

Does when you review your Anki cards matter?

Anthropic released research today showing that carefully designed prompts can elicit blackmail, corporate espionage, and other harmful strategic behaviors from AI models across the industry.

Between 2014 and 2023, Computer Science bachelor's degrees in the US doubled from 50,000 to 112,000 annually during a brutal job market.

I went to my first ML conference the last few days and I realised there is a lot of stuff I didn’t know about how to attend a large ML conference.

Nobody has lived through a technological change that in theory at least promises to take all of our jobs, change our existence as the apex species on the planet and potentially kill us.

It was only after the release of the Deepseek R1 model that I realised that RL was our best shot to AGI, and possibly even super intelligence To summarize the past (2019 to 2024): When you perform regular language model pre-training, what you’re doing is…

“Endogenous!”

Tl;DR The Index of Economic Freedom ignores an important measure of government size: government ownership of the economy.

TLDR Under slow takeoff, general AI is an indefinite race.

Thanks to Sam Enright, Fergus McCullough, Nihal Sahu and Judah for their comments on this Making public policy is hard.

https://distill.pub/2020/circuits/zoom-in/ This by Olah et. al is a very interesting explanation of how neural networks work Patronage and Research Labs - the history of scientific research and how it connects to patronage from the rich Lessons from the G…

29/1/2023 - The High Cost of Expensive Housing (and How Auckland Might Fix it) Two mistakes here I named it “The High Cost of Expensive Housing (and How Auckland Fixed it)” but that was misleading.

This piece was previously titled “The High Cost of Expensive Housing (and How Auckland Fixed it)”.

Behave in good faith.

In lieu of the usual post for November and December, I’m here to give you two book recommendations and some thoughts about them.