weekend ai reads for 2025-01-31 (#100!)

Shares of Microsoft — widely seen as a frontrunner in the AI race because of its ties to industry leader OpenAI - were down 6% in early trading on Thursday after the company said growth in its Azure cloud business would miss third-quarter estimates. "We really want to start to see a clear roadmap to what that monetization model looks like for all of the capital that's been invested," said Brian Mulberry, portfolio manager at Zacks Investment Management, which holds shares in Microsoft.

It goes beyond that though— the very programming framework that coders use to write low-level code that is optimized for GPUs, CUDA, is totally proprietary to Nvidia, and it has become a de facto standard. If you want to hire a bunch of extremely talented programmers who know how to make things go really fast on GPUs, and pay them $650k/year or whatever the going rate is for people with that particular expertise, chances are that they are going to "think" and work in CUDA.

The way I would tell this story is that the modern world has created conditions in which it is incredibly lucrative to get very good at statistical inference.

But these same models sometimes struggle with basic tasks, like arithmetic or writing metered poetry. That has given them a reputation as astoundingly brilliant at some things and totally useless at others, and it has created vastly different impressions of how fast A.I. is improving, depending on whether you’re looking at the best or the worst outputs.

Second is the low training cost for V3, and DeepSeek’s low inference costs. This part was a big surprise for me as well, to be sure, but the numbers are plausible. This, by extension, probably has everyone nervous about Nvidia, which obviously has a big impact on the market.

This calculation showed that South African workers using gen AI provided 40% more value for money than U.S. workers using gen AI, and 80% more than U.S. workers not using gen AI.

Today we’re announcing ChatGPT Gov, a new tailored version of ChatGPT designed to provide U.S. government agencies with an additional way to access OpenAI’s frontier models.

the animations (1, 2) are funny and curious

Some differences here are: firstly, a higher recall is better, holding summary length constant. You don’t want to score 100% recall with a summary the same length as the source. Secondly, you’d ideally want precision to be close to 100% as possible — hallucinating information is really bad.

The authors note that the LLMs did not always associate pleasure or pain with straightforward positive or negative values. Some levels of pain or discomfort, such as those created by the exertion of hard physical exercise, can have positive associations.

As of right now, it’s unclear what role AI plays in the project, or whether McAfee’s widow is operating his official page or the AIntivirus account, which has over the past day tweeted up a storm in the late programmer-turned-criminal’s voice.

Social media represents a shift in power from the content as a product of consumption, to the creator as an artist. Every creative has a unique set of knowledge and identities that can be celebrated because it allows for exactly what audiences are searching for – a curated, niche place to find solace.