Should I consume less content, or is there a lot happening? Either way, there’s a lot to talk about.

We probably should start an “open source rug pull” counter somewhere because Redis just pulled one off from right underneath our feet.

Days since the last open source rug pull: 0

We enter an election year with the looming question of AI’s influence. So far, knee-jerk reactions that paint AI as good or bad (mostly) have been the only attempts at addressing something that clearly requires better solutions.

Thinking in terms of absolutes is the real problem. Instead of viewing AI as either good or bad, it can be viewed as better or worse. “How better are we than the status quo with AI,” or “How worse off would we be than we are now with AI?” are the questions we need to answer.

Thinking at the margin reveals that the perceived risks of open AI models are far greater than the actual risks for most applications. The only reasons to stick to and advocate for closed models are strategic and economic.

An analogy to this AI-led geopolitics battle is the nuclear arms race between the USA, the Soviet Union, and the rest of the world in the latter part of the last century. India managed to build nuclear missiles in the 90s, even with all the checks and regulations in place.

Similarly, regulating AI is futile because no single entity has the power to coerce all nations who are incentivized to leverage this technology for economic and strategic advantage.

Open models are the way to go and there is no doubt about it.

New research suggests useful AI models (comparable to equivalent proprietary models) can be trained from publicly available datasets. This is excellent because companies with access to proprietary data have an unfair advantage over new entrants.

For example, Google, Amazon, and Microsoft have vast amounts of proprietary data to train their models using their own compute infrastructure. You basically win Monopoly at this point.

Another advantage of keeping models open is that it reduces misrepresentations, biases, and other issues that have plagued them since their invention.

We could take a step further for the elections and call for open source voting machines. It’s a little late for this election cycle, but how great would that be?

What’s Happening?

The most important AI news that dropped last week wasn’t all this, and it wasn’t even Twitter open-sourcing Grok, but the release of Devin, an AI programmer that can seemingly do it all.

It was another “our jobs are over” kind of week for programmers, and the demos look neat, but it’s far from being able to replace programmers.

Devin’s creators report that it can solve 1/7 GitHub issues correctly. Not bad considering the novelty of the technology.

Thousands of AI Authors on the Future of AI
Thousands of AI Authors on the Future of AI

If science continues undisrupted, the chance of unaided machines outperforming humans in every possible task was estimated at 10% by 2027, and 50% by 2047 (source).

The nature of our jobs will change inevitably. That has been the case for every technological leap before. But this isn’t the first attempt to make programmers obsolete.

COBOL was developed to allow business teams to program. It ended up creating COBOL programmers. The myriad of no-code tools are great, but there are still programmers.

AI is a radical change, but is it radical enough to make programmers obsolete?

There was also some buzz around AI hardware, with Hugging Face announcing an open source robotics project and Extropic.ai announcing their hardware platform. The AI hardware space is neat, and the research around it will be critical in the coming years.

I don’t care about the OpenAI drama anymore, but they released the investigation report concerning Sama’s termination from the company. The Mira Murati interview was funny. If I were involved in OpenAI, the least I could do is learn about what the company actually does when going to an interview or at least prepare for an obvious question.

In other news, Williams Racing Team Principal James Vowels is appalled to find his new team still using Microsoft Excel to manage 20,000 Formula 1 car parts. Does he know Excel can run Python now?

We focus so much on the now, so it’s time for some classics. Here’s what I have for you this week:

Hot off the Press

I wrote about the Freenginx fork, but the article ended up focusing more on the interpersonal aspects of the scenario. I have previously written about Nginx and how it is just fine for most of us.

Read here: “Wasn't Nginx Free?

Read here: “Nginx is Probably Fine

I’m working on a new article about CloudFlare’s newly open source project, Pingora, and I expect to publish it next week.

Now, who’s forking Redis?