🏴‍☠️ X marks the spot

X-ing out your info

Elon Musk’s plaything X has recently made changes to its platform policies, one of which will allow the company to use posts and other data to train AI models.

As noted on Y Combinator’s Hacker News forum, X says it “may use the information [it collects] and publicly available information” to teach its AI models.

So far so shruggie emoji, right? After all, Musk already made a stink about OpenAI purportedly scraping the service formerly known as Twitter to make ChatGPT smarter, so why wouldn’t he use the content from his own platform.

Of course, the wrinkle here is that Musk has talked up xAI and TruthGPT, and now this X policy update makes a whole lot more sense. It’s plausible that X will act as a database for xAI, though considering how much of a tarpit the site has become, one has to wonder just what kind of verbiage his large language model (LLM) might ultimately spit out.

Why it matters:

Considering X/Twitter has been around since 2006, Musk has access to a lot of possible training data. And with the AI industry estimated to be worth $300 billion by 2026, Musk might somehow be able to justify that $44 billion he spent.

The LLM that shows you what it’s thinking

Wired profiles Aleph Alpha, a Germany-based startup working on a ChatGPT competitor.

Founded by ex-Apple engineer Jonas Andrulis, Aleph Alpha is an LLM that can handle several languages including English, German, French, and Spanish. This multilingual approach has seen the LLM adopted by numerous governments and corporations.

Andrulis highlights what might be the most interesting part of Aleph Alpha - humans can check its homework.

When you click on generated words or images, the LLM will show you a brief overview of why it chose that phrase or picture. In the piece, Andrulis demonstrates how Aleph Alpha interpreted the word “sunset” in a graphic by placing a highlighted box around layers of yellow and red on the horizon.

Wired highlights how Aleph Alpha might fare better than some of its US counterparts in Europe once the European Union (EU) finalizes its AI Act. Of course, the bloc spent decades talking about a unified currency before the Euro showed up, so don’t expect AI rules in a hurry.

Why it matters:

Aleph Alpha’s usage is in the tens-of-thousands, which is dramatically less than ChatGPT or Bard. But its ability to present how it makes its decisions could set a precedent for LLMs - not just in the EU, but around the world