🫡Wiki's AI Cleanup Crew

Wikipedia editors launch project to combat rising tide of AI-generated misinformation

Key Points 

• A group of Wikipedia editors started WikiProject AI Cleanup to tackle the issue of poorly written, unsourced AI content on the site. They’re using specific AI phrases to help identify misleading content.

• One major challenge is fake citations, which can go unnoticed for months.

☕News - A group of Wikipedia editors have started WikiProject AI Cleanup—a project focused on dealing with the rise of poorly written, unsourced AI-generated content on the site. Their goal is to keep Wikipedia safe from the misleading AI content that's been showing up in places like Google searches, Amazon books, and academic journals.

🥸The situation from an insider's perspective - A few editors noticed more and more unnatural writing on Wikipedia that looked like it was AI-generated. Ilyas Lebleu, one of the founding members of WikiProject AI Cleanup, said they were able to replicate that style using ChatGPT. They picked up on certain AI catchphrases, which helped them spot some of the worst examples, and that led to formalizing the project to organize their findings and methods.

Most of the time, WikiProject AI Cleanup finds AI-generated content by looking for phrases commonly used by ChatGPT—similar to how others find AI-generated work in scientific journals and Google Books. One obvious example is a Wikipedia article about the Chester Mental Health Center that included the phrase, “As of my last knowledge update in January 2022,” showing it was clearly written by a language model.

But, not everything is that easy to spot. Fake citations are a bigger issue, according to Lebleu, because they can go unnoticed for months. Even if the AI-generated text reads well and has properly formatted citations, it often doesn’t link the citation to the specific claim in the article, making it hard to catch.

🕵🏻‍♂️What about AI-generated images? WikiProject AI Cleanup doesn’t remove images just because they’re AI-generated. In some cases, they’re actually appropriate, especially if the article is about or references the AI-generated image itself. 

Nevertheless, it seems that, for now, Wikipedia is doing a much better job at catching and filtering out misleading AI-generated content compared to platforms like Facebook and Google, which have repeatedly failed to do so.

Nvidia faces new competition as AMD launches new AI chip to rival Blackwell

Key Points 

• AMD has launched the Instinct MI325X AI chip to challenge Nvidia’s stronghold in the data center GPU market.

• They expect to start production of the MI325X by the end of 2024, going head-to-head with Nvidia’s upcoming Blackwell chips.

• With the data center market projected to hit $500 billion by 2028, AMD is looking to position its AI chips as viable alternatives to Nvidia's offerings, potentially increasing pricing pressure on Nvidia's products.

♨️News - AMD has unveiled a new artificial intelligence chip called the Instinct MI325X, targeting Nvidia’s data center graphics processors, or GPUs. 

Production is expected to start by the end of 2024, putting the MI325X in direct competition with Nvidia's Blackwell chips, which are set to begin shipping in large quantities early next year.

🧐What to expect - Nvidia has dominated the data center GPU market for the past few years, with AMD typically in second place. Now, AMD is setting its sights on capturing a significant portion of the market, which it estimates will reach $500 billion by 2028.

If developers and cloud companies begin to see AMD's AI chips as viable alternatives to Nvidia's, it could create pricing pressure on Nvidia, which has enjoyed about 75% gross margins thanks to high demand for its GPUs over the past year.

However, AMD faces a significant challenge: Nvidia’s chips use their own programming language, CUDA, which has become the standard among AI developers. This essentially keeps developers locked into Nvidia’s ecosystem.

In response, AMD has announced improvements to its ROCm software, aiming to make it easier for AI developers to switch their models to AMD’s chips, referred to as accelerators. AMD believes its accelerators are particularly competitive for tasks like content creation and making predictions, rather than processing large datasets for improvements. They claim that the advanced memory in their chips allows them to run Meta’s Llama AI model faster than some Nvidia options, so it’ll be interesting to see how it stacks up against Nvidia’s chips.