The internet’s steady fall into the AI-garbled dumpster continues. As Vice reports, a recent stud y conducted by researchers at the Amazon Web Services (AWS) AI Lab found that a “shocking amount of the web” is already made up of poor-quality AI-generated and translated content.

The paper is yet to be peer-reviewed, but “shocking” feels like the right word. According to the study, over half — specifically, 57.1 percent — of all of the sentences on the internet have been translated into two or more other languages. The poor quality and staggering scale of these translations suggest that large language model (LLM)-powered AI models were used to both create and translate the material. The phenomenon is especially prominent in “lower-resource languages,” or languages with less readily available data with which to more effectively train AI models.

In other words, in what the researchers believe to be a ploy to garner clickbait-driven ad revenue, AI is being used to first generate poor-quality English-language content at a remarkable scale, and then AI-powered machine translation (MT) tools transcribe said content into several other languages. The translated material gets worse each time — and as a result, entire regions of the web are filling to the brim with degrading AI-scrambled copies of copies.

One brain to rule them

Two researchers have revealed how they are creating a single super-brain that can pilot any robot, no matter how different they are.

Sergey Levine and Karol Hausman wrote in IEEE Spectrum that generative AI, which can create text and images, is not enough for robotics because the Internet does not have enough data on how robots interact with the world.

The HiP framework developed MIT_CSAIL creates detailed plans for robots using the expertise of three different foundation models, helping it execute tasks in households, factories, and construction that require multiple steps.

“All we want to do is take existing pre-trained models and have them successfully interface with each other,” says Anurag Ajay, a PhD student in the MIT Department of Electrical Engineering and Computer Science (EECS) and a CSAIL affiliate. “Instead of pushing for one model to do everything, we combine multiple ones that leverage different modalities of internet data. When used in tandem, they help with robotic decision-making and can potentially aid with tasks in homes, factories, and construction sites.”

These models also need some form of “eyes” to understand the environment they’re operating in and correctly execute each sub-goal. The team used a large video diffusion model to augment the initial planning completed by the LLM, which collects geometric and physical information about the world from footage on the internet. In turn, the video model generates an observation trajectory plan, refining the LLM’s outline to incorporate new physical knowledge.

This process, known as iterative refinement, allows HiP to reason about its ideas, taking in feedback at each stage to generate a more practical outline. The flow of feedback is similar to writing an article, where an author may send their draft to an editor, and with those revisions incorporated in, the publisher reviews for any last changes and finalizes.