The Deeplocal Séance Bot uses a large language model trained with “personas” from spooky literature, urban legends, and history. Think Dracula or Frankenstein answering questions about their food preferences.
Experience design company Deeplocal said in an email that it used speech-to-text models to recognize people’s questions. It also rigged the table with magnets to move the planchette.
The company said in a blog post that it is experimenting with LLMs to flag abuse of its products to enforcement teams. “Our aim is to be able to rapidly build and train a model in a matter of days — instead of weeks or months — to find specific kinds of abuse,” it said. “We’re still testing these new techniques, but the prototypes have demonstrated impressive results so far.”
Google did not specify which of its many LLMs it is using to track misinformation, nor offer examples of specific ways they could be used to target it.
Crusoe Energy, a former crypto miner turned AI server landlord, reportedly raised $200 million to buy Nvidia’s H100 chips and will use the hot commodity as collateral. Bill Libby, CEO of Upper90, which provided the loan to Crusoe Energy, said Nvidia chips are a “new asset class.”
The Verge’s Elizabeth Lopatto reported in August on another company doing the same thing: CoreWeave, which also uses H100 chips as collateral to get debt to buy more H100 chips. An Ouroboros of AI chips, if you will.
