HiddenLayer says the Jev model is open to attacks that LLMs stop
The TokenBreak attack works on Jev without changes, but not usually on large LLMs, says Kasimir Schulz.
Claimed, not confirmed
HiddenLayer published a report on Wednesday. It says that the Jev model from TypeSafe, released last month, is more open to some attacks than other LLMs. Prompt injection and adversarial text attacks can change the output of the model and cause incorrect results. Jev is a fast model made for work from machine to machine. TypeSafe did not answer the questions of HiddenLayer.
Sources
Posted
