[tomek@vz]

I get it. Support is often viewed as a cost center, and agents are often working against a brutal, endlessly increasing backlog of tickets. There is pressure at every level to clear those tickets in as little time as possible. Large Language Models create plausible support responses with incredible speed, but their output must still be reviewed by humans. Reviewing large volumes of plausible, syntactically valid text for factual errors is exhausting, time-consuming work, and every few minutes a new ticket arrives.
Companies must do more with less; what was once a team of five support engineers becomes three. Pressure builds, and the time allocated to review the LLM’s output becomes shorter and shorter. Five minutes per ticket becomes three. The LLM gets it mostly right. Two minutes. Looks good. Sixty seconds. Click submit. There are one hundred eighty tickets still in queue, and behind every one is a disappointed customer, and behind that is the risk of losing one’s job. Thirty seconds. Submit. Submit. The metrics do not measure how many times the system has lied to customers.
↫ Kyle Kingsbury

Ništa neočekivano.