$CBRS A FURU called this a terrible example and apparently Cerebras building its own agent to demonstrate what its hardware can do was also an offense. LOLOL. Please submit all future demos to the FURU Committee for Acceptable AI Agent Examples. Personal agents are going to be a huge market. People will want them booking dinners, arranging trips, scheduling appointments, shopping, and handling the daily admin that eats up their time. Speed is fundamental to that experience. Nobody wants to wait 5 minutes for an assistant to finish a simple errand. According to @andrewdfeldman , the Cerebras prototype booked the table in 22 seconds. The manual attempt took 37 seconds. Other assistants took up to 7 minutes and 40 seconds. The team combined a task-tuned agent, parallel restaurant searches, advance knowledge of the booking site, and fast Cerebras inference. That is impressive. It shows how their hardware’s speed, paired with good agent design, can translate into something consumers immediately understand and appreciate. Dinner reservations are an excellent example precisely because the task is familiar and shows how "personal agents" could do tasks faster on ultrafast hardware. Everyone understands what it involves, how annoying it can be, and the appeal of having it handled in 22 seconds. As agents take on more complicated errands involving repeated model calls, the time spent waiting for each response adds up. Apparently a personal assistant successfully assisting a person is now “cringe” and somehow reflects badly on Cerebras? Tough crowd. If you want to criticize Cerebras, make the attacks good. At least explain why completing a useful task in 22 seconds is supposed to embarrass the company. I would love to hear that investment thesis.