On the other hand, the only way to get good training data is to generate data indistinguishable from the real-world scenario and then have humans mark it up the way you want the system to do it. You might as well have the data actually be from the real world and recoup some of the costs with sales.
As noted in the YT comments, you can also set your brownie pan in a larger pan of water to nagate the hot-metal effect. Might not be quite as precise, but way easier.
This was comments, not posts. They were using a model to approximate the demographics of a poster, then using an LLM to generate a response counter to the posted view tailored to the demographics of the poster.
They likely only have a dialogue script for masculine and feminine, which swaps both player and npc dialogue. They’d need four scripts and cutscenes timings to make both player voices work with both NPC pronoun uses.
JWO hasn’t shut down. The system got polished enough for them to sell it to other companies, so they don’t need their own test-platform locations anymore.
JWO and similar systems do not reduce labor. The people working cashier become customer service attendants. These systems are valuable when the issue is throughput and sales are being lost at peak times. Airport convenience stores and stadium concession stands, for example, can get significantly higher revenue for the same footprint.