As with most companies that are releasing AI assistants, the alignment issue in regards to what it will answer is horrible. I am getting constant “Sorry, I can’t answer your question. Please rephrase it and try again.” and “Sorry, I can’t assist with your question as it violates Epic’s terms of use.” responses when trying to ask various questions about the engine and coding features into the game. Two such examples include:
- “how do i enable and use the SearXNG mcp server for online research”
- “how do i enable and use my web search mcp server for online research by using the MCP plugins that are within unreal engine 5.8”
One of the biggest problems that LLMs have currently (and why I started using a locally hosted, uncensored model even if I am only using an RTX 3080) is the amount of utter frustration I have to get it to actually answer questions about various topics (you try getting any brainstorming assistance when discussing a grimdark fantasy novel).
My personal philosophy is that your morals are not my morals, so don’t try to enforce your beliefs on me. So while having the EDA is a good idea, because of corporate philosophy and risk-aversion with how it might be used, it ultimately creates a tool that is mostly worthless. And like other companies investing heavily into AI and LLMs, congratulations on wasting money on development.
Oh, and why do I want it to browse the web? Because the EDA was trained on UE5.6 and does not have up-to-date information on the engine nor third-party plugins.