3 min read
Releasing Basewell Intelligence 3.0
By Evan Stewart — July 16, 2024
Today we're releasing model update 3.0, with improved support for ambiguous, nuanced, casual, real-time queries. Basewell now understands questions in your context (how you work and speak) because smart tools shouldn't require prompts or habit changes to work well.
We’re achieving this goal through two critical improvements:
Reflective, not reactive query architecture
Consensus paramaters
Reflective, not reactive query architecture
Most AI tools require detailed prompts for accurate responses, burdening users with complexity and decreasing reliability. This process is inherently reactive, forcing users to tweak their actions after negative experiences. By interpreting questions with context on each user, company, and other paramaters, Basewell becomes reflective: rephrasing, prompting, and fixing errors before they impact response accuracy.
Reflective, not reactive architecture is critical for long-term viability, ensuring natural language can actually be natural. After all, most human questions are casual or ambiguous. Coworkers already have this context, which is why they can often produce answers. Now your learning tools can do the same.
Consensus paramaters
When working in person or in groups, solutions to hard problems are often realized in consensus. Multiple people agree that a certain direction is the right one. Alternatively, when working with async tools, consensus can be hard—or impossible—to find. This issue leads to:
💬 Reliance on outdated communication tools
📆 Delayed answers (even when time is of the essence)
🤔 Loss of context
What if there was a way to get answer consensus, without losing context, measured against verified company info, in seconds instead of days? With our 3.0 model update, Basewell does just that; double-checking for accuracy before returning an answer.
Behind the scenes, Basewell automatically runs multiple checks-and-balances to ensure responses are contextually relevant, verifiably accurate, and retrieved correctly. After consensus has been reached across these checks, a response is returned in language familiar to the user.
Once again, everything happens automatically. Users can just focus on asking questions and getting back to doing their best work.
Through understanding questions regardless of "prompt" and ensuring accuracy through a series of response checks-and-balances, this new model is a substantial step towards making Basewell your most helpful co-worker—not just a co-pilot.
Learn more by giving Basewell a try below: ↓
