Spaces:
Running
Running
use term Information Foraging
Browse files
demo/media_analysis_cell.py
CHANGED
|
@@ -414,7 +414,7 @@ The contextual transcription demo above nudged Gemini with a simple text prior (
|
|
| 414 |
further: it lets you describe your **baseline script** for a talk β who is speaking, what you expect to hear, what you already know, and
|
| 415 |
which questions you actually care about β and then asks the model to surface where the session *deviates* from that script.
|
| 416 |
|
| 417 |
-
These structured priors are the heart of the expectation-driven β
|
| 418 |
prediction errors. Instead of a neutral recap, Aileen 3 Core asks Gemini to focus on surprises, newly introduced actors or systems, and
|
| 419 |
concrete commitments, while only briefly acknowledging content that matches your baseline.
|
| 420 |
|
|
|
|
| 414 |
further: it lets you describe your **baseline script** for a talk β who is speaking, what you expect to hear, what you already know, and
|
| 415 |
which questions you actually care about β and then asks the model to surface where the session *deviates* from that script.
|
| 416 |
|
| 417 |
+
These structured priors are the heart of the expectation-driven βinformation foragingβ idea: they turn a long conference video into a search for
|
| 418 |
prediction errors. Instead of a neutral recap, Aileen 3 Core asks Gemini to focus on surprises, newly introduced actors or systems, and
|
| 419 |
concrete commitments, while only briefly acknowledging content that matches your baseline.
|
| 420 |
|