Spot AI LLM-Based Copilot Examined
Silicon Valley Spot AI has launched "Copilot," similar in both name and goals of the large-language models (LLMs) of Microsoft and Github but aimed for video surveillance.
This note, based on an interview with Spot's co-founder, Rish Gupta, examines Copilot, what it is aiming to do, and how it compares to other offerings / broader trends in VSaaS / VMSes.
For more on Spot AI, see:
Executive *******
***** ** ** ***** * *.* release, *** **** **** *** ***** us ** *** ***** ******** ** test *********, ** ***** **** ******** is * **** ** ***** *** video ************ ****** ** *******, ******** in ******** ** ****** ***** *** searching *******, ** * *********** ******* of * *****/***.
** *** **** ****, *** ******** includes ****** **** ****** ***** **** the ******* ****** / *** ********* in ******* *** ****** ******* ***** users *** ****** *** ***** ***** own *****. ** *** ******** ****, even ***** **** ** ** ******** some ******** **** *** ******* ************* they *** ********* *** ******* ** improving ***********. *** **** **** ************ take *** *** **** **** ********** works **** ** ******** ** *********** how **** **** ** ** ********* demo ****** ** ******* ******** *******.
Analytics *****-** ******** *********** ** *** / *****
*** ******* ***** ** *** ******** is ************* ** *** / ***** essentially ******** ********* ******** **** *** video ********** ****** ****** **** ***-** analytics, ***** ** ***** **** ******* and ****** (*.*., ******** **** *********,********* ***** ****** ******, ***) **** ********* ***** *****-** and *********** ***** ******* ** **** as *** *** ** ** *** cloud, ** ****** ******* **** * few ******.
**** **** **** ***** ** *********** their "***" ** *** **-******* *******:
**** ** *** *** **** **** AI. **** ** **** ** **** been ******** ********* *** *********. ** believe *** ***** ******** ******** ** going ** **** ******* * ***** data ********.
** **** **** **** **** ******* primarily ** ******** ***** *** / Analytics ************:
***** ** **** ** *** ****** why ** **** *** ****** ** build ***** ****** ** ***** **** of *** ******** ********. ** *** much **** ******* ** *** *** we ****** *** ***** *** **** of *** **** *** **** ** video **** *** ***** **.
**** ********* **** **** ***** ******-**-***** ecosystems **** *** ******* ************ ********* into ******** *********, *.*., (******, *********, etc).
Copilot ********
**** **** *** **** ** ** understand **** ** ********* ** ******** video ** **** ***** *** *** results ******* ****** ** ***** ******:
** ***** ***** ***** ******, ** are **** ** ********** **'** ****** to ********** ********** ****'* ********* ** the ******....
*** *** **** ** *** ****** to ***** *** ***** ****** ** time ******** ****** *** ******* *** to **** ****'** ***** ******* ****** to ***** ****** ** ****** ** configure * ***** ** ****** ** their *******, ********, ***.
******* **** *** **** ** **, and *** ** *** ********** *** context ** **** ****** *** ** quickly **** ** ***** *******, ***** search, *** ***** ****** ** *** of **.
**** **** ****** ** *** *****, e.g., **** ** ********* ********'* *** case *** * ****** ** ****, though **** ** ****** ** ******* this *****-**, ** * *** ***** cost *** *** *** *****.
******* ******** "***** **** ***" / custom ********* ***** **** ******** ** objects ***** **. ******* *** ******** in ******** ** **** *** ** included ** *** "***" ******* ** Spot **, ***** ** ***** ******-**** subscription. ***** ** * ***** ** Spot's *** ********** *******:
Copilot ******** ** "*** *********"
*** **-******* ******* ** *** **** differences ******* *** "***" *** "****" licensing. *** ***:
** *** *** ****** ** ** beyond ****** *** ***** *** **** AI *******, **********, ****** ********, ******* plate ***********, *******. ** *** *** trying ** ** ****** **** ****** based *** *****, ** **** *** cases, *** ******** ** **** ********* tend ** ***. ********* **** **** a *** ***** ***** ***** ** an ********* **** *** ******* **** out ** ** ** ****** *** security ** ****** ** ********** ***** towards *** *** *******.
*** ***:
**** ** * **** ***** *** that **** ********** **** * ***** VMS ***** **. **** ********* **** a ****** ***, ***** * **** search ****. ***** **** ** ****** search, **** *** *****, ***** *** video, **** ** ****** **** ** give ****** ** *** *****, *** of **** ** *****. **** ** powerful *** **** *****. **** ** where *** ***** ******* **** **** of *** ***** *****.
** **** **** *** ***** ******* customers ***** **** / *** ******** is ***** **/**. ************, * ******* comparison *******"****" *** "***," **** **** "********" ******** ***** included ** *** "***" ****:
MSRP ******* **********
*** **** ******* ********** ******. ******* ***** ** ****** ***** or *** "****" / ******* *****. For *******, **** ** **** ** retention *** ** *-******* ***, *** Pro ******* ** ~$** **** *** camera / ******* **** ~$*** **** for *** *** *** ****:
**** ** ******** ** **** **** charge ** ****** ************ *** **** cameras *** *********. **** **** **** away **** ******* *** ***** *** want ****.
Performance ******** ** ******* **** ****
***** **** ********** *** ******* ** train / ******* ** *** ******, he ************ **** ** ** ** early ******* *** ** ******** ** improve **** ****:
***** ****** *** ******* ********. *** the ******** *** ********* *** ****** that *** ******* ****** ******* *** they *** ***** ** ***** ****. This ** *** ** *** ******** this ** *** *********, ******** **** to ***** [*******] **** *** *********** that **** ** *.* ** *.* of * **** ***** **********.
*** *** **** ** **** ****** with **, *** **** ** **** it ****** ** *** *** *** more *** ***** ** ** ******* in *** **** ****** ** *****, you **** **** *** **** ******** AI ******* ** **** *******.
**** **** "** *** **** ** is ** *.*, *.*. ** **** be ********* ******** *** ********."
Live **** ** ****
*** ***** ***** **** ***** **** demo *** ********* ******** ****** **** some ************ ** *** ********* / labeling (*.*., * ****** ******* "***********" or *** "****," ***** ** ******** the *****):
*******, ** ********* ***** **** ** models *** ******** ********, *** ********* performance, ** *******, *** **** ********* over *** **** *** *****. *** most ******** ******* *** **** ** time *** *** ******* **** *** iterate *** ******* ** ***** ******.
Copilot ********* **. **** / ****-****** ******
***** **** ********* **** ****** ***** used ****** *** ********, ** ********* it ** ******* *** *** *********** in *** ******** ** ********:
**** ****** **** **** ** ** an ****-****** ***** **** *** **** trained ** *** *** **** ******** objects. **** ******* **** **** ***** on *** ** **** ** **** VMSes, *** ********* ********* **** ***** YOLO *** ***** ********* ******, *** just ** *** ******** *** ** other ********** ** ****.
* ***** ** ** ******** **** [YOLO], ** **** **** ***** ** it. *** *** ********** ** *** are ***** ******** ****** ** ****** and *** **** ******* ** ******* expanding ** ******* * *** ** objects.
[*******] ** **** ** * ***** language *** ******** [**** **] ******* on ******** *** ******** ** ****** to ******** ********** ** ** *****. It ** **** **** * ***** being.
****** ** **** *********** ****-***** *********, *** ** *** ******* ***** comings *** ****** ******** ** ***** analytics *****-** ** ******* / ******* from **** ** ***** ****** *************.
*** **** ** *** "***** **** own" ********* *******, ***:
- **** ** ***** ****** / "******** Recognition" ******
- ***** ** ****** ******* ******
- ******* ***** ******* ****** (***** **** Own *********)
Copilot **. **-****** "****** *********"
************, ** ********* ****'* ******* ********* and *********** ****** ********* *****-** ** cameras:
***** ** * ******* ****** ** AI **** *** *** ***** **** on * ****** **** ******* ******* have **** ****** ***********. *** *** not *** ********** ** *** ***** because ** ********* *********** *** ******* bandwidth ** ************* *** ****** ** growing ** **** ** *** *** cases *** ******* ************ *** *****.
** ******* ** *** ************ *** the ********** ** ***** ****:
******* **** *** *******. *********, *** architecture ***** **** ****, *** ********* of **** **** *********** ***** ********* and ****** **** **** ******* ***** at *** ****. ****** ** ************ which ** ***** ** *** ** some ** *** ****** **** ** take ********* **** ***** ********* ********** wins ***** ***** **** ***** ******* very ********* **** ******.
***** ****** ********** / **** ** run / ******* ** ********* ******** than ******* *************, **-***** / ****** manufacturer ********* *** *** **** ******** used ** * ***** ****** (***:******** ***** ******** ********* ****). **** ** **** ******** ******* they *** *********** ******* **** *** hardware ** ****** / ******* ********* suffices ** **** *****, *** **** have ****** ******** **** *** **** few *****.
*******, ** ****** *** ***** ** analytics **** *** **** / *****-** to *** *** / ***** ** grow ** ********* ***** ***** **** the ********* *** ******** **** ******.
*** **** ** *** ********* *******, see****** ********* ******** **** - ** Manufacturers, ** *********.
~1,000 ******* "*******" ** ********* *****
**** ***** ** **** ******* *** already **** ******* ** **** *,*** objects ** *** *********:
***** *,*** ********* ******** ********* **** happened ** *** ********. **** ** where ** **** ****** ********, ** are ******* **** **** **** ********* about **** **** **** ** **********. That ** ********* ** ******** ** us ********* *** **** ********** ** video ** **** ******.
*******, **** *** ** ***** ***** your *** ********, *** *** ******** will ** ******** ********* ***** *******, as **** ** * ****** ******* than ****** ********* *** ** *** box ******* **** ** ****** ** vehicles.
100% ** ******** ******* ******* ** *** ******
***** **** *** *** **** ****** metrics, **** **** ** **** *** focused ** ********* ***** ******** ******* the *******, ********* ** ** ******* fully ******* *** *******:
** **** ******* *** ******** ** are ***** **** *** *******. ** are ********** ** **** ** * point ***** ***% ** *** ******** will **** ** *** ******* ** not * ******* ******.
** ** ****** *** *****-***** ******** to **** ******** *** *** **** Spot ** ******** **** *** ***** year, ** **** ** ********* ** see *** *** *** **** ** evolves.