ZKTeco Launches New LLM For Security
ZKTeco says it has launched a 'self-developed' Large Language Model (LLM) that will integrate into its products, a first for a large security manufacturer.
The launch raises broader questions about whether LLMs will find clear use cases in security.
Inside this post, IPVM examines ZKTeco's LLM, what it is supposed to do, and the future of this technology in security.
******'* ******* *** **** ****** *** ChatGPT-powered '******** **' ****************** *** **** ******************** *** ******** ********** *******.
Executive *******
****** **** *** *** '***** *** 3.0' **** ***** '*****' ****** *************, 'interactive' ******** ******* ****, '******' *** shoppers, *** ******* ***** *** *****.
******'* ***** *** **** *** ****** vague, **** ******* ****** *******, ***********, and *********** ***********.
****** **** *** *** ** "****-*********" and "*** ******* ** **" **** ChatGPT ******* ******, ** ******** ** its ***** ******* ******* ****** "**************."
**** *** **** ******** *** ***** eventually, *** ** **** ***** *** tech ***** **** ******* *** ********* purposes. ****** *** *** ******* ** comment ********.
LLM *********
***** ******** ****** (***) *** ** programs ******** ** ********** *** ******** human ********, **** ******* ***** *** best-known ******* ***** ** ** *** (***-*). **** ********* ******* **** ******* of ******** **** ** ***.
ZKTeco ******** ***** ***
****** *** "******** *** ***** ***," touting ** "**** ** ********** **** various *********" *** ** ******* ****************:
** ***** ** *** ****** ** complex ***** ********* *** ******* **** demands,has ******** *** ***** *** (***** ******** *****), ***** ** ********* ** ********** *** **** *** *.* *******. In the future, the BioCV LLM 3.0 large language model will ** ********** into various scenarios of the four core businesses. [emphasis added]
******* *** **** ** ******'* ***-***** software ******** ***** ********** ******'* ****** control, ***** ************, **********, ******* **********, and ***** *********.
****** *********************** *** ***** *** ** **** 2023 ** *** ******* ******** ****,********* ********* "********** *********".
Use ***** ******: *******, *******, ******
************* *** **** ***** ******** (****** vague) *** *****, **** ** '*****' office *************, '***********' ******** ******* ****, 'visitor ********' *** ******* ********, *** 'smart ******':
*** *******, ** ***smart ****** ******* ********** ***** ******, ** *** ******* smart ****** ********** ********* ******* ********, visitor *********** ************, ***** ***********, *********** push ***virtual ********* etc. In the smart ****** ********, it can empower smart office businesses, including intelligent interactive attendance and intelligent ******** ******* etc. In the intelligent ****** ********, it can empower smart ****** businesses, including intelligent advertising push and retail guidance etc. [emphasis added]
****** ****** *** *** ** ******* in "********** ***" (********* ** ******* details) *** **** ** **** ****** across ******* "*********" ** *** ******.
No ****** ******* ** *******, ***********
** ***, ****** *** *** ********* any ******* ** ******* **** *** costs ***** ** **** **** ******* into *** ************* ********** ***** ********.
****** *** **** *** ********* *** technical ********** ***** *** ***, **** as *** **** ** *** ******** data ** *** ***** ** *** processing, ********* ***** ********* ********* **** often ** (*.*. ** *** ** platform***********).
***** ** ** ********** **** **** LLM ** **** ******, ******* ** will **** ** ********* ** ****** customers.
Showcased ** ****
****************** *** *** ** *** ***** at****, *** ***'* ******* ******** **********, publishing **** *****:
**** **** *** ***** ***** * pre-recorded ***** ** *** *** ****** than * **** ****.
Claims "****-*********", "******* ** ** **** ******"
****** **** *** *** ** "**** developed" *** *** "******* ** **" with ******* ******* ******, *** **** "sometimes *** ********* **** ******* *** use ** ******’* *** ** **** scenarios", ***** ******** *&*:
BioCV *** ** * ***** ***** ** ********* ** ********* *** *** ******* ** ** **** ******. When we use BioCV LLM, we will not use OpenAI’s API, but sometimes *** ********* **** ******* *** *** ** ******’* *** ** **** *********, so the interface design is compatible with OpenAI's API, which means that it can be switched to OpenAI's API.
***** *** 是我们自研的大模型，与 ****** 没有关系，我们用 ***** LLM 的时候，就不会用到 ****** 的 ***，但是有时候部分场景我们的客户会要求用 ****** 的 ***，那么我们在接口设计上是兼容 ****** 的 *** 的，就是说可以切换到 OpenAI 的 ***。[******** *****]
******* ******* ** ****, ***************** ******* "********* ******* ** *******" claiming ** *** ************* **** ******/*********, e.g.:
- ***** **:"*** *******'* *********** **** *********'* ****** is ** *** ***** ** *********** and **********"
- ***** **: "*** *******'* *********** **** *********'* OpenAI ** ** *** ***** ** cooperation *** **********"
- ***** **:"*** ******* *** ********* *** ******* OpenAI-related **********, *** **** ** *** interfaces **** **** ****** ** ***"
****** **** ***** ****** **** * broader*** ***** ****** ********* *** **-******* ****** *** ** the ****** ** *******.
**** *** ***** ****** ** ******* whether *** *** *** ** ******* to ***** ******** *************, *******, ** did *** *******.
Other *** ******** ** ********
**** ***** *** ***** ***** ******** manufacturer ** ****** ** *** *** its ********. (*** *********, ****** ******** *,********* ***** ****~$****** ***** *** **** ********* *** ******* ***** $*** ***).
*** ***** **** ******* *********************** * *******-******* ******* ** ****** users ** *** ******** ********** *******, as ** ******** ** *** ******** ****:
*******, ****** **** *** *** ** "self-developed," *** *** ***** *** **** ambitious, ******* ****** **** **** *** cases (******, *******, *******, ***).
Impact ** **** ** ********: ***** *****
***** *** ******* ******* *** *********** so ***, ** ** *** ***** to **** ** ******'* *** **** have * *********** ******.
**** *** **** ***** *** ***** in ********, **** ** ********* ******** assistance ********. *******, *** ************** ****** of *** **** ** *******, ** least ** *** **** ****.
*** ***, *** **** ****** ** LLMs ** ****** ** ** ********* using **** *** ********* ********, *.*.,***** ********* ** *** *****.
*******, **** ******!
* ** ******* ** *** ***** they *** **. *** **** **** it **** ******* ********* ********* ** train, ** ***** *********** ********* (*** beyond **** ** ********* ** ****** security *******) ** *** ** ** an ******* *****.
****, ******* *** **** ******** ** will ****. **** ** *** *** it's **** ** **** ** *******, ChatGPT ***** ********* ***** ****** *********** efforts. ** *** ******* ********* *** we *** **** ***** ** *** risks *** **** ** ****** *******, etc. ***** **** ** * ******* security ********* *** ****** ** ******* serious ******** (*** ******, *** **** ZKTeco).