During the past week, DEF CON took place in Las Vegas, gathering numerous highly skilled IT engineers with a common objective: to contribute to the creation of a more inclusive and reliable AI environment.
The NVIDIA Generative Team Red Challenge was a prominent part of this event, hosted by a collaboration between AI Village, SeedAI, and Humane Intelligence. The challenge encompassed a series of workshops and training sessions that aimed to both understand and showcase the vulnerabilities present in the current generation of AI technology, which suffers from a bunch of exploitable weaknesses.
The cooperation between hacker communities and enterprises played a significant role in advancing the cause of prioritizing AI safety alongside technological progress, particularly within the realm of cutting-edge generative models.
Thanks to the efforts from the hacker community, they themselves are glad to provide skepticism, independence, and transparency to field-test emerging security standards, making sure everything is rigorously evaluated
On the other hand, the assessment platform developed by Scale AI will be used to test the models created by Anthropic, Cohere, Google, Hugging Face, Meta, NVIDIA, OpenAI, Stability, Microsoft, and others.
As anticipated, The NVIDIA NeMo Guardrails feature was showcased at the event and if you are yet to know about it, it assists developers in effectively guiding and fine-tuning generative AIs within predefined boundaries, resulting in coherent, clear, and contextually relevant outputs.
Furthermore, in addition to their participation in DEF CON, NVIDIA also took part in last week’s Black Hat conference. Here, they delivered a two-day training session focused on machine learning applications and provided demonstrations of web-scale training dataset poisoning. Panel discussions were also conducted to foster the exchange of opinions.
The event organizers are committed to analyzing and sharing their findings with the public. This transparency-driven initiative aims to expedite AI development and comprehensively explore all possible avenues of progress.
