Vera

Some big news from the Vera team today: it's official!

Just like the internet, Generative AI is "not a fad", and ignoring its popularity can create new security risks. When "22% of employees admit to knowingly violating company rules on the use of generative AI", it may be time to find a way to let your teams use these tools safely.

Chatbots can be a good way to improve customer service and get timely information to those in need. But, especially in high risk use cases, rushing to market without sufficient guardrails has consequences…

This past week, I had the delightful opportunity to share our work at Vera with Women in Data Science (WiDS) Worldwide at the University of North Carolina at Charlotte for their Women in Data Science conference. Over the course of my hour-long keynote, I revisited a talk I've given dozens of times, to update it with all the progress that's been made since 2019. In short, I'm now much more optimistic than I used to be on the direction of this field, but it's going to take all of us to ensure a "safe-enough" future for AI and society. I asked GPT-4 (through the Vera platform, of course!) to summarize my speaker notes*, and it came up with the following: "The speaker, an early tech whistleblower, discusses their work in shaping safe AI across non-profits, government, and industry, now by conforming large language models to corporate policies and minimizing risks. The talk highlights the complexities and rapid advancements in AI technology and mitigation strategies, debunking myths about AI's potential to either solve all problems or lead to an apocalypse. It addresses issues such as mathematical complexity, ethical concerns highlighted by studies like Gender Shades, and the need for robustness in AI models. Societal challenges like deepfakes, civil liberties, and economic inequality are also discussed, along with the importance of human oversight in AI and the evolving technical and ethical landscape. The speaker urges industry commitment to continual improvement and standards, and emphasizes individual involvement in policy-making and the significance of voting to shape AI governance." *lightly edited because, well, AI will never be perfect!

Good things come to those who wait, and we know you've been patiently looking forward to the latest release from Team Vera. This one is a big deal, and we can't wait to share it with the world! Take a look at our brand new chat UI...

Today on the Vera blog, a letter from our CEO about how AI tools and standards must be aligned with science, but also easy to deploy and use. Vera is a "gummy vitamin" for enterprise... it tastes great, but it's good for you, too.

Model moderation is... really hard. But with Vera's team of experts, we can help you map out the universe of exactly how you'd like your models to behave. To illustrate this process, we worked with Claire Geist, formerly of Twitter Cortex, Clarifai, and iMerit to develop a "one-click" policy on illegal substances, and you can read all about it in our latest blog post:

First February Release Notes: Growing, learning, adapting, and remaining focused on the customer experience.

This week's release contains something unique and exciting for anyone hoping to enjoy AI's latest and greatest. As new models enter the market, no one wants to be locked into a single vendor. With Vera's model router, your prompts get sent to the best (or fastest, or least expensive) model for the job, and now it happens without losing the context of your prior queries. Read on about our approach to LLM memory on the Engineering Blog: