Release Notes Version 23.4.5: aka How We Stopped Worrying and Learned to Love Building in Public
Version 23.4.5: Buckle up ’cause this one’s a doozy!
We won’t mince words on this one, because the team has worked SO HARD to get used to our new weekly release schedule. This past week, we added some impressive new functionality that we’re super excited to start showing off to our clients and waitlisters. Possibly the most exciting of which is…
Vera Gets Serious About Embeddings
Our lead data engineer has been quietly cooking up the nuts and bolts of a feature that powers a lot of cool stuff on the back-end. Andrew Savitski (as he is known to his friends), spent a long time researching this new and exciting technology to figure out what approach was best, and guess what! He’s going to write about it on the blog for us soon. But for now, we’ll just share that A) it works, and B) it sets up a critical function that will act as the precursor for a lot of cool features we have planned.
The first of these features is also live, with a fast and flexible way to search the prompt and model response monitoring logs. Want to find all of your team’s prompts that mention “plants”? Cool. You can do this even better now, through the API AND the Front End, and the results will show you anything to do with flowers, tulips, and roses, as shown below:
To get here, we also had to improve the logging capabilities, add filters, and that kind of thing. As always, we’ll continue to improve things as we go!
Routing Display Updates
Every user loves to have transparency into what’s going on in those anticipation-filled moments between when you hit [Enter] and when the platform responds. So, we’re leveling up our transparency game, and giving a little more info on what happens once a question gets sent through Vera to the LLM of your (or your organization’s) choice.
With this UI improvement, we’re now displaying exactly what the Vera platform (or OpenAI, Cohere, etc.) is doing when a prompt is sent for evaluation. Check it out!
One-Click Policies: Drugs & Drug Use
You all know by now that Vera enforces company policies for LLM use in the enterprise. But what exactly are those policies and how do they work?
Well! We plan on telling you all about it in an upcoming blog post. For now, let’s just say that in Vera, you can pick one of many pre-set policies, fully customize them, and even create your own from scratch. More detail to come here, but in this release, we updated and added a comprehensive filter for drugs, paraphanalia, addiction, and other taboo terms that you may or may not want your users to access. You do you.
Moar Model Routing Metrics
We’re always working to improve the quality of our model router, and this week’s release opened up a whole world of configuration options you can now tinker with in the API to optimize model routing according to your preferences.
And that’s all for this week! We’ll be in touch next Monday with more new features and bugfixes; and we hope you’ll follow us along in this journey!
P.S… Join the waitlist.
