Fact-Checking & Model Updates

Hey there.

I'm Drew Barontini, VP of Product at Scite, and I'm going to walk you through a couple of updates that we've made to Assistant.

So if you go into Settings here, you're going to see the settings information for Assistant.

So this is what is used for every Assistant query that you use.

You can change this.

We always encourage people to change these settings and see which different outputs you get.

There's a lot of really powerful controls in here.

So the model selector here is going to show you all the different models that we have available.

Now, we've done some cleanup recently to retire some old models that are no longer supported, as well as introduce a couple of the newest, latest and greatest models, from Anthropic.

So we have all of the standard models that you have come to expect in here.

We're defaulting to GPT5 Nano.

That gives a good balance of both speed and intelligence.

But we also have Clodsonnet 4.6 and Opus 4.0.4.6, which are models, that skew higher on the intelligent side with lower speed.

Specifically with Claude Opus 4.6.

So we're going to select a different model here.

I'm going to select Claude Sonnet 4.6.

You have the ability to change any response length you want.

Maybe you could look at an additional, set of sources.

So maybe we say 50 here.

And I'm going to apply this.

So, so now those custom settings are applied to this specific response here.

So if I want to ask a question like, does trisomy , increase the rate of chromosome mis-segregation?

I'm going to trigger this.

You're going to see that there is custom setting Assistant settings applied to this specific section here.

So you can, for this session, you can see that information starts printing out.

it's looking at 45 different publication sources within the literature for this information.

So you can see that's going to come out, it's going to print the response.

But the other thing that I want to show here, that's something that we're excited about and we're trying to improve further, is to show you what Assistant is doing behind the scenes.

There's a lot that we're doing to look at our research and make sure there aren't hallucinations, to make sure that we're looking at validated references, because that's a tool, that's being used for real research and we want to make sure that that is valid.

And we also want to make sure it's exposed because that's Part of the scientific process is making sure you're putting things out for debate and people can see it and there's visibility and transparency in that.

So you're going to see here, once this gets to that stage, Okay, so what we're going to do here is we're going to start double checking against the sources.

So you can now see if you expand this fact checks.

We're exposing what's happening behind the scenes.

So Assistant is going through and it's looking at all the literature, all the publications and sources, those 45 sources it's referenced and it's going through and determining which ones are verified and you'll see if there are any that are rejected because they're not relevant to, to the individual query.

So this is going to go through and you can see now we're up to 25.

We selected a larger number of publications to look at with a model that has higher, intelligence and a little bit lower speed, which is why it's taking longer.

But depending on the question you can change those settings as you need to.

So you can see it's going through here.

And it did reject one and it's telling you it's, it's content is largely duplicated here.

It's rejecting four so far, five and counting.

But this is really powerful because we're showing what Assistant is doing behind the scenes.

We're letting you look through this because it's your AI research assistant that's going through and looking at the literature and it's determining what's relevant to the given query and trying to answer your question in the best way possible.

So we can keep looking through here and you can see there's a lot obviously that we're verifying, and rejecting because we're looking at a larger publication set in addition to the top 25.

So once that's done, that is going to show you the printed response here with all of your inline citation references.

It's basically fact checked all of that and provided that here.

Now if you were to open up the references panel, you're going to see a section now under each of these references that'll say why was this reference selected?

So it's going to tell you here specifically that this one provides key details on trisomic cells displaying higher rates of chromosome segregation, etc.

Etc.

You have that information here.

You can scroll down further and see it across all these different references.

And we want to continue to bring more of this into the experience, but really excited about this, so please make sure you're trying out new models and you are letting us know how you feel about the fact check.

If you have any feedback, please click on the feedback here.

When you're in the assistant experience in the sidebar, you can submit feedback to us.

We're always happy to hear it, so please reach out if you have anything.

Thank you.