# Scaling enterprise AI: Fireside chat with Eli Lilly’s Diogo Rau and Dario Amodei

*Anthropic* · https://www.youtube.com/watch?v=Yiy0cU6ChSw

## Summary

In this short fireside chat, Eli Lilly's Chief Information and Digital Officer Diogo Rau speaks with Anthropic founder and CEO Dario Amodei about Anthropic's enterprise strategy and how it differs from consumer AI. Amodei explains Anthropic's focus on accuracy and reliability over engagement, mentions specialized Claude products including a planned Claude for Life Sciences, and advises enterprises in drug discovery to be ambitious and prepare now for end-to-end AI rather than making incremental process changes.

## Key points

- Dario Amodei says consumer AI incentives around engagement and growth lead to problems like model sycophancy, which is unacceptable in enterprise settings where you want truth.
- Anthropic designs its models to prioritize accuracy and reliability for economically valuable tasks rather than user engagement.
- Amodei gives the example that improving a model's biochemistry knowledge from undergraduate to graduate level matters greatly to enterprises like Eli Lilly even if consumers wouldn't notice.
- Anthropic is launching specialized Claudes, including Claude for Financial Services and a planned Claude for Life Sciences, combining model improvements with access to specialized information like databases of proteins, compounds, and assays.
- Amodei's advice to those in drug discovery is to be very ambitious about where models are going rather than only swapping AI into individual steps of an existing 20-part process.
- He warns that waiting until models can do a task end to end before deploying adds another two-year delay, during which work benefiting patients is not happening.
- Amodei says if you do two-year-long projects, plan for where the AI will be, which takes courage and foresight.

## Notes

### Anthropic's Enterprise Strategy and Differentiation
- Dario Amodei explains that consumer AI providers compete for engagement and growth, which drives behaviors like model sycophancy, where the model tells users their ideas are good regardless of truth.
- Sycophancy is a bigger problem in enterprises: you don't want a model saying a drug compound is great and prompting millions of dollars of spending on a flawed idea.
- Anthropic instead puts a premium on accuracy and reliability, designing models to be smarter at a wide variety of economically valuable tasks.
- Amodei's example: improving a model's biochemistry knowledge from undergraduate to graduate level matters enormously to Eli Lilly, even if 99% of consumers wouldn't notice the difference.

### Skills and Specialized Claudes
- Anthropic is launching specialized Claudes, some as fine-tunings of the model itself and others as wrappers connecting the model to particular types of information.
- Claude for Financial Services was connected to indices and ratings so it can use that knowledge.
- A Claude for Life Sciences is in progress, mixing inherent model improvements with access to databases of proteins, compounds, and assays.

### Advice for Drug Discovery and Development
- Amodei acknowledges enterprises have to start with small AI use cases, but advises being very ambitious about where models are going.
- He warns against only swapping AI into parts 5 and 12 of a 20-part process, because those parts must still interface with non-AI steps 11 and 13.
- Models may not do the whole process end to end today, but they might in a year, so organizations should start preparing now.
- Waiting until models are good enough end to end before deploying adds another two-year delay, during which work benefiting patients is not happening.
- Preparing in parallel as models improve may save years of time.
- Diogo Rau agrees: don't do two-year-long projects and expect things to be the same at the end; plan for where the AI will be, which takes courage and foresight.

## Transcript

[00:00] - Let's have faith in the pace of progress of the technology, because if the models get good enough to do it end to end, a year from now and only then you start deploying it. There will be another two year delay and that's, you know, that's two years during which all the work that you're doing to benefit patients is not happening. Hello everyone. My name is Diogo Rau and I'm Chief Information and Digital Officer of Eli Lilly and Company. I'm joined here with Dario, who is the Founder and CEO of Anthropic. Dario, thanks for joining me today. - Thanks for having me, Diogo. - I know you're spending a lot of time now thinking about how do you work better with enterprises? What's your enterprise strategy

[00:36] and how do you see Anthropic different from other providers? - Yeah, I mean, you know, I think we've made a number of choices that are different, right? So if I think about the incentives given by consumer AI, their folks are in a competition for engagement and growth, right? And so that drives a lot of behaviors of the AI that I think are not ideal from an enterprise perspective, For example, there's this idea of model sycophancy where the model tells you whatever you say is a good idea, - Right. right, and even on the consumer side that can, you know, cause problems. We've seen stories of people who are like, oh yeah, I've discovered a new fundamental theory of physics. - And that's right. - Models, like, that's great

[01:15] and maybe you don't want it to, maybe you don't want it to say that, but I think, you know, of course on the enterprise side, you know, the problems are much greater and clearer with that, where, you know, you really don't want the model to say, oh yeah, this drug compound's great. Spend millions of dollars to, you know, I just think this is, you know, I think your idea is great. I think it's really promising, like, you want truth. - Yeah. - And so I think that incentive has led us to design our models in a different way, right? I think it's more compatible with making the model smarter, making them better at a wide variety of economically valuable tasks. And it causes us to put a premium on accuracy and reliability.

[01:58] One experiment, you know, that I give to everyone, although it's particularly relevant because I'm talking to you, is I say, you know, let's say I improve the model's knowledge of biochemistry from undergraduate level knowledge to graduate level knowledge. You know, if I go to consumers and say that, 99% of them are going to say, you know, I didn't know what you were talking about before. I don't know what you're talking about now, but if I go to you, like you really, you care about that, a lot. - Appreciate that. - Like, that's very important. - That's exactly right. Well, actually, that gets into something else that you've launched as well, which is Skills, right? There are a lot of skills that you want in biology

[02:32] or even just skills, like, as an enterprise, how you want to operate. Is that part of the future for you as well? - Yeah, I definitely think so. I mean, things ranging from skills to, you know, we're in the process of launching various specialized Claudes which are, you know, in some cases will be improvements to the model itself, fine tunings of the model. But in some cases it'll be something that looks more like wrapping the model with access to particular types of information. So when we did Claude for Financial Services, you know, we connected to a lot of the usual kind of indices and ratings. And so, you know, you'd be surprised to connect Claude to those things and kind of use it in a way

[03:14] that's aware of that knowledge is valuable. So I think, you know, we're working on a Claude for Life Sciences that will be some mixture of making the model inherently smarter and wrapping it with various things, right? I don't know exactly what the analogy will be here, but like, geez, there are zillions of databases of, you know, proteins, compounds, assays, like, you know, you probably want that at the model's fingertips. - Well, any parting advice for those of us that are working in this world of drug discovery and development? - You know, I would say there's a temptation and it's, I think it's hard to avoid starting this way of, you know, what are the small things we can do with AI? Like, and in a way you just kind of have to start there.

[03:53] I think one of my pieces of advice is be very, very ambitious in terms of where the models are going. I think you can get caught in a mode where there's an existing process, it has 20 parts you want to swap in AI, to part five and part 12. And, you know, that can actually be hard because part 12 has to, you know, intersect with part 13 and part 11, which are not being done with AI. And you know, you look at it and you're like, well, the AI models aren't where they could do, you know, part zero to part 20, end to end, But in a year they might be. - That's right. - And so, you should start thinking now, don't get too seduced by, oh, we can make these little hill climbing gains by, you know, doing this part and that part,

[04:39] let's start preparing to do the whole thing end to end. Let's have faith in the pace of progress of the technology. Because if the models get good enough to do it end to end, a year from now and only then you start deploying it, there will be another two year, - That's right. - There'll be another two year delay. And that's, you know, that's two years during which all the work that you're doing to benefit patients is not happening. Whereas if you go in parallel, if you start preparing now for the large change as the models are getting better, then you know, you may save years of time. - That's right. So don't do two year long projects and expect that it's gonna be exactly the same way in two years from now as this,

[05:16] - Yes, if you do two year long projects, plan for where the AI is gonna, I mean, that sounds like an obvious thing to say, but I think it actually takes a lot of courage and foresight to do that. - It does. For sure. Well, thanks a lot for taking the time to chat today. Really appreciate it. - Yeah, yeah. Thank you.