Let's talk large quantitative models. How is what you're building different from the large language models we've all had to wrap our heads around?
Caroline, good to see you, first of all. Great to be here live on set in the world of Zoom.

It's great to be in person for once, but this is a fundamentally new area of AI. So I think we're all familiar with chatbots. We're familiar with large language models, a wonderful set of applications, millions of people using them around the world, as it should be.

Now come large quantitative models. And when we look at the majority of business sectors, the economy, we look at biopharma, chemicals, energy, financial services; these are all quantitatively driven, not language-driven.

And so the majority of our economy is actually depending on quantitative relationships. An example would be in biopharma. Let's say we want to create a new drug for a certain cancer, or Alzheimer's, or Parkinson's. We probably don't want to take the drug of a language model trained on Reddit and cat pictures. Probably not, probably not recommended.

But we can use large language models to summarize scientific literature, for example. That's wonderful. But when it comes to actually the quantitative relationship of creating the drug, seeing if it fits at that receptor and developing that all throughout the clinical trial, that's a job for large quantitative models. The data is fundamentally different.

The models are different. The whole field is a new area. And how practical is it in its application right here, right now, when you're raising $300 million? What is…

that for?
Is that for talent? Is that to actually drive the necessary
improvements in the models? What is it that you're going to be fine-tuning?

Well, the good news is, over the last
two or three years since our spinout,

we've now proven with customer after
customer that it actually is having an impact.
For example, UCSF, University of California, San Francisco, one of the
best places for biomedical work. Stan Prusiner is a Nobel laureate.

They're using our software to advance
his drug discovery for neuro diseases right now.
So this is having an impact today. It's running today on GPUs, on Nvidia and
CPUs from Alphabet. It's running right now.

And these are quantitative models, again, not trained on the corpus of
words of the Internet, but trained on

the relationships of chemicals and atoms
and those kinds of dynamic relationships. So it's having impact right now.
Caroline, because what's interesting, you talked about a spin-off. It was
a spin-off from Google. Your chair is Eric Schmidt.

We understand the sort of way in which
the company first was born, and then we
look at what's happening with Google Quantum, the Willow chip.

Everyone's deeply excited about real applications for Quantum in the next
five, ten years. Is that realistic?

Well, it's a great question.
Why do we call our company SandboxAQ?
What's the AQ? AQ is for Quantum.

And so years ago, we already saw that Advanced Compute really has two engines.
It has the engine of large language models, large quantitative models, a la
LLMs, LQMs, but it now also has the QPU.

the quantum and the breakthrough that Google announced just last week. Our colleague, Hartmut Neven, who announced that at the Santa Barbara campus of Google, is a wonderful example of the milestones being achieved in the quantum hardware space. We don't build quantum hardware.

We use quantum hardware as it progresses to business impact. Today, quantum computers are very early, but with Google's show last week with the Willow chip, is that as you scale the number of qubits, you can actually decrease the number of errors. That was a critical milestone, Caroline,

in terms of how we can advance quantum computing. So the future that we see is one that has GPUs and coupled quantum processing units together, working together in a mesh in the hybrid cloud. That, I think, is the future. And we're going to see, I think, very

significant progress over the next five, six years in the quantum hardware itself. All of that means your applications become, well, more broad. How quickly are you going to be making money from this for your investors that we've just been seeing?

Well, the good news is we're already generating revenue today. We're creating a lot of value for our customers today, and so be it in biopharma, be it in the chemicals and materials space, be it in battery chemistry.

We know the importance of batteries, not just for electric vehicles. Actually, the bigger market for batteries are stationary batteries to store energy from renewable sources. That's actually the bigger market. But we must leapfrog beyond lithium-ion technology. This, again, is a job for an...

Q: In a large quantitative model versus an LLM? And so I think that the number of applications are really numerous, and we raised this money in order to advance our investment in product development. We do want to hire more PhDs, more

software engineers. We have quite a few already, but it's time to hire more.

SandboxAQ Sees a Quantum Computing Future

Welcome to the Bloomberg Technology YouTube channel. Subscribe for original series and mini-docs on game-changing technologies, profiles on innovators, live conversations with experts, and breaking technology news & analysis from Bloomberg TV.

Bloomberg Technology

대규모 정량적 모델(LQM)과 대규모 언어 모델(LLM)과의 차이점에 대해 논의합니다.

LQM은 바이오 제약, 화학, 에너지, 금융과 같은 분야에 중요한 정량적 관계에 초점을 맞춘 새로운 AI 영역입니다.

LLM은 과학 문헌 요약과 같은 작업에 유용하지만, LQM은 신약 개발과 같은 정량적 작업에 더 적합합니다.

신약 개발에는 언어 기반 정보뿐 아니라 복잡한 정량적 관계를 이해해야 합니다.

3억 달러의 자금은 인재 확보, 모델 개선, 미세 조정에 사용될 예정입니다.

LQM은 이미 UCSF의 신약 개발과 같은 실제 애플리케이션에서 그 영향력을 입증하고 있습니다.

이 모델은 현재 인터넷 텍스트가 아닌 화학 및 원자 관계에 대해 학습된 GPU와 CPU에서 실행되고 있습니다.

고급 컴퓨팅의 미래에서 양자 컴퓨팅의 역할.

샌드박스AQ는 양자 하드웨어(QPU)를 구축하는 것이 아니라 성숙해짐에 따라 이를 활용하는 데 중점을 둡니다.

Google의 Willow 칩은 큐비트 수가 확장됨에 따라 양자 컴퓨팅의 오류를 줄이는 데 진전이 있음을 보여줍니다.

미래에는 GPU와 QPU가 함께 작동하는 하이브리드 클라우드 환경이 등장할 것입니다.

샌드박스에이큐는 이미 다양한 분야의 고객을 위해 수익을 창출하고 가치를 창출하고 있습니다.

리튬이온 기술을 넘어서는 것을 목표로 배터리 화학 연구에 LQM이 적용되고 있습니다.

모금된 기금은 더 많은 박사 학위와 소프트웨어 엔지니어로 팀을 확장하는 데도 사용될 예정입니다.