What are the contributions you’re excited to make?
나는 웹기술과 ai 기술에 모두 ah 그러니 두가지를 결합하는 프로덕트를 만드는 데에 기여하고 싶다
또한 optimization에도 관심이 많은데 Quantization gptq를 이용하여 비용절감이 기여하고 싶다
웹 생태계에서 사용되는 AI제품의 비용효율적인 시스템 설계에 기여하고싶다. 나는 웹 기술과 AI기술에 배포와 모델 양방향에서 Web AI 프로덕트를 개선할 수 있다. Web api들과 프롬프트 엔지니어링에 경험이 있어서 해당 부분에 참여할 수 있다. 또한 LLM 서비스의 비용 최적화에 기여하고 싶다. 백엔드만을 이용한 GPU 컴퓨팅 자원 뿐 아니라 WebGPU를 활용한 프론트엔드 자원에 컴퓨팅 파워를 분할하는 연구나 GPTQ같은 Model Quantization또는 knowledge distillation 이용하여 모델과 GPU비용 최적화도 실제 회사의 프로젝트에 적용해보고 싶다.
현재로서 scaling이 AI 산업에서 중요한 역할을 하는 이상. 일반 목적 ai에서 빅테크를 비용을 고려하며 같은 제품으로 경쟁하기에는 무리라고 생각한다. 현대 귀사가 제공하는 일반목적 AI뿐 아니라 프롬프트 engineering을 이용한 chatpdf같은 특수목적 ai가 사용자들의 결제를 이끌어내는 데에 더 효율적인 프로덕트이다. 그래서 귀사에서 다양한 구체적인 타겟을 가지고 AI 제품을 개발하는 방식으로 여러 타겟층을 커버하고 싶다.
Pi는 다양한 유저층을 커버할 수 있지만 그만큼
그래서 나는 Pi AI에 좀더 타겟 대상을 구체화하는 동시에 타겟을 다양하게 하는 전략으로 다양한 사용자층을 커버할 수 있도록 특수목적 AI들 개발에 기여하고 싶다.
Pi 외에
What are the contributions you’re excited to make?
I’m excited to optimize the cost of AI products deployed in the web ecosystem. I have experiences in AI technologies (Model Optimization) and Web technology (Deployment), thus can improve web AI products in both ways. I am experienced with Web APIs and promt engineering skills like LangChain which can contribute Web based AI application.
Model optimization is critical to serving multiple users simultaneously. Optimizing model and GPU costs using model quantization such as GPTQ or knowledge distillation, and applying them to project Pi is necessary. I am capable of sharing brilliant optimization ideas from the Web perspective. For example, I suggest using front-end Client WebGPU during inference as well as using GPU on the back-end. Transformers.js which is based on WebGPU for browser inference is used in my project.
I can share my LangChain development experience to make Inflextion AI cover a wider user base. It is hard to compete with Big Tech companies in the general-purpose AI field like ChatGPT where scaling plays an important role. Today, not only general-purpose AI, but also specific-purpose AI such as ChatPDF using prompt engineering is an attractive product that makes personal users pay. Refining Pi's user target or making new specific-purpose services will increase sales. Prompt engineering using LangChain that I can contribute to is crucial for such specific-purpose AI.
Antropic
Why do you want to work at Anthropic? (We value this response highly - great answers are often 200-400 words.)
I'm excited to optimize the cost of AI products deployed in the web ecosystem. I have experience in AI technologies, specifically model optimization, as well as web technology and deployment. This allows me to improve web AI products in both ways. Here are some ways I can contribute to optimizing costs and improving web-based AI applications:
- Model Optimization: Optimizing models is critical for serving multiple users simultaneously. I can apply techniques such as model quantization, like GPTQ or knowledge distillation, to optimize model and GPU costs. This can help reduce the computational resources required for AI applications.
- Web APIs: I am experienced with Web APIs, which are essential for building and deploying web-based AI applications. By leveraging efficient API design and implementation, we can optimize the performance and cost-effectiveness of AI services.
- Front-end Client WebGPU: I suggest using front-end Client WebGPU during inference. This can offload some of the computational load to the client's device, reducing the reliance on back-end GPU resources. This approach can improve performance and cost efficiency.
- LangChain: I have expertise in LangChain, a prompt engineering skill that can contribute to web-based AI applications. Prompt engineering is crucial for specific-purpose AI, such as ChatPDF, where personalized user interactions are important. By leveraging LangChain, we can refine Pi's user target and develop new specific-purpose services to increase sales.
By combining my expertise in AI technologies, web technology, and optimization strategies, I can help enhance the cost-effectiveness and performance of web-based AI products.