RC Town Newsletter Vol.33

Aljachi No. 32                                                               
Convert PDF Files to HTML! Turn Them into EBOOKs!
Now, let's introduce a project currently under development.
The 'Speaking AI Test e-book Builder' converts PDF files into HTML, making it easy to create e-books. Many contents are created as PDF files using Photoshop, Illustrator, etc. Educational content companies often manually rework these files, which is quite inefficient. Using the EBook Builder, PDF content can be converted into web-based HTML code and edited or added to via the builder. You can also create eBooks that read aloud by uploading MP3 files, and it will show you where you are currently reading. Implementing AI speaking tests on the web is no problem either.

For more information, click to explore the site. (In development)
"The Core of Copilot Runtime Vector Search" Building Basic Knowledge of DiskANN

One of the main components of the Microsoft Copilot Runtime Edge AI development platform for Windows is a new vector search technology called DiskANN (Disk Accelerated Nearest Neighbors). Based on a long-term project by Microsoft Research, DiskANN is a method for building and managing vector indexes within applications. It uses both in-memory and disk storage, mapping quantized vector graphs in memory to high-precision graphs on disk.
ⓒ Getty Images Bank

What is DiskANN?

hile not exactly the same, DiskANN can be thought of as the vector index version of tools like SQLite. By adding it to your code, you can easily search through vector indexes composed of semantic embeddings from small language models (SLM) such as Phi Silica in Copilot Runtime.

It is important to understand that DiskANN is not a database; rather, it is a set of algorithms provided as a tool for adding vector indexes to other storage systems that are not designed to support vector search. Therefore, DiskANN can be seen as an ideal companion to other embedded storage, whether relational or NoSQL key-value stores.

The requirements for in-memory and disk storage help explain the hardware specifications of the Copilot+ PC, which needs twice the memory of previous Windows basic memory and larger, faster SSDs. The CPU requirements are lower compared to other vector search algorithms, requiring only 5% of the CPU used by traditional methods when implemented at scale in Azure services.

Separate storage is needed for the data being indexed. Placing the index and embedding source in separate storage can present its own challenges. When handling regulated data, including personally identifiable information (PII), you must not overlook encrypting the source data. This can add overhead to queries, but it is interesting to note that Microsoft is developing a software-based secure enclave that encrypts both data at rest and data in use, reducing the risk of PII leaks or prompt manipulation by malware.

DiskANN implements approximate nearest neighbor search using the Vamana graph index. It is designed to handle frequently changing data, making it useful for agent-type AI applications that need to index data stored in local files or services like Microsoft 365, including emails or Teams chats.

Top 10 AI Graphic Design Tools (July 2024)

Thanks to the unprecedented impact of artificial intelligence (AI), the digital design realm is witnessing a revolution. AI graphic design tools are reshaping the way artists and designers express their creativity, allowing them to produce more unique designs in much less time. Let's take a look at the top 10 AI graphic design tools that surpass the limits of creative potential.

The transparent screen used by Iron Man... now anyone can use it.

How did you like the RCTOWN newsletter? I, Mr. R, will strive to deliver even more valuable content next time. It's hot and humid, but I hope you're managing well. It's the rainy season, so don't forget your umbrella. On sunny days, remember to wear a hat and use sunscreen. Enjoy lots of delicious food and have a healthy summer. We'll be back with new updates soon! Thank you.^^
Soft Labs Co., Ltd.
partners@rctown.co.kr
#1610, Digital-ro 121, Geumcheon-gu, Seoul
T | 02-6494-2848 F | 02-6494-2847

This email is an information delivery email provided by Soft Labs Co., Ltd.
This email is sent to registered customers through inquiries, job applications, event participation, etc.

댓글

이 블로그의 인기 게시물

11st App UI UX Analysis Review

アルシータウンニュースレター Vol.27

アルシータウンニュースレター Vol.29