Marktechpost AI Newsletter: Yandex Introduces YaFSDP + Gretel AI Releases a New Multilingual Synthetic Financial Dataset + LaVague’s Open-Sourced Large Action Model Outperforms Gemini and ChatGPT.... and many more....

Want to get in front of 1.5 Million AI enthusiasts? Work with us here

Featured Research..

Yandex Introduces YaFSDP: An Open-Source AI Tool that Promises to Revolutionize LLM Training by Cutting GPU Usage by 20%

Developing large language models requires substantial investments in time and GPU resources, translating directly into high costs. The larger the model, the more pronounced these challenges become.

Recently, Yandex has introduced a new solution: YaFSDP, an open-source tool that promises to revolutionize LLM training by significantly reducing GPU resource consumption and training time. In a pre-training scenario involving a model with 70 billion parameters, using YaFSDP can save the resources of approximately 150 GPUs. This translates to potential monthly savings of roughly $0.5 to $1.5 million, depending on the virtual GPU provider or platform.

Yandex has made YaFSDP publicly available on GitHub. ML engineers can leverage this tool to enhance the efficiency of their LLM training processes. By open-sourcing YaFSDP, Yandex aims to foster innovation and collaboration in the AI community, enabling developers to train models faster and cost-effectively.

 Editor’s Picks…

Gretel AI Releases a New Multilingual Synthetic Financial Dataset on HuggingFace 🤗 for AI Developers Tackling Personally Identifiable Information PII Detection. [Notebook Included..]

Detecting personally identifiable information PII in documents involves navigating various regulations, such as the EU’s General Data Protection Regulation (GDPR) and various U.S. financial data protection laws. These regulations mandate the secure handling of sensitive data, including customer identifiers, financial records, and other personal information. The diversity of data formats and the specific requirements of different domains necessitate a tailored approach to PII detection, which is where Gretel’s synthetic dataset comes into play.

Every organization has unique data formats and domain-specific requirements that may need to be fully captured by existing Named Entity Recognition (NER) models or sample datasets. Gretel’s Navigator tool allows developers to create customized synthetic datasets tailored to their needs. This approach significantly reduces the time & cost of traditional manual labeling techniques. By leveraging Gretel Navigator, developers can rapidly create large-scale, diverse, privacy-preserving datasets that accurately reflect the characteristics and challenges of their domain, ensuring that PII detection models are well-prepared for real-world scenarios and unique document types. One such dataset by Gretel is its multilingual Financial Document Dataset, released on the 🤗 platform this week.

Researchers at the University of Illinois have developed AI Agents that can Autonomously Hack Websites and Find Zero-Day Vulnerabilities

We all know AI is getting smarter every day, but you’ll never guess what these researchers just accomplished. A team from the University of Illinois has unleashed AI agents that can autonomously hack websites and exploit real-world zero-day vulnerabilities – security holes that even the developers don’t know about yet.

That’s right, the age of AI hacking is here.

The problem? Current AI hacking agents like the ones using ReAct are basically stumbling around blindly when it comes to complex, multi-stage attacks.

Here’s how it works: These ReAct-style agents iteratively take an action, observe the result, and repeat. Simple enough for basic tasks. But when it comes to the long game of high-level hacking, this approach crumbles for two huge reasons:

1. The context required balloons out of control for cybersecurity exploits. We’re talking pages upon pages of code, HTTP requests, and more to keep track of.

2. The agent gets trapped going down one vulnerability rabbit hole. If it tries exploiting some XSS vulnerability for example, it struggles to backtrack and pivot to attempt a completely different type of attack like SQL injection……

LaVague’s Open-Sourced Large Action Model Outperforms Gemini and ChatGPT in Information Retrieval: A Game Changer in AI Web Agents [📓Colab Notebook included]

LaVague is a comprehensive framework designed to simplify the creation and deployment of AI agents. Its LAM architecture allows developers to build agents capable of performing complex tasks and effortlessly sharing their functionalities. By leveraging LaVague, developers can create powerful, community-shared AI agents with just a few lines of code, offering unparalleled performance in retrieving up-to-date information.

The LaVague framework utilizes a World Model to translate objectives and current web states into executable instructions and an Action Engine to compile these instructions into action code. This setup enables LaVague agents to execute tasks autonomously on the web, significantly lowering the barrier to entry for AI agent development. For instance, creating a Gradio demo is as simple as using the command agent.demo().

A New Era AI Databases: PostgreSQL with pgvectorscale Outperforms Pinecone and Cuts Costs by 75% with New Open-Source Extensions

In a groundbreaking development, Timescale, the PostgreSQL cloud database company, has introduced two revolutionary open-source extensions, pgvectorscale, and pgai. These innovations have made PostgreSQL faster than Pinecone for AI workloads and 75% cheaper. Let’s explore how these extensions work and their implications for AI application development.

Timescale unveiled the pgvectorscale and pgai extensions, aiming to enhance PostgreSQL’s scalability and usability for AI applications. These extensions are licensed under the open-source PostgreSQL license, allowing developers to build retrieval-augmented generation, search, and AI agent applications with PostgreSQL at a fraction of the cost compared to specialized vector databases like Pinecone.

Trending AI Social Media Posts

• This AI Paper from Snowflake evaluated the performance of GPT-4 on Document Understanding tasks, and it seems even models of this parameter count underperform in image-only setup and vastly benefit from providing text in addition to the input image.

• Not Llama 3 405B, but Nemotron 4 340B! NVIDIA just released 340B dense LLM matching the original OpenAI GPT-4 performance for chat applications and synthetic data generation. 🤯 NVIDIA does not claim ownership of any outputs generated. 💚

• ✨ Introducing 𝐎𝐩𝐞𝐧𝐕𝐋𝐀 — an open-source vision-language-action model for robotics! 👐

• Nemotron-4-340B is released today!