Date: Thursday 18 January 2024
Time: 10:00-12:00
Place: KickOff Web Conference
The following NGI Search project beneficiaries have been introduced. To listen to a podcast, please use the reader below:
AALLM (Adversarial audit of LLM-based search engines)
Recent performance leap inLarge Language Models (LLMs) led to build chatbots and serach engines relying on them. These tools are widely used in this society. However, people are not always aware that the answers to critical questions they ask can be erroneous or biased. AI Forensics aims at providing proof of this undesirable behaviors to hold big tech companies releasing these tools accountable. Our focus will be to analyze Microsoft Copilot's answers on european elections' questions.
Listen to the podcast presentation:
00:00
00:00
Better Food Search
Better Food Search (opengoods.org) allows users to search for specific ingredients, such as peanut butter, leveraging structured data and categories to provide detailed results.
Listen to the podcast presentation:
00:00
00:00
E.V.A.Gallery - European Visual Arts Gallery
European Visual Arts Gallery (E.V.A. Gallery) is a European wide research project, exploring latest art and cutting edge online tech. Artists and gallerists can make virtual exhibitions or create entire virtual galleries to present art in 3D online space. They can use AI tools to protect their art from plagiarism, also mint, exhibit and sell their own NFTs. The project is aimed at new ways of presentation and protection of visual data within the next generation internet.
Listen to the podcast presentation:
00:00
00:00
InDi
lnDi will develop an open-source search engine with key principles such as explainability, human-in-the-loop, anonymity, fairness, and inclusiveness. lt will involve a community of voluntary reviewers to manually assess Al re-rankings. A blockchain ensures fair, anonymous reviews, promoting participation from discriminated minorities. Top reviewers receive crypto-assets, initially with moral value, potentially gaining extrinsic value, akin to Bitcoin's evolution.
Listen to the podcast presentation:
00:00
00:00
OKLLM (Online Knowledge Search for LLM innovations)
The increasing availability of LLMs is opening up a new era in online content discovery. However, LLMs are computationally intensive. Worse still, the elements generated by the AI may contain biases or false information.
As part of this research, the project aims to fill these gaps through knowledge distillation, knowledge graph generation and verification, bias detection and transfer learning.
Listen to the podcast presentation:
00:00
00:00
SCION Browser
SCION Browser enables path-aware retrieval of web resources over the SCION Internet architecture.
Listen to the podcast presentation:
00:00
00:00
Spare Cores
Spare Cores, a Python-based open-source ecosystem, offers standardized inventory and performance evaluations for compute resources of public cloud and server providers, and helps with selecting and actually starting the optimal instance type for containerized tasks, such as ML model training, ETL processes, rendering videos, traditional CI/CD tasks for software developers, or even microservices with varying compute requirements.
Listen to the podcast presentation:
00:00
00:00
Trust4AI (Trustable AI-Driven Internet Search)
Large Language Models (LLMs) like ChatGPT are changing how we search and decide online. Integrated into Bing, they show promise but raise safety and trust concerns. The EU AI Act aims to regulate AI, demanding trustworthiness from companies. Currently, ensuring AI compliance is manual, slow, and prone to errors. This project will investigate automating compliance checks using metamorphic testing to make the process more efficient and reliable.
Listen to the podcast presentation:
00:00
00:00
Visio Creativia
Originally called MetaVision, this project provides decentralized solutions, including blockchain and distributed ledger technologies.
Listen to the podcast presentation:
00:00
00:00
WASPer
With the leaps in generative AI and the ease of access to it, we see an expected rise in its utilization as part of cognitive warfare efforts.
Coordinated communication attacks are a segue to state capture and they have become easier to design.
The WASPer series of models and taxonomies helps media and communication professionals both in the public and private sectors, by analysing whether content is synthetic, whether it contains propaganda, and if it does - what type of propaganda it is.
Listen to the podcast presentation:
00:00
00:00