ODSC’s AI Weekly Recap: Week of May 24th

5 min readMay 24, 2024

Every week, the ODSC team researches the latest advancements in AI. We review a selection of the most impactful research, trending open-source projects, and significant startup investments. We carefully curate and share a selection of the most impactful developments, bringing you the insights that matter most to the AI and data science community.

Open Data Science Blog Recap

OpenAI Introduces GPT-4o to the World (Source)
Senate Leaders Unveil $32 Billion Plan to Address AI Issues (Source)
Breaking: OpenAI Disbands Team Focused on Long-Term AI Risk (Source)
AI Godfather Sees Need for Universal Basic Income Due to AI (Source)
New AI Safety Pledge Made by AI Giants at AI Seoul Summit (Source)
Microsoft Unveils Copilot+ PCs (Source)
Scarlett Johansson “Shocked” at OpenAI’s Chatbot Voice (Source)
IBM Announces Major Updates to watsonx Platform (Source)

AI News Highlights

Advertising That Mixes Reality with A.I.
AI eats the web
Slack users are horrified to discover messages used for AI training
Can Artificial Intelligence Make the PC Cool Again?
I Tried Microsoft’s New AI-Focused PCs. Windows Is Exciting Again.
AI computers could run in extreme environments like Venus thanks to heat-proof memory device
Microsoft’s AI chatbot will ‘recall’ everything you do on a PC
Cancer patient uses AI to help family remember him
Microsoft announces new PCs with AI chips from Qualcomm
The world is ill-prepared for breakthroughs in AI, say experts
Holistic AI Is Now H, a New French Startup With $220 Million Out of the Gate
Europe sets benchmark for rest of the world with landmark AI laws
AI-Translation Startup DeepL Announces $300 Million Investment at $2 Billion Valuation

New Trending Open Source Projects

Efficient-KAN is an OS pure-PyTorch implementation of the Kolmogorov-Arnold Network (KAN) paper mentioned in this week’s research section. It’s designed to reduce memory usage and computational complexity. It reformulates the original KAN computations to improve efficiency while maintaining compatibility with typical neural network operations, making it suitable for various machine learning tasks
CoreNet, a versatile deep neural network library developed by Apple, supports a wide range of tasks including training large language models and object detection, and has gained popularity for its robust features and open-source nature
This AI Reader open source library converts any URL into a format that is friendly for large language models (LLMs) with a simple prefix. It also allows LLMs to access and process the latest web content by searching the web for given queries
Perplexica is an AI-powered search engine that serves as an open-source alternative to Perplexity AI. It uses advanced machine learning algorithms to refine search results, supposedly ensuring accurate and relevant answers from web queries while supposedly maintaining user privacy
Firecrawl is yet another open source project that converts entire websites into LLM-ready markdown or structured data. It allows users to scrape, crawl, search, and extract data from websites using a single API
TimesFM (Time Series Foundation Model) is a pretrained model developed by Google Research for time-series forecasting. It offers efficient and accurate forecasting by leveraging a decoder-only architecture, making it suitable for various forecasting tasks with publicly available checkpoints and comprehensive usage documentation
Secret Llama is a fully private LLM chatbot that operates entirely within a browser, eliminating the need for a server. It supports models like Mistral and Llama 3, ensuring user privacy by keeping all conversation data on the local device
Cohere Toolkit is a collection of prebuilt components designed to facilitate the quick development and deployment of Retrieval-Augmented Generation (RAG) applications. It provides tools for building customizable interfaces and backends, supports various model providers, and includes deployment guides for production environments

New Trending Research From Around The Globe

The hype around Kolmogorov-Arnold Networks (KANs) as next-gen Large Language Models (LLMs) stems from their unique approach to model architecture and potential advantages over traditional models like Multi-Layer Perceptrons (MLPs). KANs are based on the Kolmogorov-Arnold representation theorem, which allows them to use activation functions on edges rather than nodes, leading to improved accuracy and interpretability with fewer parameters — Paper
Pytorch-Wildlife: A Collaborative Deep Learning Framework for Conservation — Paper
OLAPH: Improving Factuality in Biomedical Long-form Question Answering — Paper
What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functions — Paper
End-to-End Real-World Polyphonic Piano Audio-to-Score Transcription with Hierarchical Decoding — Paper
EHRMamba: Towards Generalizable and Scalable Foundation Models for Electronic Health Records — Paper
TerDiT: Ternary Diffusion Models with Transformers — Paper
TimeX++: Learning Time-Series Explanations with Information Bottleneck — Paper
TimeX++: Learning Time-Series Explanations with Information Bottleneck — Paper
Joining up the scattered anticancer knowledge on auraptene and umbelliprenin: a meta-analysis — Paper
OLAPH: Improving Factuality in Biomedical Long-form Question Answering — Paper
SirLLM: Streaming Infinite Retentive LLM- Paper
AutoSoccerPose: Automated 3D posture Analysis of Soccer Shot Movements — Paper

Start-Up Funding News

San Francisco-based Scale AI, a provider of data-oriented platform that assists in the development of AI applications, raised over $1.5B in Series F
Paris-based Holistic, a provider of artificial intelligence models within the technology sector and helps in data analysis, raised $220M in Seed
Massachusetts-based Sumo, an AI-powered music startup that hopes to allow anyone to create music, raised over $125M in Private Funding
London-based LabGenius, user of a machine learning-driven evolution engine to develop protein therapeutics, raised over $73M in Series B
California-based Praktika, an automated 1–1 tutorship powered by genAI Avatars, raised over $36M in Series A
California-based Coactive AI, a machine learning platform that unlocks analytics and insights from unstructured image and video data, raised $44M in Series B
California-based WitnessAI, a builder of guardrails that make AI safe, productive, and usable, raised over $27.5M in Private Funding

See you in London!

Join us at ODSC Europe to get hands-on with the cutting-edge AI tools and techniques that transform how we work and connect.

Blogs or News to Share?

If you have any blogs, industry news, product launches, job postings, fundraising news, or open-source launches to add to our newsletter then please send them to: blogs@odsc.com .