Global AI Weekly
Issue number: 67 | Tuesday, August 27, 2024
Highlights
}
Apple Intelligence is coming. Here’s what it means for your iPhone
Apple is about to launch a ChatGPT-powered version of Siri as part of a suite of AI features in iOS 18. Will this change the way you use your phone – and how does it affect your privacy?
theguardian.com
}
Toward a code-breaking quantum computer
Building on a landmark algorithm, researchers propose a way to make a smaller and more noise-tolerant quantum factoring circuit for cryptography.
news.mit.edu
}
This Week in AI: Gen Z has mixed feelings on AI
This week, surveys suggest that Gen Z — regularly the subject of mainstream media fascination — has very mixed opinions on AI.
techcrunch.com
}
How Newtons method for estimating π shows it doesn't really matter if LLMs are Stochastic Parrots
The phrase "stochastic parrot" has emerged as a misguided but oft repeated pejorative for Large Language Models, suggesting that these models mimic human language without true understanding, much like…
linkedin.comResearch
Exploring mixture of experts models
As foundational models get bigger it becomes harder to train them. We simply don't have enough compute power to train models in a reasonable time and budget. To fix this, people have turned to mixture of experts models. But how do they work? This paper explains one approach to use mixture of experts. It's advanced, but cool to see how much power you can push out of a deep learning model!
arxiv.orgVideo
}
Microsoft's Phi 3.5 - The latest SLMs
In this video I look at yesterday's releases of Phi 3.5 models which include the mini, a new Mixture of Experts model as well as the new 3.5 Vision model.
youtube.com
}
Designing High-Quality Synthetic Data for Training & Fine-tuning LLMs
At this critical juncture in AI development, we face a scarcity of novel, high-quality data for training and fine-tuning large language models (LLMs). This shortage poses significant challenges for organizations aiming to enhance model performance or adapt them to specialized domains. In this hour-long livestream, Yev Meyer, Ph.D., Chief Scientist at Gretel, will showcase the company's innovative solution to this bottleneck.
youtube.com
}
Copilot L33t Sp34k | Strategic considerations for AI security
Sarah and Mark Simos, Lead Cybersecurity Architect at Microsoft, chat about the strategic considerations for implementing Copilot for Security and AI more generally across an organisation. We also cover the AI shared responsibility model and how data protection is a key security control in the age of AI.
youtube.com
}
OpenAI's ChatGPT Does Research… And Breaks Itself!
❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambdalabs.com/paper
youtube.comArticles
}
GENEVA uses large language models for interactive game narrative design
Designed for interactive storytelling in games, GENEVA lets users explore narrative paths and adapt stories to diverse contexts. It uses LLMs to generate and visualize branching narratives from high-level descriptions, representing them as graphs.
microsoft.com
}
What Nobody Tells You About RAGs
A deep dive into why RAG doesn’t always work as expected: an overview of the business value, the data, and the technology behind it.
towardsdatascience.com‘Never summon a power you can’t control’: Yuval Noah Harari on how AI could threaten democracy and divide the world
Forget Hollywood depictions of gun-toting robots running wild in the streets – the reality of artificial intelligence is far more dangerous, warns the historian and author in an exclusive extract from his new book.
theguardian.com
}
Anthropic publishes the ‘system prompt’ that makes Claude tick
Generative AI models aren’t actually human-like. They have no intelligence or personality — they’re simply statistical systems predicting the likeliest next words in a sentence. But like interns at a tyrannical workplace, they do follow instructions without complaint — including initial “system prompts” that prime the models with their basic qualities, and what they should […]
techcrunch.comLLM Pruning and Distillation in Practice: The Minitron Approach
LLM Pruning and Distillation in Practice: The Minitron Approach.
research.nvidia.com
}
BERT — Intuitively and Exhaustively Explained
Baking General Understanding into Language Models,
towardsdatascience.com
}
Enabling Fast Gradient Clipping and Ghost Clipping in Opacus
Opacus is a PyTorch implementation of DP-SGD. Opacus addresses the above task by employing hook functions, which allows intervening on specific events, such as forward and backward passes.
pytorch.org
}
How to Achieve Near Human-Level Performance in Chunking for RAGs
The costly yet powerful splitting technique for superior RAG retrieval
towardsdatascience.comCode
}
How to Color Polars DataFrame
Continue working with the Polars library while being able to color and style the table.
towardsdatascience.com