LAI #92: AI Hype vs. Reality, Deepfake Detection, and Copilot+ PCs

Author(s): Towards AI Editorial Team

Originally published on Towards AI.

Good morning, AI enthusiasts,

This week, we’re looking at where AI is truly useful versus where it’s just hype. Notion makes for a strong case study in integrating AI, where it adds real value, not noise. On the research side, we cover new methods for detecting deepfakes, why single-vector embeddings are hitting mathematical limits, and what Copilot+ PCs mean for running powerful AI apps entirely offline.

Let’s get into it.

What’s AI Weekly

This week, in What’s AI, we’re going to talk about how to spot the difference between AI that actually solves real problems and AI that’s just there for the sake of being trendy. Every company uses AI, but often, it feels more like hype than something that genuinely helps you get things done. So I will dive into how Notion is a perfect example of useful AI integrations into their new Notion AI, solving problems and not just hype for the sake of using AI. Read the complete article here or watch the video on YouTube.

— Louis-François Bouchard, Towards AI Co-founder & Head of Community

Learn AI Together Community Section!

Featured Community post from the Discord

Mgsneol has been working on Coders Connect, a WhatsApp bot built to help coders, developers, and builders connect more easily. It also opens up opportunities for people to work with the UAE Ministry of AI initiatives. You set up a quick profile, share what you’re looking for, and the bot brings back relevant matches you can reach out to directly. Check it out here and support a fellow community member. If you have any questions, connect with him in the thread!

AI poll of the week!

Almost half of you said vibe coding “depends on the task,” which feels spot on. For exploratory coding, prototyping, or chasing an idea down rabbit holes, it’s a productivity booster. But when it comes to production-ready, test-heavy work? Many of you seem to pull back. What kinds of tasks do you feel vibe coding hurts efficiency rather than helps? Debugging? Large-scale projects? Something else? Tell me in the thread!

Collaboration Opportunities

The Learn AI Together Discord community is flooding with collaboration opportunities. If you are excited to dive into applied AI, want a study partner, or even want to find a partner for your passion project, join the collaboration channel! Keep an eye on this section, too — we share cool opportunities every week!

1. Superuser666_sigil is a member of RLAgentBot, a crypto trading agent. Their team is seeking individuals with spare time to contribute their financial/crypto trading knowledge to our team. If this is your domain, connect with him in the thread!

2. Cpnk75m is working on an AR + AI app for Devpost Hackathon and needs help with picking the right ML models, some quick hacks to get it running, AR objects/modals for overlays, and solid testing. If you are interested in working on this project and potentially scaling it further, reach out in the thread!

Meme of the week!

Meme shared by hitoriarchie

TAI Curated Section

Article of the week

Solving Deepfakes with Traces, Frequency, and Attention! By Shreyash Pawar

Shreyash Pawar presents a hybrid model for deepfake detection that combines three complementary techniques. AMTENnet is used to detect subtle manipulation artifacts, frequency-domain analysis (inspired by VANet) uncovers hidden generation patterns, and a CBAM module helps the system focus on the most relevant features. Tested on GAN-based datasets, this multi-layered approach achieved 98.9% accuracy, showing the effectiveness of blending trace extraction, frequency analysis, and attention mechanisms to spot synthetic images.

Our must-read articles

1. Designing Data Pipeline Architectures for Machine Learning Models By Kuriko Iwai

Kuriko Iwai breaks down three common ways to design data pipelines for machine learning. First, the traditional data warehouse with ETL, great for accuracy but slow. Then, the cloud-native data lake, built for real-time, messy data with ELT and streaming frameworks. Finally, the modern lakehouse that blends the strengths of both. The right choice depends less on trends and more on your data characteristics and business needs.

2. How I built Nano Banana AI Image Editing Agent By Gao Dalie (高達烈)

Gao Dalie walks through building an AI-powered image editing agent using Google’s Gemini 2.5 Flash Image Preview model (known as “nano-banana”). The system, built with Streamlit, lets users provide prompts and reference images, then calls the Gemini API to return both edited images and a description of the changes. It’s a practical example of how fast, instruction-based editing can be turned into an accessible tool.

3. Vector Embeddings Hit Mathematical Limits: Google DeepMind Report By MKWriteshere

A joint report from DeepMind and Johns Hopkins highlights a mathematical ceiling in single-vector embedding models. Even with more training data, they struggle with multi-condition queries due to limited representational capacity. Using the LIMIT dataset, researchers showed that even state-of-the-art models perform poorly. The study suggests that alternatives, cross-encoders, multi-vector models, or hybrid systems are better suited for complex retrieval tasks.

4. On-Device AI Is Finally Real — Build a Copilot+ PC App That Runs 100% Offline By Tarun Singh

Tarun Singh demonstrates how to build an AI app that runs entirely offline on Copilot+ PCs. The project uses ONNX Runtime to leverage NPU/GPU hardware for efficient embeddings, hybrid retrieval with FAISS and BM25, and optional integration with Ollama for local LLM-powered answers. The result is a private, high-performance AI system with clear instructions and code for hands-on implementation.

If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI