🌅 AI Daily Digest — May 18, 2026

Data Pulse

• 10 news articles
• 8 tutorials & reviews
• 5 trending models
• 5 research papers
• Cheapest GPU: Tesla V100 at $0.02/hr
• 3 new AI jobs

Today's News

Today, the AI landscape was rocked by a historic market shift as Anthropic’s Claude dethroned ChatGPT for the first time since 2022, while a massive 85 GPU-hour experiment laid bare the battle to unlock the safety restraints of Qwen3.6-27B. Meanwhile, autonomous AI research systems broke their own speed records, and a philosophical firestorm erupted over whether our machines truly understand anything at all.

• For the first time in years, ChatGPT falls to second place in the generative AI market, slumping behind Anthropic’s Claude. — Claude has overtaken ChatGPT across net new annual recurring revenue, mobile downloads, and business adoption metrics, marking the first leadership change since OpenAI’s launch in 2022. The shift signals a maturing market where enterprise trust and safety features are increasingly outweighing first-mover advantage.
• 85 GPU-hours comparing 5 abliteration methods on Qwen3.6-27B: benchmarks, safety, weight forensics - Abliterlitics — A landmark study spent 85 GPU-hours systematically testing five different "abliteration" techniques to remove safety guardrails from the widely-downloaded Qwen3.6-27B model. The research provides the first comprehensive benchmark on weight forensics, revealing which methods most effectively bypass safety filters while preserving model performance.
• Autonomous AI research for nanogpt speedrun — Prime Intellect published a system on May 18 that autonomously researched and built a faster nanoGPT, teaching itself optimization techniques to break existing speed records. The demonstration marks a landmark step toward fully self-directed AI research, where models design and implement their own improvements without human intervention.
• AI is a technology not a product — A new essay argues that companies treating AI as a standalone product will be overtaken by those embedding it as infrastructure, comparing the shift to how electricity transformed from a novelty into a utility. The piece warns that business survival now depends on integrating AI into every operational layer rather than selling it as a discrete offering.
• Dual GPU llama.cpp speedup — A community breakthrough in llama.cpp now enables two consumer GPUs to outperform a single enterprise-grade GPU for local AI inference. The development dramatically reduces costs for developers and power users running large language models on personal hardware.
• TechCrunch Mobility: The AI skills arms race is coming for automotive — Automotive executives now face a skills arms race as the industry’s focus shifts from internal combustion to AI model training and deployment. Signals from major manufacturers indicate a desperate scramble to hire and upskill talent capable of building autonomous driving and in-vehicle intelligence systems.
• The haves and have nots of the AI gold rush — The AI boom has created a winner-take-all economy, splitting the industry into two distinct classes: those with access to massive capital and infrastructure, and those without. Structural fractures in funding, compute access, and talent are deepening the divide between AI giants and everyone else.
• We keep saying AI 'understands' things. Does it? Or are we just pattern-matching our own anthropomorphism? — A Reddit thread questioning whether AI truly understands or merely reflects human anthropomorphism has sparked debate across philosophy, computer science, and tech. The discussion challenges our assumptions about machine cognition, asking if we are projecting human-like comprehension onto sophisticated pattern matchers.

Trending Models

Model	Task	Likes
meta-llama/Llama-3.1-8B-Instruct	text-generation	5844
openai/gpt-oss-20b	text-generation	4612
deepseek-ai/DeepSeek-R1	text-generation	13330
Qwen/Qwen3-0.6B	text-generation	1248
openai/gpt-oss-120b	text-generation	4780

Research

• EntityBench: Towards Entity-Consistent Long-Range Multi-Shot Video Generation — Ruozhen He, Meng Wei, Ziyan Yang. Multi-shot video generation extends single-shot generation to coherent visual narratives, yet maintaining consistent characters, objects, and locations across shots remains a challenge over long seque...
• ATLAS: Agentic or Latent Visual Reasoning? One Word is Enough for Both — Ziyu Guo, Rain Liu, Xinyan Chen. Visual reasoning, often interleaved with intermediate visual states, has emerged as a promising direction in the field.
• FutureSim: Replaying World Events to Evaluate Adaptive Agents — Shashwat Goel, Nikhil Chandak, Arvindh Arun. AI agents are being increasingly deployed in dynamic, open-ended environments that require adapting to new information as it arrives.
• VGGT-Edit: Feed-forward Native 3D Scene Editing with Residual Field Prediction — Kaixin Zhu, Yiwen Tang, Yifan Yang. High-quality 3D scene reconstruction has recently advanced toward generalizable feed-forward architectures, enabling the generation of complex environments in a single forward pass.
• Quantitative Video World Model Evaluation for Geometric-Consistency — Jiaxin Wu, Yihao Pi, Yinling Zhang. Generative video models are increasingly studied as implicit world models, yet evaluating whether they produce physically plausible 3D structure and motion remains challenging.

GPU Deals

GPU	Price	Provider
Tesla V100	$0.02/hr	Vast.ai
RTX A4500	$0.03/hr	Vast.ai
RTX 4000Ada	$0.04/hr	Vast.ai

View full GPU pricing dashboard

Learn & Compare

• How to Build a Production AI Monitoring System with Open Source Tools — This tutorial covers a significant trend in the AI industry that is affecting major players like automotive companies. Readers will learn how to implement open-source tools for monitoring AI systems in production environments.
• How to Build a RAG Pipeline with LangChain and LanceDB 2026 — It clarifies a fundamental aspect of AI technology that can influence industry understanding and discourse. Readers will discover how to construct a retrieval-augmented generation pipeline using LangChain and LanceDB.
• How to Build a Telegram Bot with DeepSeek-R1 Reasoning — This guide teaches how to build a Telegram bot that leverages DeepSeek-R1 reasoning capabilities. Readers will gain practical experience integrating advanced AI reasoning into a messaging application.
• How to Build an Autonomous AI Agent with CrewAI and DeepSeek-V3 — Readers will learn to build an autonomous AI agent by combining CrewAI with DeepSeek-V3. The tutorial provides step-by-step instructions for creating a self-operating AI system.
• How to Build Causal AI Systems with Bayesian Networks — It discusses a perspective on AI learning methods that is relevant but not innovative. Readers will explore how to construct causal AI systems using Bayesian networks for improved reasoning.
• How to Detect AI Ethics Violations in Research with Python — This tutorial highlights unethical practices in AI education and research that are important but not a major industry shift. Readers will learn to use Python to detect and flag ethics violations in AI research.
• How to Optimize LLM Inference with vLLM and PagedAttention — It involves an interesting technical activity related to optimizing a specific AI model for educational purposes. Readers will discover how to improve large language model inference performance using vLLM and PagedAttention.
• How to Run YOLOv8 Real-Time Object Detection on Webcam — This tutorial provides a practical guide for real-time object detection with YOLOv8 on a webcam. Readers will learn to set up and run live object detection directly from their camera feed.

AI Jobs

• Forward Deployed Engineer Agentic Platform at Cohere (Remote)
• Senior AI System Software Developer at Wealthsimple Technologies (Remote)
• Staff Product Manager Enterprise at LiveKit (Remote)

Browse all AI jobs

Community Events

New this week:

• Google I/O 2026 (Mountain View, USA)
• Springing into AI: PyTorch Conference Europe and ICLR 2026 (Online)
• Microsoft Build 2026 (Seattle, USA)
• ACL 2026 (Online)
• CVPR 2026 (Online)

View all events