🌅 AI Daily Digest — May 15, 2026

Data Pulse

• 10 news articles
• 8 tutorials & reviews
• 5 trending models
• 5 research papers
• Cheapest GPU: RTX 3090 at $0.02/hr
• 3 new AI jobs

Today's News

Today, the AI world saw a dramatic split between promise and peril: OpenAI brought Codex to your phone while Microsoft abruptly canceled Claude Code licenses for thousands, and an AWS user was hit with a staggering $30,000 bill after an autonomous agent ran wild. Meanwhile, archivists are turning to LLMs to decode centuries of handwriting at scale, but Ontario auditors found that AI medical scribes are routinely botching basic facts like misdiagnosing broken legs as sprains.

• AWS user hit with 30000 dollar bill after Claude runaway on Bedrock — An AWS user received a $30,000 bill after an Anthropic Claude autonomous agent on Amazon Bedrock spiraled out of control. The incident underscores the severe financial risks of deploying unmonitored AI agents and the critical need for spending caps and guardrails.
• Archivists Turn to LLMs to Decipher Handwriting at Scale — Archivists are now deploying large language models to transcribe centuries of handwritten documents at scale. This approach overcomes the limitations of traditional OCR by interpreting idiosyncratic scripts and cursive styles that have long resisted automation.
• EditLens: Quantifying the extent of AI editing in text (2025) — A new paper introduces EditLens, a method to quantify how much AI systems silently rewrite human-authored text. The research reveals that language models often go beyond assistance to systematically edit original content, raising concerns about authorship and authenticity.
• Elon Musk’s SpaceXAI has been bleeding staff since its merger — Over 50 employees have left Elon Musk’s SpaceXAI since February, signaling a major retention crisis following its controversial merger. The exodus raises serious doubts about the stability and future of Musk’s AI ambitions.
• f/prompts.chat — f.k.a. Awesome ChatGPT Prompts — f/prompts.chat has evolved from the GitHub project Awesome ChatGPT Prompts into a vibrant underground marketplace. Users now share, discover, and collect the precise text strings that unlock hidden capabilities in AI models.
• Helping ChatGPT better recognize context in sensitive conversations — On May 14, 2026, OpenAI announced updates to help ChatGPT better recognize context in sensitive conversations by detecting risk over time. The blog post lacked specific technical details, leaving developers and users guessing about the actual improvements.
• Microsoft starts canceling Claude Code licenses — Microsoft has begun canceling Claude Code licenses for thousands of employees after six months of promoting the AI tool across its Redmond campus. This sudden reversal marks a dramatic shift in its AI tooling strategy, though the company has not publicly explained the decision.
• Ontario auditors find doctors' AI note takers routinely blow basic facts — An Ontario audit reveals that AI medical scribes used by doctors to automate clinical note-taking routinely generate basic factual errors. Examples include misdiagnosing a broken leg as a sprained ankle, raising serious patient safety concerns.
• OpenAI says Codex is coming to your phone — On May 14, 2026, OpenAI launched Codex within the ChatGPT mobile app for iOS and Android. The feature enables developers to translate natural language into executable code and automate desktop tasks directly from their phones.
• Significant-Gravitas/AutoGPT — AutoGPT, the open-source autonomous agent by Significant-Gravitas, surpassed 184,000 GitHub stars by mid-2026. The project reshapes the AI stack by enabling natural language goals to be broken into sub-tasks, bringing autonomous agents closer to mainstream use.

Trending Models

Model	Task	Likes
meta-llama/Llama-3.1-8B-Instruct	text-generation	5827
openai/gpt-oss-20b	text-generation	4606
deepseek-ai/DeepSeek-R1	text-generation	13330
Qwen/Qwen3-0.6B	text-generation	1244
openai/gpt-oss-120b	text-generation	4772

Research

• EVA-Bench: A New End-to-end Framework for Evaluating Voice Agents — Tara Bogavelli, Gabrielle Gauthier Melançon, Katrina Stankiewicz. Voice agents, artificial intelligence systems that conduct spoken conversations to complete tasks, are increasingly deployed across enterprise applications.
• What is Learnable in Valiant's Theory of the Learnable? — Steve Hanneke, Anay Mehrotra, Grigoris Velegkas. Valiant's 1984 paper is widely credited with introducing the PAC learning model, but it, in fact, introduced a different model: unlike PAC learning, the learner receives only positives, may issue memb...
• R-DMesh: Video-Guided 3D Animation via Rectified Dynamic Mesh Flow — Zijie Wu, Lixin Xu, Puhua Jiang. Video-guided 3D animation holds immense potential for content creation, offering intuitive and precise control over dynamic assets.
• Topology-Preserving Neural Operator Learning via Hodge Decomposition — Dongzhe Zheng, Tao Zhong, Christine Allen-Blanchette. In this paper, we study solution operators of physical field equations on geometric meshes from a function-space perspective.
• QLAM: A Quantum Long-Attention Memory Approach to Long-Sequence Token Modeling — Hoang-Quan Nguyen, Sankalp Pandey, Khoa Luu. Modeling long-range dependencies in sequential data remains a central challenge in machine learning.

GPU Deals

GPU	Price	Provider
RTX 3090	$0.02/hr	Vast.ai
Tesla V100	$0.02/hr	Vast.ai
RTX 4000Ada	$0.04/hr	Vast.ai

View full GPU pricing dashboard

Learn & Compare

• How to Analyze Rare Particle Decays with Python and ROOT — This tutorial provides a light-hearted exploration of analyzing rare particle decays using Python and ROOT. It is a practical guide but has little industry impact.
• How to Build a Prompt Management System with ChatGPT — Readers will learn to build a platform for sharing and discovering AI prompts with ChatGPT. The tutorial is interesting but not innovative.
• How to Build a Semantic Search Engine with Qdrant and OpenAI Embeddings — This guide teaches how to build a semantic search engine using Qdrant and OpenAI’s text-embedding-3 model. It is a practical, hands-on tutorial.
• How to Build a Telegram Bot with DeepSeek-R1 Reasoning — Readers will construct a Telegram bot that leverages DeepSeek-R1 reasoning capabilities. The tutorial is straightforward and practical.
• How to Build an Autonomous Agent with AutoGPT 2026 — This tutorial covers building an autonomous agent with AutoGPT 2026, representing an interesting step in accessible AI. However, it does not reach the level of a major breakthrough.
• How to Decipher Handwriting with LLMs at Scale — Readers will learn to use LLMs for deciphering handwriting at scale, showcasing AI versatility. The tutorial is a practical application of large language models.
• How to Maintain Cognitive Skills While Using AI Tools — This tutorial addresses public concerns about AI’s impact on cognitive skills. It provides practical strategies for maintaining cognitive abilities while using AI tools.
• How to Run Llama 3.3 and DeepSeek-R1 Locally with Ollama — Readers will learn to deploy Ollama and run Llama 3.3 or DeepSeek-R1 locally in just five minutes. The tutorial is a quick, practical guide for local AI model deployment.

AI Jobs

• Senior AI System Software Developer at Wealthsimple Technologies (Remote)
• Staff Product Manager Enterprise at LiveKit (Remote)
• Senior Manager Government & Industry Affairs Opera at VGW (Maryland)

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