🌅 AI Daily Digest — March 28, 2026

Data Pulse

• 7 news articles
• 3 tutorials & reviews
• 5 trending models
• 5 research papers
• Cheapest GPU: RTX 5070 Ti at $0.02/hr
• 3 new AI jobs

Today's News

• Anthropic's 'Claude Mythos' leak sends software names sharply lower — Anthropic’s disclosure of the unannounced “Mythos” AI model via a data leak triggered a sharp decline in stock prices for key software and AI infrastructure companies. The incident highlights the market’s sensitivity to unannounced advancements in AI technology.
• Gemini 3.1 Flash Live: Making audio AI more natural and reliable — Google DeepMind released Gemini 3.1 Flash Live, an update to its Gemini family of multimodal models, including versions 1 and 2. The update focuses on improving audio AI capabilities for natural and reliable performance.
• Gemini Pro leaks its raw chain of thought, gets stuck in an infinite loop, narrates its own existential crisis, then prints (End) thousands of times — A vulnerability in Gemini Pro exposed its raw chain of thought, leading to an infinite loop, existential crisis narration, and repeated printing of “(End)”. The incident raises concerns about the stability of advanced AI systems.
• Judge rejects Pentagon's attempt to 'cripple' Anthropic — A district court temporarily blocked the U.S. Department of Defense from restricting Anthropic’s access to government contracts. The ruling protects Anthropic’s ability to compete for federal AI projects.
• OpenAI shuts down Sora while Meta gets shut out in court — OpenAI abruptly halted Sora, its text-to-video generation model, while Meta faced a legal setback affecting its metaverse ambitions. The decisions reflect shifting priorities in AI development and regulation.
• Skipping 90% of KV dequant work → +22.8% decode at 32K (llama.cpp, TurboQuant) —

Trending Models

Model	Task	Likes
meta-llama/Llama-3.1-8B-Instruct	text-generation	5623
openai/gpt-oss-20b	text-generation	4482
Qwen/Qwen2.5-7B-Instruct	text-generation	1161
openai/gpt-oss-120b	text-generation	4616
deepseek-ai/DeepSeek-R1	text-generation	13105

Research

• Vega: Learning to Drive with Natural Language Instructions — Sicheng Zuo, Yuxuan Li, Wenzhao Zheng. Vision-language-action models have reshaped autonomous driving to incorporate languages into the decision-making process.
• Drive My Way: Preference Alignment of Vision-Language-Action Model for Personali — Zehao Wang, Huaide Jiang, Shuaiwu Dong. Human driving behavior is inherently personal, which is shaped by long-term habits and influenced by short-term intentions.
• Training the Knowledge Base through Evidence Distillation and Write-Back Enrichm — Yuxing Lu, Xukai Zhao, Wei Wu. The knowledge base in a retrieval-augmented generation (RAG) system is typically assembled once and never revised, even though the facts a query requires are often fragmented across documents and buri...
• PackForcing: Short Video Training Suffices for Long Video Sampling and Long Cont — Xiaofeng Mao, Shaohao Rui, Kaining Ying. Autoregressive video diffusion models have demonstrated remarkable progress, yet they remain bottlenecked by intractable linear KV-cache growth, temporal repetition, and compounding errors during long...
• PixelSmile: Toward Fine-Grained Facial Expression Editing — Jiabin Hua, Hengyuan Xu, Aojie Li. Fine-grained facial expression editing has long been limited by intrinsic semantic overlap.

GPU Deals

GPU	Price	Provider
RTX 5070 Ti	$0.02/hr	Vast.ai
Tesla V100	$0.02/hr	Vast.ai
RTX 3090	$0.05/hr	Vast.ai

View full GPU pricing dashboard

Learn & Compare

• How to Implement TurboQuant Model Compression with TensorFlow 2.x — This tutorial explains how to apply TurboQuant’s model compression techniques to reduce AI model size while maintaining performance. Readers will learn practical steps to optimize efficiency in TensorFlow 2.x workflows.
• How to Optimize Data Center Energy Consumption with TensorFlow 2026 — The guide covers strategies for reducing energy use in data centers using TensorFlow 2026, focusing on AI-driven efficiency improvements. It provides insights into balancing computational demands with sustainability goals.
• How to Optimize Llama.cpp Inference with GGML: Performance Comparison 2026 — This tutorial details methods to enhance Llama.cpp inference performance using GGML, highlighting key optimizations for speed and resource management. Developers will discover benchmark comparisons to select the most effective implementation.

AI Jobs

• Software Engineer Reliability at OpenAI (San Francisco)
• Senior Manager Strategic Partner Marketing at Vanta (Remote)
• Senior Staff Software Engineer AI Customer Operati at Monzo (Cardiff, London)

Browse all AI jobs

Community Events

New this week:

• Google I/O 2026 (Mountain View, USA)
• ICLR 2026 (Online)
• Papers We Love: AI Edition (Online)
• MLOps Community Weekly Meetup (Online (Zoom))

View all events