Episode Details

Back to Episodes
AI proves an Erdős conjecture & Data filtering in AI pretraining - AI News (May 22, 2026)

AI proves an Erdős conjecture & Data filtering in AI pretraining - AI News (May 22, 2026)

Published 1 month ago
Description
Please support this podcast by checking out our sponsors:
- SurveyMonkey, Using AI to surface insights faster and reduce manual analysis time - https://get.surveymonkey.com/tad
- Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron
- KrispCall: Agentic Cloud Telephony - https://try.krispcall.com/tad


Support The Automated Daily directly:
Buy me a coffee: https://buymeacoffee.com/theautomateddaily

Today's topics:

AI proves an Erdős conjecture - OpenAI says an internal reasoning model generated a verifiable proof overturning Erdős’s planar unit-distance conjecture, validated by external mathematicians—major news for both discrete geometry and AI reasoning.

Data filtering in AI pretraining - A new arXiv study from Mohri, Duchi, and Hashimoto suggests that with enough compute, the best data-quality filter may be no filter, challenging common dataset curation assumptions in LLM pretraining.

Long-video understanding gets cheaper - DeepMind and Seoul National University introduced LiteFrame, a compact video encoder that cuts latency and enables much longer context windows, improving long-form video understanding on key benchmarks.

Open multimodal models push downscale - ByteDance released Lance, an open-source 3B unified multimodal model spanning image and video understanding and generation, signaling a continued push toward smaller, single-model multimodal tooling.

Audio generation shifts to open - Stability AI launched Stable Audio 3.0 with open weights and licensed training data, while Meta Research shared WavFlow for video-to-audio generation, highlighting rapid progress in generative audio.

Compute deals and AI infrastructure - A securities filing says Anthropic plans massive payments to SpaceX for compute, underscoring how GPU-scale infrastructure access is becoming the central competitive moat in AI.

AI pricing pressure hits margins - Enterprises are warning that LLM inference costs are hurting margins, while API pricing shifts and cheaper competitors pressure premium models—raising questions about long-term profitability.

OpenAI IPO timeline heats up - The Wall Street Journal reports OpenAI is moving toward an IPO as early as September, after a Musk lawsuit loss removed a major overhang—setting up a marquee market event.

China accelerates its chip stack - Alibaba unveiled the Zhenwu M890 accelerator to reduce reliance on Nvidia amid export restrictions, reflecting China’s broader push to build a domestic AI compute stack.

AI plagiarism, burnout, and backlash - A creator’s plagiarism complaint, an essay on engineer AI burnout, and cultural pushback around AI in media and commencements show the human and incentive-side costs of AI adoption.

Websites optimized for AI agents - Chrome Lighthouse added experimental checks like llms.txt under “Agentic Browsing,” hinting at a new category of web best practices aimed at AI agents rather than traditional search ranking.



-Blogger Says Generative AI Enables Large-Scale Plagiarism and Rewards Copycats
-
Listen Now