The AI Water Usage Issue Is "Fake" — Analysis Article Sparks Debate
An [analysis article by Andy Masley](https://blog.andymasley.com/p/the-ai-water-issue-is-fake) that picked up 56 points and 34 comments on Hacker News argues that "the AI water usage issue is exaggerated," stirring up controversy.
Claims that training and running AI models requires massive amounts of water have been frequently covered in the media over the past few years, but the article, drawing on actual data, calls into question the very premise of this "water issue."
That said, the community remains divided over the validity of the claims, with commenters also noting that there are many variables to consider, such as data center location and differences in cooling methods.
University of Arizona Students Boo Eric Schmidt's AI Praise
According to [The Verge](https://www.theverge.com/ai-artificial-intelligence/932203/university-of-arizona-students-boo-eric-schmidt-ai-commencement), when former Google CEO Eric Schmidt delivered a speech at the University of Arizona's commencement ceremony, students booed his optimistic remarks about AI.
It looks like a symbolic episode showing that concern over AI's impact on jobs and society is growing among the younger generation.
Gemini 3 Flash Takes Top Spot on the LongMemEval Memory Benchmark
According to [research that drew attention](https://www.reddit.com/r/MachineLearning/comments/1tfwd69/1_on_memory_benchmark_longmemeval_with_gemini/) in Reddit's MachineLearning community, an experimental memory retrieval system built with Gemini 3 Flash achieved 96.4% (top-50) on the LongMemEval benchmark, surpassing existing systems that use Gemini 3 Pro (such as Mem0's 94.8%).
The key point is that it delivers high accuracy through a clever retrieval architecture while using the smaller Flash model. Three design choices grounded in cognitive science — query decomposition, temporal salience scoring, and coherence re-ranking — are credited with making the difference.
"What AI Did to My College Classes" — An NYT Column
A [New York Times opinion piece](https://www.nytimes.com/2026/05/17/opinion/chatgpt-ai-college-school-graduation.html) published a candid account of the actual impact AI has had on college education. It also gathered 10 points on Hacker News, where readers are debating the reality of AI use in education.
LinkedIn Recruitment Spam Manipulated via Prompt Injection
As reported by [Tom's Hardware](https://www.tomshardware.com/tech-industry/artificial-intelligence/linkedin-recruitment-spam-becomes-olde-english-prose-after-user-hides-ai-prompt-injection-in-bio-bots-also-also-manipulated-to-address-user-as-my-lord), a LinkedIn user planted a prompt injection in their bio and succeeded in getting an AI recruitment bot to address them as "My Lord" in archaic English.
It's a humorous case, but it highlights the vulnerability of AI agents that trust and act on external text.
Developments in the Local LLM Scene
On Reddit's LocalLLaMA community, discussions continue around hopes for a [124B-parameter Gemma model](https://www.reddit.com/r/LocalLLaMA/comments/1tfv8li/i_hope_that_someday_we_will_have_a_124b_gemma/) and the practicality of [meeting memory systems running on local models](https://www.reddit.com/r/LocalLLaMA/comments/1tfukyl/are_local_models_good_enough_yet_for_ai_meeting/). Some also say that Claude MCP integration, which enables searching meeting history in natural language, is expanding what local AI can do.