9.28.26 | Google's oddities, Nvidia stock worth a billion dollars, Ember-1

SPEAKER_1 (0:00)

Welcome to the Hacker News Highlights, where we explore the top 10 posts on Hacker News every day. Today, we dive into Google's increasing empathy interfering with searches, a billionaire discovering unpaid Nvidia stock options from 30 years ago, and Ember One reducing token costs without sacrificing quality. Let's dive in.

Title, When Did Google Get So Weird? Source, Sancho Panza's Thoughts. The post recounts a personal experience with Google's AI-powered search in 2026, which provided an empathetic overview instead of direct links when searching for a niche meme from the 2010 S. The author criticizes how Google's AI now often prioritizes emotional engagement and filler responses over accurate, straightforward results, reflecting a shift from traditional search to a more conversational, yet less reliable interface. The story emphasizes that these changes make Google's search feel less like a tool for finding information, and more like a probing, intrusive digital friend. In the comments, the community largely agreed that Google's transition to AI summaries and overcontent has degraded search quality. Users debated whether this shift is intentional or a result of cost-cutting and market pressures, with many noting that alternatives like DuckDuckGo, Kagi, and Yandex often produce better, less hallucinated results. Several discussed how this new focus on empathetic AI responses fosters confusion, misinformation, and even dangerous consequences like military misjudgments in legal issues. The overall sentiment was mostly critical, pointing out that Google's move towards AI overviews damages both trust in the search process and the integrity of information dissemination.

Title, I was owed a billion dollars in Nvidia stock.

Source, Eric Gullickson, September 2026 The post tells the story of the author's early involvement with Nvidia in 1993, including technical work on texture mapping and his role on the advisory board. He discovered decades later that he was allegedly owed about a billion dollars in Nvidia stock due to a legal misunderstanding about vesting schedules and stock options. While he exercised some options in 1996, he believed that a clerical error in paperwork over the vesting period meant he was entitled to more shares, which had grown in value over time. Attempts to resolve the issue legally were hampered by the statute of limitations, and he ultimately accepted the situation with some humor and lessons about reading contracts carefully. In the comments, the community largely supported the author's recounting, with discussions about the legal validity of his claim, the importance of carefully reading stock agreements, and the role of statutes of limitations. Many noted that the claim was unlikely to succeed in court due to the elapsed time, but acknowledged the fascinating nature of the story. Some debated whether the paperwork error could be considered a mutual mistake, or if the claim was just a missed opportunity.

The community expressed that the situation highlighted the importance of diligent contract review and understanding legal timelines, with a generally skeptical but intrigued tone.

Title, Ember One from Fireworks Research offers similar performance to KimiK3 with 40% fewer tokens. Source, Fireworks AI. The article explained that Ember One is a new AI model designed to deliver the same quality as KimiK3, but with fewer tokens by making reasoning more efficient. It was trained through extensive experiments on Fireworks serverless training without using customer data, and tested successfully on public benchmarks, internal evaluations, and live customer traffic, achieving token reductions of up to 50% without losing accuracy. Ember One has been released for use alongside the original KimiK3 in a research preview, targeting agentic workloads like coding, where reasoning tokens are costly.

Fireworks aims to develop specialized models further supported by custom training options. In the comments, the community mostly supported the innovation, but expressed skepticism about the practical benefits relative to cost and licensing limitations. Many discussed how the concept of the Pareto frontier is used as a marketing phrase to highlight efficiency improvements, though some doubted whether the token savings truly translate into real world savings or advantages. Users debated if fireworks release was mainly for marketing or specifically aimed at maintaining market relevance through continuous innovation. Concerns about proprietary licensing, data security, and the true cost effectiveness of the model also appeared, alongside appreciation for the progress, but skepticism about whether the approach truly offers a competitive edge or just an incremental step in AI efficiency.

Title, Show HN, Lofi Cities, Pixel Art City Nights with Browser Generated Lo-Fi, Source, Lofi Cities. The project is a free web app that shows animated pixel art scenes of cities at night, like Paris, Tokyo, New York, paired with live generated Lofi music. Every city scene is a seamless four-minute animation with weather, landmarks, and city sounds, while the music is procedurally created in real time in the browser. It provides a calming background for studying or relaxing, with features like changing weather, city sounds, and adding business billboards. In the comments, the community was mostly supportive, praising the visuals and the concept. They discussed the potential for AI to generate city scenes on demand. With some skepticism about how well AI-generated pixel art could replace hand-made work. Users debated whether the scenes feel authentic or AI-made, with some expressing concern over AI's role in art, while others highlighted the effort behind the curated assets. There was also interest in expanding the app with indoor views, more cities, and better customization options, along with criticism of the ads cluttering the scenes. Overall, the community appreciated the project's creative design, but expressed a desire for more authenticity and smoother integration of user preferences.