Latest / UpNext AI / DuckDuckGo’s No-AI Search Push, Alphabet’s $80 Billion AI Buildout, and Bias in Multimodal AI Judges | UpNext AI – June 2, 2026
Transcript
- Welcome to the UpNext AI podcast. It's Tuesday, June 2nd, 2026, and here's what matters in AI today. Our lead story is DuckDuckGo’s push to make AI-free search easier to reach. TechCrunch reports that DuckDuckGo has launched new browser extensions for Chrome and Firefox that let users set its no-AI search experience as their default. Once enabled, users are sent to DuckDuckGo’s AI-free search page, where the company says there are no AI-assisted answers, no chat prompts, and fewer AI images in results. What makes this notable is the positioning: DuckDuckGo is turning “no AI” into a consumer-facing product choice just as the biggest search platforms are moving hard in the other direction. TechCrunch says traffic to DuckDuckGo’s no-AI search page has been rising, which suggests at least some users want search to stay closer to the old link-first model. And DuckDuckGo is not abandoning AI entirely; it still offers its own AI chatbot and subscription tools. So this is really a product packaging bet: make opting out of AI in search easy, visible, and defaultable. For our second act, Alphabet is signaling just how capital-intensive this race has become. According to TechCrunch, Alphabet says it plans to raise eighty billion dollars to help fund AI infrastructure and global compute. The company said demand for its AI solutions and services from enterprises and consumers is exceeding available supply. That line matters because it shifts the conversation from model launches to the physical bottlenecks underneath them: chips, power, networking, and data centers. TechCrunch also reports that part of the plan involves selling ten billion dollars in stock to Berkshire Hathaway. The bigger takeaway is that AI is increasingly a financing story as much as a software story. At this scale, capacity itself is becoming a competitive advantage. Now to the research story. A paper posted to arXiv on June 1st looks at a weakness in multimodal LLM-as-a-judge systems, meaning models used to evaluate other model outputs when both images and text are involved. The researchers argue these judge models can show perceptual judgment bias: when visual evidence and textual cues conflict, the judge may reward the answer that sounds more plausible in words instead of the one that is actually grounded in the image. To address that, the paper studies controlled visual perturbations and proposes perceptual perturbation plus reward modeling as a mitigation. The authors say that approach improves perceptual fidelity, ranking coherence, and alignment with human evaluation. The practical point is simple: if you use AI as a judge for mixed-media outputs, you need to test whether it is actually following the image, not just the caption. Bottom line: multimodal AI judges can be biased by the wrong cue, and this paper offers one concrete path to making those evaluators more reliable. ...Are you building apps with voice? Elevate your app's voice capabilities with ElevenLabs. Their API is a game changer for embedding dynamic, responsive voice interactions in your applications, providing unprecedented realism, flexibility and latency. In fact, you're listening to one of their voices - right - now. If you are a developer looking to elevate user experience with natural voice interfaces, this is your solution. Visit up next dot fm slash eleven to check out their latest offerings. ... In headlines, GlobalData says autonomous AI agents are exposing the limits of traditional graphical user interfaces across software development, cloud operations, automation, and digital infrastructure. The argument, picked up by Fudzilla, is that agents need cleaner, machine-readable execution paths than point-and-click software was designed to provide. Google also published a behind-the-scenes look at how it used Gemini and other AI tools to help build Google I/O 2026, including work on film, visual design, music experiments, interactive experiences, and speaker assets. Nvidia, meanwhile, used GTC Taipei to launch a set of physical-AI offerings for robots, autonomous vehicles, and video systems, according to The Decoder, with a new world model, a driving brain, and an open humanoid robot effort at the center of that push. And Simon Willison highlighted reporting that attackers were able to use Meta’s AI support flow in an Instagram account takeover scenario. The reporting is still filtering through, but it is an early reminder that AI-powered support systems can create very real security problems if sensitive actions are too easy to trigger. Before we wrap up, a quick note: this podcast is generated with the assistance of AI and is intended for informational purposes only. All referenced articles, research, and commentary remain the property of their original authors and publishers. If you enjoyed this episode, don't forget to subscribe, rate, and leave us a review! And that's your briefing for today. Full source links are in the episode notes, and we'll be back tomorrow with what's up next!