AI News Archive: June 12, 2026 — Part 2
Sourced from 500+ daily AI sources, scored by relevance.
- Microsoft Discovery Platform Brings Agentic AI to Scientific Research
Microsoft has moved its Discovery platform into general availability, calling the service a production-ready environment for scientists and researchers that want to apply AI agents.
- Ainnocence Brings Its Protein, Antibody and Peptide AI Models to Production as Secure, On-Demand APIs
Ainnocence Brings Its Protein, Antibody and Peptide AI Models to Production as Secure, On-Demand APIs azcentral.com and The Arizona Republic
- Bluechip Technologies announces the acquisition of YarnGPT at the Bluechip Data and AI Summit
Bluechip Technologies announced the acquisition of YarnGPT, an AI text-to-speech model, at the Bluechip Data and AI Summit on Wednesday, as Africa's tech leaders gathered to define the continent's role in the AI future.
Score: 75🌐 MovesJun 12, 2026https://techpoint.africa/coverage/bluechip-announces-acquisition-of-yarngpt/ - Ukraine’s AI drones have given Kyiv a fresh edge on the battlefield
Ukraine’s AI drones have given Kyiv a fresh edge on the battlefield The Japan Times
Score: 75🌐 MovesJun 12, 2026https://www.japantimes.co.jp/news/2026/06/12/world/ukraine-ai-drones-battlefield-edge/ - Apple AI now runs on Google, Nvidia tech: What happens to privacy promise
Apple's most advanced AI features are no longer powered solely by Apple infrastructure. As Google and Nvidia enter the stack, questions around privacy may surface
- China’s Nvidia challenger MetaX eyes Hong Kong share sale to fund AI ambitions
MetaX, a Shanghai-based graphics processing unit (GPU) designer, said on Friday that it plans to pursue a Hong Kong listing, less than half a year after its debut on Shanghai’s technology-focused Star Market. The company said in a filing that the proposed H-share offering would support its business expansion, strengthen corporate governance and competitiveness, and advance its international growth strategy. MetaX plans to issue H shares equivalent to no more than 5 per cent of its enlarged share...
- Over half of Americans fear losing both their jobs and their independent thinking to AI, survey finds
Anthropic surveyed nearly 52,000 Americans about their hopes and fears around AI. Sixty-four percent fear job losses, and 56 percent worry about losing the ability to think for themselves. Daily AI users are far less concerned. Still, most people reject AI in their own workplace, even for tasks they think it can handle. The article Over half of Americans fear losing both their jobs and their independent thinking to AI, survey finds appeared first on The Decoder .
- Neura CEO on raising Europe’s biggest ever robotics round: ‘It’s not science fiction’
Neura CEO on raising Europe’s biggest ever robotics round: ‘It’s not science fiction’
Score: 73🌐 MovesJun 12, 2026https://sifted.eu/articles/neura-robotics-ceo-interview-funding-round-tether/ - Pine Labs Launches ‘India’s First Autonomous Agentic Payment’, P3P
Pine Labs Launches ‘India’s First Autonomous Agentic Payment’, P3P india.entrepreneur.com
Score: 73🌐 MovesJun 12, 2026https://india.entrepreneur.com/business-news/pine-labs-launches-indias-first-autonomous-agentic-payment-p3p - AI has become latest US inflation problem with memory chips at ‘insane’ prices
AI has become latest US inflation problem with memory chips at ‘insane’ prices The Straits Times
- Former xAI staffer says he was fired for questioning Grok safety
Former xAI staffer says he was fired for questioning Grok safety The Mercury News
Score: 72🌐 MovesJun 12, 2026https://www.mercurynews.com/2026/06/12/former-xai-staffer-says-he-was-fired-for-questioning-grok-safety/amp/ - Nvidia preps to sell its Vera CPUs into China as its GPU sales stay frozen — customers encouraged to place orders for CPU shipments as early as August
Nvidia has told Chinese clients that its Arm-based Vera server CPUs could be available as soon as August.
- Singapore, Microsoft to explore ways to test the safety of frontier AI models
Singapore, Microsoft to explore ways to test the safety of frontier AI models The Straits Times
Score: 72🌐 MovesJun 12, 2026https://www.straitstimes.com/tech/singapore-microsoft-to-explore-ways-to-test-the-safety-of-frontier-ai-models - Huawei arms HarmonyOS with 2,000 AI agents in challenge to Apple
Huawei Technologies unveiled HarmonyOS 7 on Friday, a major upgrade to its self-developed mobile operating system that the company said marks its entry into the “agent era”, as the Chinese technology giant ramps up competition with Apple’s latest iOS 27 software. The new platform introduces an “agent-friendly” architecture and a significantly enhanced voice assistant capable of understanding context, remembering past interactions and connecting with more than 2,000 specialised AI agents,...
- AI Is Turbocharging the Spamosphere, Amping Up Prolific Text-Message Scams
Google sued swindlers accused in losses totaling $1.9 billion.
- India’s Avataar AI launches a video model that costs $0.005 per second, 27x cheaper than rivals
Bangalore-based Avataar AI has launched Varya, one of India’s first homegrown video AI models. It generates video at roughly $0.005 per second, or 0.48 rupees. Founder Sravanth Aluru, a former Deutsche Bank investment banker and Microsoft and IIT Mumbai alum, says that is 27 times cheaper than comparable open-source video models. The cost advantage comes […] This story continues at The Next Web
- Tencent's AI Bet Is Not a Chatbot. It Is WeChat as an Action Layer
Tencent is not trying to win the chatbot beauty contest. It is wiring AI into the places where users already message, pay, order, and work.
- Police officer suspended over alleged misuse of AI
Police officer suspended over alleged misuse of AI The Telegraph
Score: 71🌐 MovesJun 12, 2026https://www.telegraph.co.uk/news/2026/06/12/police-officer-suspended-misuse-ai/ - Anthropic’s Claude Fable 5 curbs to create new hurdles for China’s AI labs
Anthropic’s move to restrict access to its cutting-edge model, Claude Fable 5, is expected to create new hurdles for Chinese artificial intelligence labs, experts say, even as the US firm walks back part of its enforcement plan following a backlash from the global AI research community. Earlier this week, Anthropic released Claude Fable 5, the public-facing version of Mythos – the company’s most powerful model to date. First announced in April but withheld from general use, Mythos alarmed...
- Google AI Mode starts rolling out Search agents that keep track of information for you
At I/O 2026, Google announced the concept of “Search agents,” with information agents now rolling out in AI Mode for AI Ultra subscribers.
- Anthropic Disputes Fable 5 AI Jailbreak
An AI hacker claims to have achieved a prompt-based jailbreak shortly after Fable 5’s launch, but Anthropic says it’s not a real jailbreak. The post Anthropic Disputes Fable 5 AI Jailbreak appeared first on SecurityWeek .
- VA’s AI chatbots not designated high-impact, despite clinical use, watchdog says
VA’s Inspector General noted that the agency’s two internal chatbots “are not designed specifically for clinical use,” although they have been deployed for such purposes.
- KAIST Launches Mind Care & Growth Center, an Integrated Mental Health Platform for the AI Era
KAIST Launches Mind Care & Growth Center, an Integrated Mental Health Platform for the AI Era EurekAlert!
- The First Fully AI Movie to Screen at a Festival Has Nearly Glitch-Free Tribeca Debut
It took only $2,000, less than two months, and two Iranian brothers to make a resistance film that may ruffle feathers. “The disruption AI is going to cause is obviously something that is not easy to digest,” says producer Tom Rogers.
- This AI Stock Is Australia’s Answer to CoreWeave. It Just Notched a Deal With Nvidia.
This AI Stock Is Australia’s Answer to CoreWeave. It Just Notched a Deal With Nvidia. Barron's
- Apple finally delivers on original Siri promise
Apple finally delivers on original Siri promise USA Today
- Counsel's Concerns of AI-Related Legal Exposure Spike in Healthcare Industry, Survey Finds
Norton Rose Fulbright's survey found 53% of healthcare industry counsel said AI exposure across clinical and operational functions has increased at the federal level.
- Alleged victim in AI deepfake case describes what trauma-informed means to her
Alleged victim in AI deepfake case describes what trauma-informed means to her CBC
- Samsung reportedly in talks to help make Google's next AI chip
Samsung Electronics is reportedly in talks to make a key portion of one of Google's most advanced future artificial intelligence chips. The deal would give the South Korean company a role in a supply chain as the industry is being forced to widen, with Taiwan Semiconductor Manufacturing Co. running out of capacity to meet demand. Google is considering using Samsung's 2-nanometer process to produce a memory input-output die for its 10th-generation tensor processing unit, the in-house chip Google
- Don't know if Claude AI used in strike at Iran school: Anthropic CEO
Dario Amodei said the use case in this instance didn't violate the company's policies, arguing military decision makers make terrible mistakes even at the best of times
- India wont be second to anybody in developing foundational AI models: IT Secretary
India won't be second to anybody in developing foundational AI models: IT Secretary
- AI chatbot's BMW buy-back offer sparks questions over accountability in automated customer interactions
A customer's attempt to sell his BMW back to a dealership turned into a lesson in AI accountability when an artificial intelligence chatbot mistakenly negotiated a higher buy-back price than intended. After initially revoking the offer, the dealership ultimately honoured it following media scrutiny. The incident highlights growing questions around the legal and operational responsibilities businesses face as AI systems increasingly interact with customers and communicate on behalf of companies.
- Theker just raised $85M to build the factory robot that doesn’t specialize in anything
Unlike humanoid robots designed around a fixed form — think Boston Dynamics — Theker's machines are built to be reconfigured.
- Rekise Marine raises $9.7 million from Accel, NKSquared to build autonomous naval platforms
Bengaluru-based Rekise Marine secured $9.7 million from Accel and NKSquared. The funding will accelerate the development of autonomous ships and submarines for the Indian Navy. The company focuses on full-stack development of unmanned maritime systems. This investment supports the expansion of engineering talent and the advancement of their extra-large autonomous underwater vehicle, Jalkapi.
- KPMG report contained AI hallucinations on benefits of . . . AI
Bogus case studies on UBS and transit systems exaggerated adoption of the technology
- Visa Wants to Let You Give ChatGPT Your Credit Card. What Could Go Wrong?
AI companies are increasingly excited about giving AI agents control of the shopping cart.
Score: 70🌐 MovesJun 12, 2026https://www.cnet.com/tech/services-and-software/visa-openai-credit-card-shopping/ - Labour MP Jess Asato launches legal action over Grok deepfakes
Labour MP Jess Asato is taking legal action against Elon Musk’s xAI after its Grok chatbot was used to non-consensually fabricate sexualised images of her, marking the first ever English law claim against the generation of deepfakes. In a claim submitted to the High Court in London at the start of June 2026, Asato alleged that xAI – now a subsidiary of Musk’s SpaceX, which also owns X ( formerly Twitter ) – breached UK data protection law and rules around the misuse of private information in allowing users of the site to create non-consensual deepfake images and videos of her. Asato’s claim will seek damages from xAI and attempt to set a precedent that technology companies be held responsible for their design choices and the harms of the systems they create. “Grok created deepfake pornography and sexualised content which harmed thousands of women and children,” she said. “Its ability is not an accident, nor misuse, it is a design choice by its creators. In launching this case, I am pursuing accountability for those choices. “I hope this legal action also gives voice to the thousands of victims in the UK, women, girls and horrifically even children who were abused by Grok. I am calling on anyone in the UK who experienced the misuse of their image or video by Grok to come forward and support our legal claim.” Ravi Naik, the legal director of legal firm AWO who is representing Asato, added that, “at its heart, this case is about a single principle – that AI developers must answer for the way they design their tools…No one should be subjected to abuse like this, and no one should have to instruct a lawyer to get images like these taken down.” He added while the firm has already secured the removal of the offending images, it is now seeking redress and accountability for Asato. “This content existed because of design choices made by xAI, and technology of this kind does not simply happen – it is built and it is built deliberately,” said Naik. “Grok was designed in a way that permitted the creation of non-consensual, sexualised and misogynistic images of women – and that outcome was a choice, not a glitch. This is one of the first claims to test liability for the design of an AI system, and we hope it will make it clear to AI developers that safety cannot be an afterthought.” Asato’s case has since been backed by more than 100 campaigners and organisations – including Women’s Aid, Refuge, Rape Crisis England & Wales, the Fawcett Society, the Mental Health Foundation and the Molly Rose Foundation – which have published a joint statement backing the MP. “We hope that this will be a first step towards accountability for those responsible and that it will open a path to redress for the many, many other victims who have suffered,” they said. “Researchers found that in an 11-day period – from the start of December 29th 2025 to the end of January 8th 2026 – Grok generated an estimated 3 million non-consensual sexualised images of women and children, which were widely disseminated on X, causing untold harm. Technology of this kind does not simply happen – it is built and it is built deliberately Ravi Naik, AWO “To date, there has been no justice for any of the victims. We believe that xAI must be held legally accountable to ensure that no AI tool or social media platform can ever repeat such awful harms against women, children or anyone.” The UK government previously condemned X in January 2026 after Grok was used to produce vast quantities of sexualised images based on real women, and in some cases children, with senior politicians claiming the company “is not doing enough to keep its customers safe online”. The media regulator, Ofcom, then launched a formal inquiry the same month , saying it would work to establish whether X has failed to comply with its legal obligations under the Online Safety Act . Among the areas of investigation is an assessment of the risk of people in the UK seeing content that is illegal in the UK, and whether X is taking appropriate steps to prevent people in the UK from seeing “priority” illegal content such as non-consensual intimate images. The investigation will also look at how quickly X takes down illegal content when it is made aware of it, and how it is protecting users from a breach of privacy laws. With regards to protecting children, Ofcom said the investigation will also assess the risk the Grok AI service poses to UK children, and the effectiveness of X to use age assurance to protect UK children from seeing pornography. In a post on X from 9 January 2026, Musk said the UK wants “any excuse for censorship”. Computer Weekly contacted xAI about Asato’s legal action, as well as Ofcom’s investigation, but received no response. Emma Pickering , head of technology-facilitated abuse and economic empowerment at Refuge, the charity which provides specialist support for women and children experiencing domestic violence, previously called for tech companies to be held accountable for implementing effective safeguards and preventing perpetrators from causing harm. “Legislation to criminalise creating, or requesting the creation of, non-consensual deepfake intimate images has progressed through Parliament, but we are still waiting for the law to come into effect,” she said. While the sharing of real and synthetic intimate images without consent is illegal in the UK, she pointed out that in practice, the law is not being effectively enforced, with woefully low conviction rates. Read more about online safety Age verification tech could put children at greater risk, says think tank : UK proposals for mandatory age verification will not mitigate children’s exposure to harmful content and ‘addictive’ app design, and risks excluding vulnerable groups from online services, says Foundation for Information Policy Research. Businesses may be caught by government proposals to restrict VPN use : Labour proposals to restrict social media use to people aged 16 and under could have unintended consequences for businesses using virtual private networks. Government and Ofcom disagree about scope of Online Safety Act : MPs heard different views from the online harms regulator and the UK government about whether and how the Online Safety Act obliges platforms to deal with disinformation.
Score: 69🌐 MovesJun 12, 2026https://www.computerweekly.com/news/366644374/Labour-MP-Jess-Asato-launches-legal-action-over-Grok-deepfakes - Huawei is considering deploying Ascend AI chips in Latin America, cloud chief says
Huawei Technologies is studying whether to run its newest Ascend artificial intelligence chips inside its cloud and AI services in Latin America, a senior executive has confirmed, in a move that would push Chinese-designed hardware deeper into a region long courted by US suppliers. Mark Chen, president of Huawei Cloud Latin America, gave the confirmation in an exclusive interview with the South China Morning Post after his presentation at the Rio Web Summit, the technology and innovation...
- LLM collapse: The danger of training LLMs on AI-generated data
Human-generated content has edge over synthetic content
- PixelRAG beats text parsers on accuracy and cuts AI agent token costs 10x
Most enterprise RAG pipelines start the same way: a text parser converts web pages and documents into plain text so they can be chunked and indexed for retrieval. That conversion step destroys retrieval signals — and according to new research, it's responsible for the majority of wrong answers. A research team from UC Berkeley, Princeton University, EPFL and Databricks published a paper this week introducing PixelRAG, a system that skips that conversion entirely. Instead of parsing pages into text, PixelRAG renders them as screenshots, indexes those images and feeds retrieved tiles directly to a vision-language model reader. Tested across 30 million screenshot tiles covering all of Wikipedia, it outperforms text-based RAG across six benchmarks, improving accuracy by up to 18.1% over text-based baselines. Parsers are the wrong place to look for fixes, according to the research team. "Improving parsers is an endless process because every website requires special handling," Yichuan Wang, lead author and UC Berkeley doctorate student, told VentureBeat. "Our goal was to explore whether recent advances in VLMs make it possible to bypass that entire problem and build a retrieval system that works across websites without site-specific engineering." HTML parsers destroy the retrieval signals that enterprise RAG depends on The goal of the researchers was to develop a clean end-to-end architecture. "Modern web RAG pipelines often involve rendering, parsing, cleaning, chunking, and many other handcrafted stages," Wang said. "Every stage introduces potential cascade errors and abstractions that move us further away from the original webpage. We were interested in whether we could eliminate most of that complexity and operate directly on the rendered page." Wang also noted that parsing inevitably loses information. Images, visual hierarchy, typography, emphasis (e.g., bold text), tables, and layout are either discarded or converted into imperfect textual approximations. "No matter how good a parser becomes, some information is fundamentally lost during the conversion," he said. The research identifies three ways text-based RAG loses the answer before it reaches the reader. All three were measured on SimpleQA, a standard benchmark of 1,000 factual Wikipedia questions: Parser loss (36.6% of failures). HTML-to-text conversion destroys structured content so completely that no text chunk in the corpus contains the answer. Rank loss (55.2% of failures). The answer exists in the corpus but gets outranked by keyword-dense infoboxes that land at rank 1 for 75.9% of queries, pushing answer-bearing paragraphs to rank 20 or lower. Reader loss (8.2% of failures). The correct content reaches the reader but flattened structure causes misattribution. How PixelRAG works Unlike a standard LLM that reads only text, a vision-language model takes images as input alongside text, meaning it can read a rendered web page the way a human does, with layout and structure intact. "For many structured information extraction tasks, we believe modern VLMs have an inherent advantage because they can reason jointly over both content and layout rather than relying on a flattened text representation," Wang said. PixelRAG is built around that principle, replacing the text parsing pipeline with a four-stage system that operates entirely on rendered screenshots. Rendering. Pages are rendered using Playwright, a browser automation library, at a fixed 875-pixel viewport and sliced into 1024-pixel-tall tiles. Wikipedia's 7 million articles produce roughly 30 million tiles. Assets are cached locally and rendered entirely offline. Indexing. Each tile is encoded as a single 2048-dimensional vector using Qwen3-VL-Embedding-2B and stored in a FAISS approximate nearest-neighbor index. The full index runs to approximately 120 GB in fp16 and supports incremental updates without full re-indexing. Training. The retrieval model is fine-tuned on synthetic contrastive data generated from the datastore, using dynamic hard-negative mining to filter false negatives. LoRA, a lightweight fine-tuning method that updates a small fraction of model weights, is applied to both the language model backbone and the visual encoder. Training on approximately 40,000 pairs completes in under three hours on a single H100. Storage. Raw screenshot tiles for Wikipedia require 5.6 TB, but a render-on-demand approach eliminates persistent storage: embed all tiles, delete the screenshots and re-render pages on demand at query time. The vector index requires approximately 120 GB. Six benchmarks, 10x agent token savings and one unsolved problem Researchers tested PixelRAG across six benchmarks spanning factual Wikipedia QA, table-based queries, multimodal QA and live news retrieval. They said it outperformed text-based RAG on all six, including on tasks where questions are answerable from text alone. On SimpleQA it reaches 78.8% accuracy versus 71.6% for the strongest text parser, widening to 48.8% versus 42.5% on structured table queries. Teams need Qwen3-VL-4B class models or above to see the benefit. Smaller models trail text retrieval by more than 12.5 percentage points. The agent cost advantage is the strongest near-term case for PixelRAG. In benchmark testing, an AI agent using PixelRAG as its search backend ran on 3.6 million prompt tokens versus 37.5 million for text retrieval, at 2 to 4 times lower cost than alternatives including Google, while achieving higher accuracy. Image compression can cut that token budget by a further third. Visual chunking is the main unsolved problem . Text-based RAG systems have spent years refining how to split documents into meaningful retrieval units based on topic, section or semantic content. PixelRAG currently has no equivalent: it slices pages by fixed pixel height, meaning a table or paragraph can get cut in half mid-tile with no awareness of content boundaries. "The text retrieval community has spent years studying chunking strategies, while visual retrieval has received much less attention," Wang said. "We think this is an important area for future research." What this means for enterprises The retrieval quality problem PixelRAG addresses reflects a broader market shift already underway. VB Pulse Q1 2026 data from qualified enterprise respondents found intent to adopt hybrid retrieval tripling from 10.3% in January to 33.3% in March, the fastest-growing strategic position in the dataset. PixelRAG's own authors point to hybrid deployment as the most practical near-term path — layering visual retrieval on top of existing text systems rather than replacing them. For teams already running RAG pipelines, the path to those savings is more straightforward than a ground-up rebuild. "A practical path is to use PixelRAG as an enhancement layer alongside existing text retrieval systems," Wang said. "Hybrid retrieval that combines both text and visual search is straightforward and is likely how many production deployments would evolve."
Score: 69🌐 MovesJun 12, 2026https://venturebeat.com/data/pixelrag-beats-text-parsers-on-accuracy-and-cuts-ai-agent-token-costs-10x - Cisco Live 2026 Fits Collaboration Into Cisco’s AI Platform Strategy
The tech giant is framing itself as the control plane for AI-driven enterprise infrastructure, with greater focus on how collaboration fits the "one Cisco" strategy.
- ChatGPT Was 76% of the AI Market. It Just Fell to 54%, and Quality Isn’t Why.
For anyone who assumed OpenAI already won: what a new market report actually shows, why ChatGPT is bleeding share despite getting better, and the one factor that is quietly deciding the whole race. Made by Author. For most people, ChatGPT is not an AI product. It is the AI product. The name that became shorthand for the entire technology, the way Google became shorthand for search. Which is why a figure in a recent market report should land harder than it has. In early 2025, ChatGPT accounted for more than three-quarters of all visits to AI chatbots. By this spring, that share had slipped to barely half. The product that supposedly won the AI race is steadily losing it, and the reason is not that it got worse. That last part is what makes this interesting. The easy explanation for a falling market leader is a failing product. ChatGPT did not fail. It kept improving the entire time its share was dropping. So something other than quality is moving these numbers. And figuring out what it is tells you more about who wins the next decade of AI than any benchmark does. 76 Percent to 55, and Claude Up 306 Percent The data comes from market analyses tracking visits to AI chatbots, and the shape of it is clear even if the exact figures will keep moving. ChatGPT’s share of chatbot web visits has fallen from roughly 76 percent in early 2025 to around 55 percent this spring, by Momentic’s reading of Similarweb data. Google’s Gemini has climbed to about 27 percent, growing at a triple-digit rate over a matter of months. Anthropic’s Claude, while smaller in absolute terms, has been the fastest riser of all, up around 306 percent in a single quarter off a small base. DeepSeek and Grok fill out the long tail at a few percent each. Made by Author. One caveat matters before you read too much into any single number. These figures track visits to websites, not total usage, and they miss enormous channels: the ChatGPT mobile app, the API that powers thousands of other products, and the AI now embedded directly into software people already use. ChatGPT’s real footprint across all of those is larger than a web-visit chart suggests. But the direction is the part that is hard to argue with. Across surface after surface, the one-product era is ending, and the share is flowing toward specific rivals for a specific reason. The Quality Story Stopped Fitting My first instinct was to explain the shift the way the industry usually does, as a story about model quality. Maybe Gemini finally caught up. Maybe Claude pulled ahead on certain tasks. There is some truth in that, and the models really have converged in capability. Then I looked at where the share is actually going, and the quality story stopped fitting. The two products eating into ChatGPT are not winning primarily because someone declared their models the best. They are winning because of where they live. Gemini Ships Pre-Installed. ChatGPT You Have to Open. Here is what the rising products have in common, and it is not a benchmark. Gemini is made by Google, which means it does not have to win users one at a time. It arrives pre-installed in the search bar, in Android, in Gmail and Docs, in the Workspace that hundreds of millions of people open for work every morning, and now, after a new deal, inside Apple’s Siri. Google does not need to convince you to try Gemini. It puts Gemini in front of you in the places you already are. Claude’s rise runs on a narrower version of the same logic. It became the default in a high-value niche, the coding and developer and enterprise workflows where it is wired directly into the tools professionals already live in. ChatGPT, for all its fame, has to do the hard thing. It has to get you to open a separate app or a separate website and choose it, deliberately, every time. That is the most expensive kind of growth there is, and it is the kind that erodes the moment competitors with built-in distribution decide to show up. The lesson underneath the numbers is blunt: in consumer AI, distribution is beating quality. Whoever owns the front door, the search box, the operating system, the work software, the phone, gets the usage. Almost regardless of which model would win a head-to-head test. The Front-Door Test If you want a way to predict who wins this race, stop reading benchmark comparisons and start asking a different question. Call it the front-door test. For any AI assistant, ask: does it have to be sought out, or does it arrive by default? Where does it already live without the user choosing it? Who owns the surfaces, the devices, the search bars, the work tools, where people will meet AI without going looking for it? Made by Author. Apply that and the market report stops being surprising. Gemini lives behind doors Google already owns, which is most of them. Claude lives behind the doors of the professional tools that pay for it. ChatGPT lives behind a door you have to decide to walk through, which is why the most famous product in AI is the one losing share. Being the best is a real advantage. Being the default is a bigger one, and the companies that own the defaults are using them. Even If the Chart Is Wrong, the Pattern Holds The skeptical reading deserves a fair hearing. A web-visit chart, the argument goes, is a weak proxy for the real AI market, and ChatGPT’s strength in mobile, in its developer platform, and in paid subscriptions is invisible to it. Lean on the wrong metric and you will see a collapse where there is only a maturing market with more players. There is also a case that some of the share shift is simply the field normalizing after ChatGPT’s unnatural early monopoly, which was never going to last. Both points are fair, and neither rescues ChatGPT from the underlying dynamic. Even if you throw out the exact percentages and grant that ChatGPT’s total reach is bigger than any single chart shows, the structural force remains. The competitors gaining ground are doing it through ownership of distribution that compounds over time, not through a one-off product win that can be reversed with a better release. A maturing market is real, but it does not explain why the share is flowing specifically to the two companies with the deepest built-in distribution rather than spreading evenly. The metric is imperfect. The pattern it is pointing at is not. Can OpenAI Build Its Own Front Door in Time? The question that decides the next phase is whether OpenAI can build distribution of its own before the companies that already have it close the door. That is what the talk of OpenAI building hardware, striking its own platform deals, and embedding itself into other products is really about. It is an attempt to stop being the app you have to choose and become the default you simply encounter. Made by Author. Watch whether it succeeds. Because if it cannot, the most recognizable name in artificial intelligence could end up in the strange position of having taught the world to use AI, and then watching the companies that own the phones, the search bars, and the office software collect most of the users it created. Being first made ChatGPT a household word. Whether that is enough is the thing the falling numbers are quietly asking. ChatGPT Was 76% of the AI Market. It Just Fell to 54%, and Quality Isn’t Why. was originally published in Towards AI on Medium, where people are continuing the conversation by highlighting and responding to this story.
- Nvidia's high-speed AI data center storage servers break cover, touting 2.9 petabytes of storage and extreme PCIe 6.0 performance — Wiwynn shows off SCADA server with GPU-accelerated storage
Wiwynn is among the first to demonstrate Nvidia SCADA server that promises to offer AI systems petabytes of ultra-fast storage thanks to GPU-accelerated storage acceleration.
- Baidu, PostBus get regulatory nod for robotaxi service in eastern Switzerland
Baidu's self-driving service, AmiGo, has gained regulatory approval in Switzerland. This joint venture with PostBus plans to launch Europe's largest automated public transport operation. Fully driverless services are targeted for early 2027. The initiative utilizes electric RT6 vehicles, equipped with advanced sensors. AmiGo will be accessible via a dedicated app, marking a significant step in autonomous public transit.
- Google researchers introduce 'faithful uncertainty,' allowing LLMs to offer best guesses instead of hallucinations
Large language models continue to struggle with hallucinations, presenting a major roadblock for real-world enterprise applications. Reducing these errors is a messy business, forcing model developers to navigate a strict tradeoff where eliminating factual errors often suppresses valid answers. In a new paper , Google researchers introduce the concept of "faithful uncertainty," a metacognitive technique that aligns a model's response with its internal confidence. This alignment allows the model to offer appropriately hedged hypotheses, such as "My best guess is," instead of defaulting to an unhelpful "answer-or-abstain" binary. In real-world agentic AI applications, this metacognitive awareness acts as an essential control layer. It empowers autonomous systems to accurately determine when their internal knowledge is sufficient and when they must dynamically trigger external tools or search APIs to resolve deficits. The utility tax of current mitigation strategies Understanding why LLMs hallucinate hinges on separating two capabilities: a model knowing facts versus knowing what is known. Historically, most factuality gains in AI have come from expanding the knowledge boundary, meaning developers simply pack more facts into the model's parameters through larger scale and more training data. However, expanding a model's knowledge does not automatically improve its boundary awareness, which is its ability to distinguish the known from the unknown and recognize its own limitations. “There are broadly two ways to improve LLM factuality,” Gal Yona, Research Scientist at Google and co-author of the paper, told VentureBeat. The first is continuing to teach the model more facts. But, Yona notes, “model capacity is finite, and the long tail of knowledge is effectively infinite.” Once models hit this limit, the hope is they know what they don't know and simply abstain from answering. However, this is inherently difficult for LLMs. “This is why most practical attempts to reduce hallucinations through various interventions don't actually make it to deployment,” Yona explains. “They do reduce hallucinations, but they also hurt utility, because the model ends up refusing to answer questions it actually does know.” This inability to distinguish between knowns and unknowns creates what the paper's authors call the "utility tax." Enforcing a zero-hallucination standard requires the model to abstain whenever it is even slightly uncertain, discarding massive volumes of completely valid information. For example, the authors demonstrate that reducing an underlying 25% error rate down to a strict 5% target forces developers to discard 52% of the model's correct answers. Treating all errors as hallucinations forces enterprise systems to choose between trustworthiness and helpfulness. Application developers are generally unwilling to pay this massive utility tax and render their models unhelpful. Consequently, they optimize systems to prioritize coverage, forcing models to operate in a state where they continue to generate confident hallucinations. Reframing hallucinations as confident errors To move past the utility tax, the researchers propose to stop treating any factual error as a hallucination. Instead, they reframe hallucinations as "confident errors": incorrect information delivered authoritatively without appropriate qualification. This subtle reframing dissolves the strict "answer-or-abstain" dichotomy and allows the model to express its uncertainty. In this new framework, if a model makes a factual mistake but appropriately hedges its response (e.g., by stating, "I am not completely sure, but I think..."), it isn't a hallucination. It is simply a hypothesis offered to the user for consideration. By expressing uncertainty, the AI preserves its utility—sharing whatever partial or likely knowledge it has—without violating the user's trust. However, if an AI assistant hedges all its responses with a disclaimer, the user is forced to double-check everything, defeating the purpose of the tool entirely. The solution the researchers propose is "faithful uncertainty." This approach requires aligning a model's linguistic uncertainty, or the words it uses to express doubt, with its intrinsic uncertainty, which is its actual, internal statistical confidence in that specific answer. This ensures the model only hedges when its internal state genuinely reflects conflicting or low-probability information. Faithful uncertainty forms a core component of “metacognition,” the AI's ability to be aware of its own uncertainty and act on it. To understand this practically, consider the intuitive example of consulting a doctor. We do not trust doctors because they are all-knowing. We trust them because they reliably distinguish between a confident diagnosis ("You have a fracture") and an educated hypothesis ("It might be a sprain, but let's run some tests"). Practical implications for enterprise AI Under the new framing, errors where a model is genuinely confident but factually incorrect are categorized as “honest mistakes.” This casts knowledge expansion (training the model on more data) and faithful uncertainty as completely complementary efforts. Knowledge expansion pushes the absolute knowledge boundary outward to minimize honest mistakes, while faithful uncertainty honestly communicates wherever that boundary currently lies. This new framing has important implications for agentic applications. The shift to agentic AI might make it seem like knowing what the model doesn't know is redundant, since models can just search external databases. However, access to external tools actually amplifies the need for faithful uncertainty. In agentic systems, metacognition becomes the central control layer that governs the entire system. External tools solve the storage problem because the model no longer needs to encode every fact into its parameters. However, this introduces a new control problem: managing when to retrieve information, verify facts, and orchestrate these external tools. Without faithful uncertainty, an agent is essentially flying blind and must rely on external, static heuristics or over-engineered scaffolds. “The model might search for something it already knows confidently—wasting latency and cost for no gain. Or the opposite: it confidently answers from memory when it should have searched, producing a plausible but wrong output,” Yona said. Today’s agent harnesses try to solve this externally with query classifiers or always-search rules, but Yona notes that these are "static and brittle." By using its intrinsic uncertainty to regulate its own behavior, the agent dynamically optimizes its tool use, choosing to invoke a search tool only when its internal confidence is genuinely low. Beyond deciding when to search, faithful uncertainty is critical for evaluating the results of a search. If a tool returns low-quality or unexpected information, a metacognitive agent does not blindly accept whatever appears in its context window. Instead, it uses its uncertainty awareness to weigh the retrieved external signals against its own internal priors. This prevents sycophantic behavior where the system might otherwise trust external sources that conflict with its actual known knowledge. The bootstrapping paradox: The catch to teaching uncertainty For enterprise builders, achieving this faithful uncertainty is trickier than it sounds. It requires teaching models the syntax of uncertainty through supervised fine-tuning (SFT). Because pre-trained models are mostly fed authoritative text, they must be explicitly taught to say things like, "I'm not entirely sure, but I think VentureBeat was founded in..." But SFT introduces a "bootstrapping paradox." Unlike standard training datasets where the "right answer" is the same regardless of the model, the ground truth for uncertainty is the model's own dynamic knowledge base. “Here's the catch: the 'correct' expression of uncertainty is inherently dynamic, because it depends on what this particular model knows or doesn't know at this particular point in training,” Yona said. “If you train on a label that says 'I don't know X' but the model actually does know X, you've taught it to hallucinate uncertainty... The training data is static, but the target is a moving one, and that's the fundamental tension teams need to grapple with.” The road to self-aware AI For enterprises looking to implement these capabilities without expensive retraining, prompting serves as the most accessible entry point. “Prompt engineering is already something most engineers do today, this provides the lowest-friction path to improving metacognitive behavior today,” Yona said. Enterprise developers can explore frameworks like MetaFaith , an open-source project previously co-authored by Yona, to begin applying metacognitive prompting to off-the-shelf models. However, Yona cautions that "there is still substantial headroom that prompting alone doesn’t solve," meaning the industry will eventually need to rely on advanced reinforcement learning (RL) to bake metacognition deeply into model training. Ultimately, as enterprises transition from isolated chat applications to complex, multi-agent workflows, self-awareness will become a defining prerequisite for reliable autonomy. But evaluating whether a model truly possesses this awareness remains a profound technical challenge. “How do you actually evaluate whether a model can sense its internal states?” Yona asks. “Even in humans, it’s hard to define or separate 'true' self-monitoring abilities from a capable reliance on proxies. We face exactly the same challenges with LLMs: a model might learn to mimic the style of uncertainty without truly sensing its internal state. Developing evaluation frameworks that can tell the difference is one of the most important open problems in this space.”
- Gemini 3.5 Flash lands on Google’s Android coding rankings, but it’s 3x the cost for slower performance
Google has released another set of benchmark results to determine the best AI models for Android coding, along with how much each model costs per token. Google’s Gemini 3.5 Flash is easily the most resource-intensive in Android development, and it doesn’t even make the top five.
Score: 68🌐 MovesJun 12, 2026https://9to5google.com/2026/06/12/gemini-3-5-flash-on-googles-android-coding-rankings/ - The Palantir CEO Just Accused AI Labs of Tokenmaxxing at Corporate Customers’ Expense
Many AI providers bill business customers according to how much information their models process. Alex Karp says companies are now questioning what they receive in return.
- Walmart plans to bring delivery drones to Philly. What that means for Amazon and the same-day delivery wars.
Walmart plans to bring delivery drones to Philly. What that means for Amazon and the same-day delivery wars. Inquirer.com
Score: 68🌐 MovesJun 12, 2026https://www.inquirer.com/news/walmart-wing-amazon-same-day-drone-delivery-philadelphia-20260612.html - Craig Federighi Swipes at AI Rivals Who Are 'Racing Forward' Without Regard for Users
Apple software chief Craig Federighi used the WWDC 2026 keynote to draw a pointed contrast between Apple's approach to artificial intelligence and the broader industry, suggesting that some competitors are developing AI without meaningful consideration for the people using it. During Monday's WWDC 2026 keynote, Federighi said: AI is incredibly powerful technology. Still, some appear to be racing forward, seemingly pursuing AI for the sake of AI, without clear regard to the people, all of us, that it's ultimately meant to serve. The remarks appeared to be aimed squarely at rivals including OpenAI, Google, and Meta , all of which have aggressively shipped AI products and services over the past two years. Federighi argued that Apple's conservative approach is more useful because it draws on personal context. The comments arrived alongside Apple's unveiling of Siri AI , a ground-up rebuild of its digital assistant powered by the next generation of Apple Intelligence . Federighi described the effort as "a big leap forward," with "an innovative architecture that unlocks a new Siri across platforms." Apple said it has created a second version of its Apple Foundation Models capable of understanding speech and reading text and images, with a new system orchestrator coordinating capabilities across its platforms. The implicit dig at competitors carries some irony given Apple's own recent history with AI. The company spent the better part of two years struggling to deliver a meaningfully improved Siri, and earlier this year parted ways with John Giannandrea , its former head of AI and machine learning, following a prolonged restructuring of its AI teams. Federighi pushed back against the idea that the new Siri is simply another " bolted-on chatbot ," saying the company sees it as "an integral but conversational tool that you use in the moment." Privacy, he said, is "non-negotiable," with data used only to execute a user's request. Tag: Craig Federighi This article, " Craig Federighi Swipes at AI Rivals Who Are 'Racing Forward' Without Regard for Users " first appeared on MacRumors.com Discuss this article in our forums
- "Age of Robots Making Human-Like Judgments, KAIST Solves Key Challenge in Physical AI
"Age of Robots Making Human-Like Judgments, KAIST Solves Key Challenge in Physical AI EurekAlert!