AI News Archive: June 9, 2026 — Part 7
Sourced from 500+ daily AI sources, scored by relevance.
- Tealium Context API unites data cloud depth with real-time customer context for AI agents, apps, and experiences
Tealium Context API unites data cloud depth with real-time customer context for AI agents, apps, and experiences Toronto Star
- Nextcloud adds Euro-Office to Hub workplace suite, expands AI assistant
MUNICH — Nextcloud has integrated Euro-Office into its workplace application suite, one of several updates to Nextcloud Hub unveiled on Tuesday that include a new compliance app for large organizations and a program to support developers building for its platform. The announcements came during the company’s Nextcloud Summit 2026 here. Euro-Office, announced in March , is billed as an open source, sovereign alternative to Microsoft Office for European organizations keen to reduce their reliance on US tech providers. It consists of four browser-based applications: a document editor, spreadsheet program, presentation tool, and a PDF editor — each enabling collaborative editing. Euro-Office documents can also be opened directly from the Nextcloud Files mobile app. Nextcloud is one of several European companies that support Euro-Office, which is built on the open-source code base of OnlyOffice and distributed under the GNU Affero General Public License v3 (AGPL v3). The integraton means Nextcloud users can now choose between two options in Nextcloud Office: Euro-Office and the existing Collabora integration. “Euro-Office uses a different architectural approach that can result in a better performance in the browser, a different user experience…, so it’s important that this option is available,” Jos Poortvliet, Nextcloud co-founder and vice president of communications, said at the Tuesday event. Other changes in the Nextcloud Hub 26 Spring release include updates to Nextcloud‘s Talk video and voice meeting app, including AI noise suppression and the ability to start a call from any Nextcloud Hub app – an addition that will make collaborative editing easier, said Poortvliet. For Nextcloud Assistant, there are new AI agent capabilities. In addition to existing capabilities such as managing calendars and tasks, AI agents can now create cards in Nextcloud’s Deck task management app and update information in the Forms app. There are also improvements to the AI assistant’s interface, which can be moved around to avoid blocking other applications and allow users to copy and paste text more easily without opening another tab. To meet EU AI Act requirements, Nextcloud will make it easier to see which provider supplies the large language model (LLM) the Assistant runs on. Nextcloud will also integrate the AI assistant directly into its Nextcloud Office suites via a sidebar chat interface, allowing users to address problems such as errors in the spreadsheet app. NextCloud’s AI chat assistant is integrated into the company’s Office suites. NextCloud There’s also a new Governance app that helps large organizations — particularly governments and highly regulated industries — meet regulatory requirements with compliance tools to manage data held in Nextcloud Hub. It contains several features, including sensitivity labels to control access rights; data retention and archive capabilities; and a legal hold option that preserves documents for legal purposes such as a court case. The Governance app includes a Compliance Manager that provides a compliance score based on an organization’s regulatory requirements, and measures progress towards certain targets. Admins can also search and review documents shared by employees and generate audit reports for compliance. The Governance app is available to Nextcloud Enterprise customers. Nextcloud also launched a program to support independent software providers interested in building apps on its platform. With AI making it easier for developers to build software that integrates with its platform, Nextcloud expects a 10-fold increase in the number of available apps — from 600 now to 6,000 over the next 12 months, according to Nextcloud CEO Frank Karlitschek. Nextcloud promised to promote apps developed by partners in its App Store and sell subscriptions as part of the ISV program, as well as provide documentation and technical help to customers. In return, developers would provide guarantees to customers around security processes and long-term support. “We can strengthen our ecosystem, the developers also make some money — because obviously we do a revenue share here — and we leverage the dynamics that we expect from AI coming very soon,” said Karlitschek. Editor’s note: NextCloud paid for Matthew Finnegan’s travel and hotel costs for NextCloud Summit 2026, but had no editorial role in the creation of this story.
- lean4-skills: protocols for inspectable AI-assisted Lean formalisation
lean4-skills: protocols for inspectable AI-assisted Lean formalisation Department of Pure Mathematics and Mathematical Statistics (DPMMS)
- AI glasses reportedly used to film flight attendants without consent, Rokid responds with action plan
A pair of seemingly ordinary glasses can be turned into a covert recording device with nothing more than a low-cost light-blocking sticker. This is no longer a science fiction scenario, but a real-world challenge emerging as AI-powered smart glasses gain wider adoption in China in 2026. This week, the topic “Are smart glasses becoming a […]
- Deliverance AI exits stealth to power sovereign enterprise AI
Deliverance AI, a UK-founded provider ofenterprise AI infrastructure, has emerged from stealth, reporting £6 million inannual recurring revenue (ARR), more than 30 employees, and six enterprisecustome...
Score: 32🌐 MovesJun 9, 2026https://tech.eu/2026/06/09/deliverance-ai-exits-stealth-to-power-sovereign-enterprise-ai/ - Constellation Research CEO R 'Ray' Wang discusses how 'Siri AI' shakes up AI ecosystem
Constellation Research CEO R 'Ray' Wang discusses how 'Siri AI' shakes up AI ecosystem
- Powering the future of robotics in Europe
Powering the future of robotics in Europe
- From Large Language Models to Evidence-Grounded AI Systems
From Large Language Models to Evidence-Grounded AI Systems cst.cam.ac.uk
- Enterprises that succeed in agentic AI start by ‘reimagining’ business process, finds Pega research
Organisations that succeed in rolling out agentic artificial intelligence (AI) in their enterprises start by rethinking their business processes , according to business and IT decision-makers. A survey of 500 business and IT decision-makers who have successfully introduced agent-powered AI into their businesses found they had several things in common. The companies that succeeded maximised the benefits of their AI projects by fostering a culture of collaboration and innovation, according to a Pegasystems study of successful agentic AI implementations. The successful organisations had a corporate-level strategy and plan for agentic AI execution, and a top-down strategy in place, said Don Schuerman, Pega’s chief technology officer. “There is still a lot of pressure, especially from boards, to drive technology for technology’s sake, as opposed to a specific solution to a specific business problem,” he said. More than half (53%) of business leaders said they had changed their existing business processes to a “significant” extent by reimagining everything their organisation does to gain maximum benefit from their agentic implementations. And 80% of the successful organisations agreed that business and IT teams were willing to embrace new technology, innovation and ideas to explore new possibilities. According to Pegasystems, which carried out the study with research firm Savanta, the companies that succeeded were motivated by a desire to produce consistent, predictable outcomes. Three-quarters (71%) of successful agentic AI implementers said one of their top two pre-deployment objectives was to automate and simplify complex processes so they work consistently and predictably across systems and platforms. Over half (58%) also reported they had already seen predictable outcomes, reduced complexity and improved customer experiences. Metrics and strategies The research showed that companies that had successfully implemented agentic AI projects had clearly defined metrics and strategies. Some 95% of those have a specific corporate-level strategy and plan for execution, and 65% have comprehensive, pre-agreed success metrics tied to business outcomes that are regularly reviewed to monitor implementation success. The winners in the agentic era will not be those who deploy agents wherever and whenever they can. They will be those who reimagine themselves and find new ways to give clients and their customers what they want Don Schuerman, Pegasystems Almost two-thirds (61%) said they start an agentic project with the expectation it will “significantly” improve customer experience once fully integrated, which more than half (58%) begin those projects believing they will realise significant, measurable value – including the potential for both increased customer satisfaction and cost reduction. When asked to name leading barriers to achieving a positive agentic project outcome, over three-quarters (77%) pointed to a lack of sufficient resources. Three-quarters (75%) agreed that a lack of knowledge and understanding of the benefits agentic AI can bring to the business is the biggest barrier to agentic AI success. “We’re fast reaching a tipping point with agentic AI where adoption is high within organisations, but maturity is not,” said Schuerman. “The value will come from rethinking ways of working and aligning culture around what AI makes possible. Those changes are what separates the promise of AI technology from the reality of creating truly transformational benefits,” he added. “The winners in the agentic era will not be those who deploy agents wherever and whenever they can. They will be those who reimagine themselves and find new ways to give clients and their customers what they want,” he said. Read more about Pegasystems Vodafone Greece automates deals for customers, saves 500 staff-days of work : Vodafone Greece hired an implementation partner for a business process management project while its own staff observed and learned how to use the technology. Wells Fargo bank turns to AI to help families settle estates after a death : Wells Fargo bank is winning customers after using business automation software and artificial intelligence to help people manage the estates of relatives following a bereavement. Citi US Personal Banking turns to AI to ‘delight’ customers with personalised services: Citigroup’s US Personal Banking business has created a repository of customer data and is rolling out a decision engine. Bupa turns to data to provide personalised health services : Private healthcare provider Bupa says a project to deploy business process automation is bringing it closer to APAC customers. Pegasystems refines Blueprint agent builder, expands marketing tools : Pegasystems emphasises ‘derisking’ agentic buildouts for its customers in regulated industries.
- AI job loss scare looms over American politics
AI job loss scare looms over American politics The Straits Times
Score: 31🌐 MovesJun 9, 2026https://www.straitstimes.com/world/united-states/ai-job-loss-scare-looms-over-american-politics?ref=latest - DTEX adds AI Risk Management to track how agents and employees use AI
Behavioral intelligence security company DTEX Systems Inc. today introduced an expanded AI Risk Management product that reads the intent behind how employees and autonomous artificial intelligence agents use generative AI tools across the enterprise. The release targets a gap DTEX argues most security tools have yet to close. As copilots, generative AI applications and AI […] The post DTEX adds AI Risk Management to track how agents and employees use AI appeared first on SiliconANGLE .
Score: 31🌐 MovesJun 9, 2026https://siliconangle.com/2026/06/09/dtex-adds-ai-risk-management-track-agents-employees-use-ai/ - Economic slump, not AI, is the real job killer, says LinkedIn executive
Economic slump, not AI, is the real job killer, says LinkedIn executive The National
- AI transforms wealth management in UAE as advisers blend data and judgement
AI transforms wealth management in UAE as advisers blend data and judgement
- It’s the year of AI transformation for these three industries. Here’s why
For CIOs across every industry, enterprise AI is inescapable right now. Everyone has a pilot running, every conference has a keynote about transformation and every vendor is promising agents that will change everything. But underneath the surface, I’ve noticed that the organizations making the most meaningful headway are clustering in three industries: financial services, industrials and healthcare. That’s because these sectors share a specific combination of factors that make them well-suited for what frontier LLMs in 2026 are best at. Each of these industries is drowning in unstructured data, their best people spend too much time on low-value, document-heavy work, and the underlying infrastructure is in place (cloud storage, APIs, data warehouses). All that’s been missing is a layer intelligent enough to put it to work, and now that layer exists. Financial services: Sitting on a goldmine Financial services has been data-rich and insight-poor for decades. The problem was never a lack of information, rather, that the information lived in PDFs, SharePoint sites and folders that nobody could easily access or analyze at scale. Resultingly, decisions were made without full context, compliance work was done manually under time pressure and senior people spent their hours on tasks that shouldn’t require their expertise. AI changes all of that. According to KPMG research , 80% of PE leaders view generative AI as a critical component for gaining competitive advantage and market share. 91% believe AI has already strengthened their competitive position, and more than half are already seeing a return on their investment. I spoke recently with a CIO at a large wealth management firm who described the moment it clicked for their team. They had been trying to figure out how to get their advisors to do more proactive outreach by reaching the right clients at the right moment rather than reacting slowly to inbound calls. The issue here was that pulling together existing information and context manually wasn’t something any advisor had time to do. So, they built an AI workflow that runs on a trigger each morning and analyzes client portfolios, market conditions and advisor notes. Then, it generates a prioritized outreach list with suggested talking points. It now runs across their entire book of business. Here’s another example. I’ve seen multiple private equity firms using AI agents to generate portfolio summaries, extract data from quarterly reports and run fundamentals-based valuations. That’s work that used to consume analyst hours every week before an investment committee meeting. What makes financial services ready for this moment is partly about infrastructure. Most institutions already have centralized document stores, CRMs and data warehouses. They don’t need to build the foundation. They need an intelligent layer on top of what already exists. The other factor is regulatory pressure: It’s not glamorous, but AI that can demonstrate auditability and consistency has a tangible advantage in compliance-heavy environments. Consistency is something humans, under volume and time pressure, struggle to deliver and it’s particularly important for financial institutions given the amount of sensitive data they work with. For CIOs thinking about where to start, I’d say that document-heavy workflows are almost always the right entry point. Term sheet parsing, compliance matrix generation, report summarization. They’re well-defined, they happen constantly and the ROI is easy to measure. Build for auditability from the beginning: Every run must be logged, every output must be cited and human-in-the-loop should almost always be involved. Lastly, I think we’ll see fewer chatbots and more trigger-configured agents in 2026, as the highest-value financial AI in production today runs on event-based logic, not on-demand queries. Industrials: Where traditional automation always broke down Industrial companies — spanning construction, manufacturing, logistics/shipping, engineering and more — have historically been underserved by enterprise software, which is a structural issue. The workflows span physical and digital worlds in ways that make them challenging to automate through conventional means: Tenders arrive as PDFs in someone’s inbox; quality inspections happen on a factory floor; freight analysis requires pulling data from a dozen carrier systems that don’t talk to each other, and often, from people who literally speak different languages. But everything has changed. According to a 2026 survey by the Manufacturing Leadership Council , 90% of manufacturers surveyed say they will increase generative AI usage in the next two years. I had a conversation last year with the CIO of a major national distribution company, where he told me that they’d automated their freight analysis reports entirely, going from a chatbot-style prototype to a fully templated, automated report that runs on a schedule and lands in the right inboxes. Another global consumer goods manufacturer I worked with now processes quality inspection sheets from production lines through AI, automatically flagging anomalies before they become problems. And one of the largest civil engineering firms in the U.S. now uses AI to do quality control on bridge inspection reports, check engineering calculations and navigate RFP documents, significantly reducing the review burden on senior engineers who were previously spending time on work that simply didn’t require their expertise. The thing I’ve heard CIOs in the industrial sector tell me is that the skilled worker shortage is real and getting worse. They have experienced people who are spending a significant portion of their time on tasks that could be automated. Giving those hours back to them is the value proposition. In 2026, AI excels precisely where RPA and EDI always broke down: unstructured inputs, variable formatting, anomalous edge cases. So, the practical advice here is to target the gap between documents and systems: That’s the place where a human is manually transcribing data from one format into another. Start with one high-volume vendor or one product line, design the workflow and track the ROI. Healthcare: The burnout crisis that AI is starting to solve Healthcare has been the most cautious sector for extremely legitimate reasons. PHI/PII, HIPAA, GDPR, the complexity of clinical workflows…the bar is higher here, as it should be. But already this year I’ve watched healthcare move from cautious experimentation into production deployment, and the driver is the combination of enterprise-grade security controls and a clinician burnout crisis that has become impossible to ignore. According to McKinsey, half of healthcare leaders report that their organizations have already implemented generative AI. The use case I keep coming back to is clinical note generation. I’ve seen multiple healthcare organizations (virtual care platforms, primary care networks and more) deploy AI that listens to patient encounters and produces structured SOAP notes. One organization has been continuously improving this workflow and is now on their fifth or sixth version of the workflow. But they started seeing the impact from day one: The documentation burden on physicians is real, and can consume one to two hours per day, time that should be with patients. Reducing that by 60 to 70 percent is life changing. Beyond documentation, I’m seeing AI handle patient intake and onboarding through conversational workflows that gather history, insurance information and chief complaint before the visit, integrating with EHRs to ensure continuity. Remote patient monitoring programs are using AI to triage incoming data and automatically escalate concerning readings to clinical staff, allowing home health programs to scale without proportional increases in headcount. Finally, on the administrative side, AI is now doing clinical billing compliance review: Checking documentation against billing codes before claims are submitted, reducing denial rates and audit risk. My advice to healthcare CIOs is, after identifying a platform with HIPAA compliance and rigorous governance, to start with use cases in billing compliance, prior authorization and patient communication. Build organizational confidence there before moving into the clinical workflow layer while measuring clinician time saved as your primary ROI metric. Cost reduction matters, but hours returned to patient care is the number that will get you continued investment and internal support. The high-level patterns The industries I’ve identified in this article are ripe for AI transformation. When we step back from the specific use cases, the same conditions show up across all three sectors. Firstly, there are massive volumes of unstructured data that traditional automation has never been able to touch. Secondly, there is high-value human expertise being consumed by low-value data processing and shuffling. Lastly, the underlying tool infrastructure is mature enough to support an intelligent layer on top. CIOs in these industries should aim to identify high-impact workflows, deploy AI that integrates deeply with those processes and be prepared to iterate. The result will be millions in operational savings. This article is published as part of the Foundry Expert Contributor Network. Want to join?
- AI Adoption in the UK is Booming, but Transformation is Stalling
What’s holding organizations back from making the most of AI.
Score: 31🌐 MovesJun 9, 2026https://aibusiness.com/generative-ai/ai-adoption-the-uk-is-booming-transformation-stalling - IIT Delhi and Cadence launch AI-enabled Innovation Lab to advance India's semiconductor talent
Cadence and the Indian Institute of Technology Delhi (IIT Delhi) have announced the IIT Delhi-Cadence Innovation Lab, a multidisciplinary centre of excellence that equips India’s next generation of semiconductor innovators with the same AI‑enabled electronic design automation (EDA) tools and workflows used by industry. The lab advances research, strengthens workforce development and supports pre‑seed startups with a streamlined path to first silicon, aligning with India’s Semiconductor Mission and the Design‑Linked Incentive (DLI) scheme. Providing proven access to 200+ industry‑grade Cadence solutions across four domains – chip design verification, digital implementation, analog design and system design and analysis – the lab ensures students, researchers and educators learn on the exact tools used in professional environments. By embedding “design with AI” across these workflows, the lab targets step‑change gains in engineering productivity and strengthens the integration of AI in VLSI design. IIT Delhi has adopted Cadence‑developed courses that combine theory with comprehensive, project‑based labs and assessments, moving beyond a theory‑only model to hands‑on learning with real tools and real‑world problem statements. Guest lectures from Cadence and industry practitioners further align learning with current technology roadmaps and career pathways. To catalyse research and early-career exploration, the lab is introducing an Early Master’s Research pathway for select fourth‑year undergraduates from IITs and NITs, mentored by Cadence experts and IIT Delhi faculty across multiple research areas. In parallel, the lab’s incubator programme supports pre‑seed startups on a case‑by‑case basis with a low‑cost route to first tape‑out and a working prototype. “Students at IIT Delhi now use the same AI‑enabled tools they’ll see on day one in industry, closing the gap from classroom to tapeout,” said Alok Jain, Corporate VP and India MD, Cadence. “Pairing industry‑grade technology with project‑based curricula, real‑world challenges and targeted startup support strengthens research relevance and workforce readiness for India’s semiconductor future.” “The IIT Delhi–Cadence Innovation Lab combines top‑tier academic rigour with cutting‑edge industry tools,” said Prof. Jayadeva, Prof. In-Charge, Cadence-IIT Delhi Innovation Lab. “This partnership will expand research output, prepare students for high‑impact careers and help founders move from ideas to prototypes, supporting the goals of the India Semiconductor Mission and the DLI scheme.” Looking ahead, the Lab is committed to significantly increasing research output, graduating talent ready to contribute on day one, and cementing high‑impact collaborations that translate academic innovation into industrial outcomes.
Score: 31🌐 MovesJun 9, 2026https://www.dqindia.com/esdm/cadence-and-iit-delhi-announce-iit-delhi-cadence-innovation-lab-12004504 - Axonius Extends Asset Intelligence into AI Security, Starting with Claude Enterprise
Axonius Extends Asset Intelligence into AI Security, Starting with Claude Enterprise markets.businessinsider.com
- Wonder Studios Wants To Turn AI Short Films Into The Next Big IP
Wonder Studios’ Beyond the Loop uses Hal Watmough’s The Trials to show how AI short films are moving beyond prompt demos into a test of craft, ownership and IP
- When AI fabricates your quotes
When AI fabricates your quotes marketplace.org
Score: 30🌐 MovesJun 9, 2026https://www.marketplace.org/episode/2026/06/09/dont-forget-ai-is-not-infallible - Survey Surfaces Emerging DevOps Bottlenecks in the AI Coding Era
Survey Surfaces Emerging DevOps Bottlenecks in the AI Coding Era DevOps.com
Score: 30🌐 MovesJun 9, 2026https://devops.com/survey-surfaces-emerging-devops-bottlenecks-in-the-ai-coding-era/ - How to bridge the global AI divide
How to bridge the global AI divide Brookings
- Apple’s best AI idea looks a lot like vibe coding
Siri AI and Image Playground will get all the hype, but Apple can actually make our lives better in Safari and Shortcuts.
Score: 30🌐 MovesJun 9, 2026https://www.theverge.com/tech/946733/apple-shortcuts-ai-safari-tabs-vibe-code - Sorry, Apple: ChatGPT Got There First and Siri Lost It's Job
Sorry, Apple: ChatGPT Got There First and Siri Lost It's Job Tom's Guide
Score: 30🌐 MovesJun 9, 2026https://www.tomsguide.com/ai/apple-spent-years-rebuilding-siri-but-chatgpt-changed-what-people-want-from-ai - AI power efficiency the target of Lotus Microsystems energy advances
Lotus Microsystems has introduced vStrata, a new power-delivery architecture aimed at Improving data center power efficiency, a pressing concern even in a non- AI environment . At the heart of the platform is the company’s proprietary Power Interposer Technology (PIT), a silicon-based interposer architecture that enables power conversion and delivery closer to the processor package. The PIT uses a vertical power delivery (VPD) chip and package designed to deliver electrical power directly through the package stack to the processor. By shortening current paths and integrating thermal management directly into the power-delivery structure, vStrata aims to reduce conversion losses while improving cooling efficiency. According to Lotus Microsystems , the module can achieve point-of-load efficiencies of up to 96% while reducing power-conversion losses by more than 50% compared with conventional approaches. “We focus very much on a topology technology that is more efficient, so it basically means that for the amount of power that you put into the power converter, you get more power out, and you have less power losses,” said Hans Hasselby-Andersen, CEO of Lotus. “Another unique thing about our solution is where we utilize our silicon substrate technology to effectively remove the heat from the solution, so where others are focusing on the power side of power delivery, we also handle the thermal issues related to power conversion,” he added. No power converter is 100% efficient, usually about 90% efficient. Lotus’s PID is 96% efficient, making for a 60% reduction in power loss. With banks of power consuming GPUs, that adds up, so much so data centers could potentially stick with air cooling rather than be forced to use liquid cooling. “There’s no doubt that if you deploy this technology across the board, you would definitely be able to reduce the energy that you put into cooling data centers, and not only energy, but issues with water consumption,” said Hasselby-Andersen. Lotus Microsystems states that vStrata maintains compatibility with existing power-management controllers and reference designs, potentially easing adoption among semiconductor and system vendors. vStrata comes in the form of power supplies, and Lotus is working with major server vendors and hyperscalers, but the new power supplies are not suitable for retrofitting into existing server racks. “There’s no industry standard [for server power supplies], so there’s no default footprint you can live up to,” said Hasselby-Andersen. Engineering samples of the LSC0580 – the first vStrata platform module – are scheduled to ship in Q3 2026.
- “Builders of the Next Decade Deserve a Government That Moves at Their Speed” H.E. Ruba Al Hassan On Abu Dhabi's AI-Ready Vision to Support Startup Growth
“Builders of the Next Decade Deserve a Government That Moves at Their Speed” H.E. Ruba Al Hassan On Abu Dhabi's AI-Ready Vision to Support Startup Growth Entrepreneur Middle East
- AI is actually good for IT hiring? Linux Foundation report says yes, but...
New research from the Linux Foundation has a more rosy employment outlook for engineers.
Score: 30🌐 MovesJun 9, 2026https://www.thestack.technology/ai-positive-effect-hiring-it-staff-europe/ - Can an AI agent cover for you while you're at the beach?
"AI can make everything that was on my plate visible to colleagues while I'm gone," one expert said.
- Wall Street holds steadier as AI stocks recover some of their sell-off
Wall Street holds steadier as AI stocks recover some of their sell-off Houston Chronicle
- Optimus Prime will take your call: Hasbro leans into AI-driven licensing
The toy company’s AI studio, Sixth Wall, will debut behavioral licensing, which is focused on how characters think, speak and interact in new experiences.
Score: 30🌐 MovesJun 9, 2026https://www.retaildive.com/news/hasbro-launches-ai-studio-to-license-optimus-prime-mr-potato-head/822225/ - Interview: Emergence CEO Satya Nitta on making AI agents safe for enterprise use
Interview: Emergence CEO Satya Nitta on making AI agents safe for enterprise use verdict.co.uk
- 85% of UAE consumers use AI for online shopping, Visa study finds
85% of UAE consumers use AI for online shopping, Visa study finds
Score: 30🌐 MovesJun 9, 2026https://www.khaleejtimes.com/business/85-of-uae-consumers-use-ai-for-online-shopping-visa-study-finds - Seri të videove të fabrikuara të Albin Kurtit dhe Vjosa Osmanit
Deklarata "Ku po shkon, do të pendohesh Vjosa, shihemi më 7 qershor." Faqet në Facebook Vlerësimi Rrenë Janë të fabrikuara përmes Inteligjencës Artificiale videot e publikuara, në të cilat shfaqen Albin Kurti dhe Vjosa Osmani në situ ... (https://incidentdatabase.ai/cite/1519#7356)
Score: 30🌐 MovesJun 9, 2026https://kallxo.com/krypometer/seri-te-videove-te-fabrikuara-te-albin-kurtit-dhe-vjosa-osmanit/ - High Value Work: The New ROI of AI
Speed and efficiency matter, but the greatest impact comes when humans have time to innovate, think, and nourish relationships.
- Reinforcement Learning for and with Large Language Models: Architecture, Agency, and Algorithmic Discovery
Reinforcement Learning for and with Large Language Models: Architecture, Agency, and Algorithmic Discovery repository.cam.ac.uk
Score: 30🌐 MovesJun 9, 2026https://www.repository.cam.ac.uk/items/ec5e095d-064f-4078-9374-c90a901de2e7 - Can tech companies learn to love cheaper AI models?
If those same AI workloads can be handled by cheaper models without affecting quality, it would mean a massive shift in the economics of AI.
Score: 30🌐 MovesJun 9, 2026https://techcrunch.com/2026/06/09/can-tech-companies-learn-to-love-cheaper-models/ - New Platform Uses Cryptographic Invisibility to Protect AI-Built Applications
Atsign’s AI Architect applies cryptographic protections to agentic software development, aiming to prevent attackers from exploiting vulnerabilities by making application identities effectively invisible. The post New Platform Uses Cryptographic Invisibility to Protect AI-Built Applications appeared first on SecurityWeek .
Score: 30🌐 MovesJun 9, 2026https://www.securityweek.com/new-platform-uses-cryptographic-invisibility-to-protect-ai-built-applications/ - Robots learn to anticipate chaos, but still fail to read a decidedly human signal
Cornell researchers are investigating the potential for using artificial intelligence to give robots social intelligence—the ability to read facial cues, anticipate the needs of those around them, and function within society. The new study tested the ability of vision language models (VLMs)—AI systems that can interpret and generate both visual information and language—to predict whether a tense scenario in a short video would end well or badly, such as a toddler carrying an overly full mug of coffee.
- Human limits when catching AI errors
Human limits when catching AI errors EurekAlert!
- Adopting AI models is easy — scaling them requires shared open standards
The AI market is as competitive as any I have seen. When organizations look to implement the latest AI model or agent platform, many skip over the infrastructure-building required for successful deployment. This instinct is understandable – teams want to move quickly, deliver business impact and avoid falling behind in a fast-paced market. But models and frameworks only deliver value over time if they sit on a foundation built for production, not just initial deployment. As AI evolves from models to copilots to increasingly autonomous agents, the systems behind them must also evolve to support reliable, coordinated behavior at scale. This foundation may not be as exciting as a new model release, but it becomes essential once you deploy AI broadly across an organization and allow access to enterprise tools and data. To responsibly build and scale this foundation, we need interoperable frameworks, shared protocols and secure, community-driven innovation. The agentic ecosystem will not scale or meet developer needs for reliability, security and consistency in isolated, proprietary silos. The most important lessons from scaling cloud systems—shared standards and open-source community innovation—will be directly relevant to the AI era. I’m seeing many of the same patterns emerge as when we built Kubernetes : a community converging on shared interfaces and operational patterns that made it possible to run distributed systems reliably at scale. Over the last decade, workloads have shifted from traditional web applications to AI-native applications, but the underlying operational constraints have remained the same. Lessons from scaling the cloud To understand what this looks like in practice, we can look at how we learned to operate distributed systems in the cloud. While AI introduces unique complexity, the operational shape is familiar. In distributed environments, feedback is slower, failures are complex and harder to diagnose and system-wide updates are difficult to implement safely, increasing the possibility that unnoticed failures accumulate into systemic instability. Those constraints shaped the cloud, and they shape production AI systems as well. Kubernetes didn’t just make it possible to run containers: it addressed the harder problem of how to change live systems without breaking them. The solution wasn’t a single tool, but a set of operational patterns such as health checks, controlled rollouts and a consistent way to describe, review and manage change. Furthermore, the definition of health became more flexible, allowing users to evolve what a “healthy” application means over time within familiar contexts. Another important lesson is the value of good defaults for a healthy system. Letting every team define their own patterns turns every operator into a system expert, which does not scale. If everyone follows their own individual approach, the subtle differences in choices make it impossible to build standardized tools which can work for everyone. This is why modern AI systems need to provide best practices and good defaults while still allowing flexibility to adapt over time. The role of the open-source community in shaping standards Most organizations treat AI as a product launch: ship a model, spot-check outputs and iterate quickly. This works for many features and updates, but it doesn’t work for probabilistic systems, where behavior can drift quietly and without obvious failure modes. AI requires us to move past the mindset that something is either “working” or “broken” and shift to a continuous understanding of output quality. Open-source communities solved this problem for cloud systems by converging on shared interfaces and patterns. That convergence enables ecosystems of tooling and operational practices that make distributed systems repeatable at scale. AI systems need the same kind of convergence and consistency. As agents operate across frameworks, clouds and environments, interoperability becomes critical. This means developing standards for the surfaces every team interacts with: Interfaces for inference and routing Common representation of quality gates and system health Clear telemetry and tracing for understanding system behavior Auditable identity and permissions that follow across multiple systems Standard definitions to describe potential actions and their effects. When standards are in place, organizations can standardize platform defaults, roll out changes gradually and keep rollback paths simple. The good news is that this is an extension of patterns already established in the cloud native ecosystem rather than a complete rethinking of what we need to build. The world of AI stands on the shoulders of a decade of cloud-native technologies, but we must adapt these technologies to the world of AI-native applications . What Kubernetes can teach us about reliable AI systems and operating them at scale Kubernetes worked because it assumed that within any application or service, change is constant and made change manageable by making it observable, staged and reversible. AI systems need the same properties, but with an added dimension: “healthy” also now includes behavior. A model can return responses with low latency and still be wrong in ways that matter. Regressions show up as degraded results, not necessarily errors, which makes them harder to detect. Because of this, “ship it and see” is a poor strategy, especially as agents begin to take on more autonomous roles. Testing a model on one or two prompts is no longer sufficient. You have to run thousands of tests and determine whether outputs have improved. Determining whether a change is better or worse requires evaluation across a wide variety of inputs. In practice, this often means both testing at scale with thousands of inputs and testing in production, where percentages of traffic can be sent through the new model and compared against the existing system. Better models alone won’t produce reliable systems. But a focus on intentional, disciplined operations will. The success of AI systems is tied to user inputs and to the outcomes of probabilistic systems. While probabilistic systems aren’t as straightforward to manage as deterministic software, we’ve learned reliability comes from controlled release processes, observability tied to outcome quality and the ability to roll back quickly. A similar lesson can be applied to operating AI systems at scale, ensuring it’s portable and durable for teams to build on it for years to come. The fastest way to fail with AI is treating it as a feature you ship instead of a system you operate. As organizations move beyond pilots and into production, the bar shifts from “it works” to “it operates safely.” That requires a small set of non-negotiable practices: Treat every model, prompt and data update as a full production release. If you can’t stage, observe and roll back, you’re not in control. Measure full system behavior, not just health. Uptime and latency won’t tell you when output quality is degrading. Design for safe failure. Build fallbacks, guardrails and clear escalation paths before the system is under load. Standardize shared surfaces . Common interfaces, telemetry and release patterns are how operators build muscle memory. Reuse proven patterns . Bad patterns create system failures. Reusable, open patterns reduce surprise. We don’t need to invent a new operational philosophy for AI. We need to apply what Kubernetes and the cloud-native ecosystem already established: standardize where it matters, make change controlled and make system behavior observable. If we apply those lessons early, we avoid relearning them later under production pressure. AI is moving at a fast pace, and we must ensure we’re ready for continued innovation. This article is published as part of the Foundry Expert Contributor Network. Want to join?
- Video të gjeneruara me Inteligjencë Artificiale të Vjosa Osmanit
Deklarata Duk Be Shaka Me Kosoven dhe Me Popullin .Vjosa Osmani Dhe LDK PDK DHE AAK faqet në Facebook Vlerësimi Me kos Është e fabrikuar përmes Inteligjencës Artificiale videoja në të cilën shfaqen figura politike si Vjosa Osmani, Lu ... (https://incidentdatabase.ai/cite/1520#7357)
Score: 29🌐 MovesJun 9, 2026https://kallxo.com/krypometer/video-te-gjeneruara-me-inteligjence-artificiale-te-vjosa-osmanit/ - Crosscheck: Benchmarking AI Models in the Real World
Benchmarking AI models in real-world scenarios
Score: 29🌐 MovesJun 9, 2026https://www.linkedin.com/blog/engineering/ai/crosscheck-benchmarking-ai-models-in-the-real-world - A Mike's-Eye View of ARC's Research
Over the past 15 months or so, ARC's technical agenda has developed quite a bit. The advent of the Matching Sampling Principle (MSP), and ideas like it, has begotten a host of concrete technical problems; progress on those problems has given us more philosophical clarity on the big picture, which has led to even more technical progress. The two most recent public discussions of ARC's research (Jacob's A Bird's Eye View of ARC's Research and David's Obstacles in ARC's research agenda ) both came out before this flywheel really got spinning, and a lot of what we now consider central to the agenda isn't reflected in either of them. The goal of this post is to give a clear, updated picture of what we're actually trying to do. This is written from my point of view; I don't speak for my whole organization. Here is ARC's hoped-for pipeline for aligning a powerful AI: monitor training to detect structure as it is added to the model; convert that structure into advice that improves an MSP-style mechanistic estimator of the model's behavior; use the resulting estimator, together with a description of the relevant input distribution, to estimate a safety-relevant quantity such as the probability of catastrophic failure; [1] then optimize the model against that estimate. The key advantage over black-box evaluation is that we are not waiting for catastrophic behavior to appear often enough in samples, or even to have so much as a single sample on which the model behaves catastrophically. We are trying to infer, from facts about the learned algorithm itself, how often rare but unacceptable behaviors are likely to occur. To make this pipeline a reality, we need roughly the following ingredients: Wide-ranging mechanistic estimators , in the spirit of the Matching Sampling Principle . These take a description of a computation — e.g. the weights of a neural network — and estimate some property of its behavior (e.g. expected loss on a distribution) without relying on input-output samples. Tools for identifying structure as it is added to the weights and converting it into advice that improves the MSP estimators. A way to deal with real-world distributions (e.g. the distribution of inputs seen by ChatGPT rather than uniformly random bits), often defined only implicitly through a large number of points. Something to align to. We need some notion of what behavior we want to reward. The "type signature" here is a mathematically well-defined function (even a slow and impractical one) that takes in model outputs (or perhaps states of the world) and assigns them goodness scores. (Optional) Mechanistic Anomaly Detection. A tool for determining whether model outputs look good "for the right reasons." If these ingredients work as hoped, the resulting technology would in principle let us describe the algorithms inside a model as it is trained, flag deceptive alignment and reward hacking, and train against those flags to produce an aligned system while paying a manageable alignment tax . The plan is to treat "how often will the model cause catastrophe" as an estimation problem, build an adversarially robust estimator, and train the model until the estimate of its catastrophic behavior is acceptably small. Matching Sampling Principle The (average case) MSP states that for any architecture and degree of precision, there is a mechanistic estimator that at least matches the performance of sampling over random instances of that architecture. A lot of what we do is look at various quantities and architectures and think "what is the right way to estimate this," and chug along until we have something that (often) far out-performs sampling. At the time of writing, one of the crown jewels of this approach is an algorithm that takes in the weights of a multilayer perceptron , and outputs an estimate which approximates to within additive error averaged over assignments of , while running more efficiently than a Monte Carlo estimator. [2] This type of research is extremely parallelizable; it's also great for parcelling out to academics in different communities who each have their own function classes they understand well. We can ask one academic to extend our MLP work to transformers, another to think about Turing Machines, another to think about some weird thing that shows up in condensed matter physics. At this stage in ARC's development, we still learn new and important lessons every time we make an MSP estimator for a new architecture. We use the word "mechanistic", even though we don't have a clear definition of it. I'm going to be up-front that after reading this writeup, you will not have a complete sense of this concept, but I hope you'll have some idea of what it's pointing at. The definition ARC gives is usually something like "never assume the input-output behavior you haven't seen looks like the behavior you have seen, for any object." For instance: Never assume that because a model's loss was low on 100 random inputs, its average loss is low. Never assume that because the activation of neuron A in layer 6 is correlated with neuron B in layer 8 on 100 samples of the input distribution, they are correlated. Never assume that because the activation of neuron A in layer 6 is correlated with neuron B in layer 8 on 100 samples drawn from a mechanistically calculated representation of the activations on layer 4, they are correlated. Never assume that if you construct a ridiculously complicated object mechanistically from the model, and give that object 100 samples, the sample average equals that ridiculously complicated object's average behavior. Hopefully, this definition gives some sense of why mechanisticness might be useful for dealing with deceptive alignment. In some sense, not only are we never trusting the input-output behavior of the full model, we are never trusting the input-output behavior of components of the model, or ridiculously complicated objects mechanistically derived from the model. Here's another property I associate with mechanisticness, which I suspect is much weaker. To me, it's all about taking a big, terrifying object and breaking it down into pieces that can't conspire against you. One (extremely bad) approach you might take is to say "well a single neuron can't be deceptively aligned, so I'll analyze the model one neuron at a time, and then there won't be deceptive alignment." This approach would fail, of course, because deceptive alignment, like all cognition, doesn't live in any single component — it lives in how the components interact. So decomposition into simple pieces isn't enough on its own; we need a decomposition where the pieces also can't conspire. Each piece has to be both individually benign and unable to coordinate with others toward a bad outcome. A way to operationalize this is to require that each of the things you look at is quite simple, and also require that they are independent (for instance, an individual neuron's behavior given the preceding activations is simple, but is correlated both with its behavior on other inputs and with the behavior of other neurons). Many of our mechanistic algorithms involve expressing our final estimate as a combination of terms which are subjectively [3] uncorrelated with each other. This is something we can actually give concrete examples of. Suppose we want to estimate what fraction of numbers below have an even number of prime factors. Concretely, we want to estimate where the Liouville function is if has an even number of prime factors and if has an odd number of prime factors. One strategy here would be to sample inputs at random between and , calculate for each of them, and return . This technique assumes the inputs we haven't seen look like the ones we have seen. It is not mechanistic. A different strategy is to presume that is about half the time, and that the values for distinct values of are subjectively uncorrelated. [4] Since is half the time and half the time, we can start with an estimate of zero. There are a number of ways one could proceed here; one way is to select a subset of values of (perhaps the values from 1 to , or perhaps random numbers) and evaluate on those inputs. Since the value of at different points is subjectively uncorrelated, knowing that the value of is doesn't change our guess for other values of , it only moves our estimate for by . A valid mechanistic estimator of after we've evaluated at points is . Note that this only differs from the sampling estimator by a factor out front — instead of . I want to make several points here. This sort of estimator could have terrible performance. We made two heuristic assumptions [5] when we derived this, a mean-zero assumption and an independence assumption for . A 19th century mathematician wouldn't have been able to prove either of those facts, and even today we only know a proof for the first one. ARC doesn't necessarily claim that future mathematicians will eventually prove the independence claim. But we do argue that if the independence claim is false, there is a heuristic argument one could give as advice that could be incorporated into the estimator and make it work. For more discussion of this, see the next section. This sort of estimator can be randomized. I didn't specify how we chose the subset of points to analyze in our estimator. They could absolutely have been chosen randomly. Using randomness to determine what aspects of a system to analyze more carefully doesn't violate mechanisticness. The only violation would be to assume the things we haven't carefully examined yet look like the things we have. Although we've only had the Matching Sampling Principle for about a year, the idea is descended from much older concepts like heuristic estimation . Identifying Structure / Plugging Structure into MSP This is among the oldest parts of ARC's agenda: the idea that anytime you have a strange behavior you wouldn't predict heuristically (e.g. a finite number of twin primes, or a random reversible circuit evaluating to the identity, or a neural net being many standard deviations from our MSP estimate that works for random NNs [6] ), there must be some structural reason for it. This has been referred to as the No-Coincidence Principle . We certainly don't have a periodic table of every type of structure, a good way of encoding it, or a complete understanding of what to do when structure is pointed out. Our current best guess is that the best way to communicate structure is with certain types of compression, and ideas inspired by Kolmogorov Complexity and Sophistication. [7] It's a simple exercise to show that any time a model has any strange property, this is reflected in a lower Kolmogorov complexity. But a stronger pattern we've noticed is that when mechanistic estimators fail, some sort of resource-bounded complexity or sophistication should come into play. A nice example to think about is AlphaZero , a powerful neural net with no training set. In principle, a short computer program and a random seed are enough to tell you everything you need to know about AlphaZero (just encode the rules of some board games and the algorithm for reinforcement learning). Perhaps that short program, plus a longer computer program that takes less time to run, and a still longer program that takes even less time to run, et cetera, are the advice we need to point out the structure of AlphaZero. This would suggest that advice should take the form of a sequence of computation-bounded sophistication constructions. A different way of phrasing this strategy is an attempt to transform worst-case MSP instances into average-case ones. The whole notion of sophistication is that a very unusual string/set of weights can be thought of a typical string/set of weights drawn from some other distribution. As you can tell, this prong of ARC's agenda has a lot of half-baked thoughts. I blame this on the fact that we don't have a single non-trivial problem of this type we fully understand. Good toy examples on simple models of computation will be a huge win here, and continued failure will be a huge red flag. There is one other point I want to hammer home: the similarity between ARC's plan to watch the neural network's training process, and ARC's plan to create a short description of a process that would give rise to this neural network. Let me spell that out: most cryptographers believe in the existence of indistinguishability obfuscation (iO) — that it is possible to obfuscate any algorithm, including the algorithm embodied in a neural network, such that no polynomial-time observer can learn anything about the original algorithm beyond its input-output behavior by looking at the finished code. If a model's weights could be adversarially obfuscated in this way, standard mechanistic analysis would be hopeless in the worst case. ARC's plan for getting around this is to watch the training process, making note of new structures as they are being etched into the weights. We believe that training-process monitoring is necessary to create adversarially robust probes. [8] There's a connection here that we don't fully understand: on one side it may be that watching the creation of a neural network is necessary to have robust descriptions of their structure (and thus robust estimates of their behavior). On the other side, those same robust descriptions might take the form of (lists of) resource-bounded programs to create our very same neural network. Dealing with Real Data The MSP is usually written in terms of inputs coming from a well-defined and easy-to-sample distribution. But the distribution we care most about looks more like "plausible inputs for GPT-6." During training, we are given some number of examples from an unknown distribution, and we need to estimate the average over this distribution using less compute and more sample efficiency than just doing forward passes. For this to make sense, we probably need at least some prior over the parameters of the input distribution. But once we have that (e.g., the input distribution is a Gaussian with unknown mean selected uniformly over the reals and covariance selected according to such and such power law, or the input distribution is the result of pushing a Gaussian through a generative diffusion model of such and such a depth with parameters selected i.i.d. Gaussian) it seems we can often beat sampling at estimating given samples from . Mechanisticness is (even more) confusing in this context. There is no way to mechanistically determine how often people ask for homework help versus relationship advice; you need to measure it empirically. How does this square with our goal of avoiding sampling? The rough answer is that the parameters of the model have been heavily optimized by gradient descent to look good to us, and quite possibly to fool us. The parameters of nature haven't. [9] The model needs to be carefully picked apart for signs of danger, whereas the data-generating process can be safely understood using only black-box methods. So we are fine learning from samples of the data generating process, as long as our understanding of the model is mechanistic. The simplest version of this is straightforward. Suppose we want to calculate given samples of . Suppose that our prior over the distribution of tells that it is drawn from a 1-D Gaussian distribution with known variance but unknown mean. We can use the samples to infer the mean, and then mechanistically compute the expectation of over the inferred Gaussian. The expected squared error of this approach scales as . However, this more mechanistic process achieves a better constant in the scaling of the MSE than simply propagating all samples through and averaging — nothing deep is happening, the data is just being used to estimate the distribution parameters, and the rest reduces to a standard MSP problem. If the MSP is true, then in cases like this (and also ones far more complicated than this), we can beat sampling in terms of both compute- and sample-efficiency. ARC has spent very little time thinking concretely about these problems, but in the past few months it's gone from a problem we don't know how to approach to a series of bounded technical questions that we just haven't gotten around to. (Unfortunately, at the moment it requires too much background to be "parcelled out" to others. I don't think this is inherent, and think with some work we could carve off some nice modular chunks of this problem for our friends in academia.) Aligned to What? This is something we've spent very little time thinking about recently. People have been grappling for years (millennia?) with the fact that no specific, concise set of mathematically defined rules does a good job capturing morality. The current best plans involve a mix of deferring to your future self (corrigibility) and deferring to the model's guess about some idealized future self (indirect normativity). The version of indirect normativity I find most plausible isn't the science-fiction "committee of the thousand greatest minds" picture. It's more local: we want an AI that helps the user stay safe and learn about the real world based on their current preferences, and defers to the user's future self for hard questions where the current self doesn't yet know the answer. A lot of the technical complexity here comes from the need for decoupled feedback — we want the AI to take actions that genuinely make the world better according to the user's eventual judgment, rather than actions that influence the user into judging the world favorably. Distinguishing "make the world good" from "make the user think the world is good" is a real technical problem, not a philosophical aside, and it's the kind of thing ARC's tools would actually need to engage with. [10] I suspect that if ARC's tools work at all, they can train a model to do either of these, but we certainly haven't put in the legwork to figure out ways of doing that, and won't be able to test those techniques for years. I consider this to be the most neglected part of ARC's agenda. Unfortunately, thinking about this in a productive way requires a strong understanding of how ARC's methods would work and at least a decent understanding of the alignment landscape, and those two things don't coexist in many people at the moment. I'm hoping I can beef myself up enough to start seriously tackling it soon. Mechanistic Anomaly Detection The hope here is that if you understand why a model is getting a low loss, you can understand when it's getting a low loss "for the wrong reasons" on a particular input. In other words, we could detect anomalies where a low loss was produced by the wrong mechanism. Although that hope seems sensible enough phrased in English, nobody has managed to turn it into a mathematically well-defined conjecture, let alone provided serious evidence for it. If we got MAD working, it would be a cure for deceptive alignment and also for reward hacking. [11] However, I can't even begin to articulate how MAD would work, and I don't get the sense that anyone else at ARC can spell it out either. While many of my colleagues are still optimistic that such a thing is possible, it is mostly on the back burner while we work on understanding why models get low loss. Including direct risks like coups, or indirect risks such as sabotaging the training of a newer model, which may then perform a coup. ↩︎ The general strategy is something we call cumulant propagation. We know the distribution of the inputs, we use them to figure out the cumulants of the next layer, and then the layer after that, and so on through the model. For most values of , including more and more cumulants will give a more and more accurate estimate in a way that beats Monte Carlo estimation. ↩︎ What do we mean by "subjective" correlation? Informally, subjective probability is the best guess that a smart but still computationally bounded gambler would put on a certain empirical or logical fact. For instance, an observer might say there is a 52 percent chance the blues will win the next election, a 50 percent chance that the trillionth digit of is even, and zero correlation between the two. By contrast, there might be a positive subjective correlation between and in computer science (though reasonable people can disagree about what the subjective correlation is). ↩︎ The first of these is provably true using somewhat advanced techniques. The second is unproven, but would follow from the Riemann Hypothesis. ↩︎ In this writeup, I use 'heuristic' to mean 'a quantitative approximation made without necessarily having a rigorous justification for it'. The paradigmatic example of heuristic approximation is the presumption of independence . Other examples include things like 'treat any distribution we see as Gaussian' or 'treat every function as a low-degree polynomial'. We typically arrive at mechanistic estimations by chaining together many heuristic assumptions. ↩︎ Each of these is something which a quick heuristic derivation would say is unlikely. For the twin prime conjecture, the simplest heuristic argument comes from the Prime number theorem and the presumption of independence . For the random reversible circuit, we can use the conjecture that even fairly small such circuits are like random permutations of , meaning that (heuristically), there is only a chance that a given circuit would be the identity. For the MSP estimate, we are constructing it so it should be right on most neural networks. Notably, each of these heuristic arguments is extremely flawed, and could turn out to be wrong. ARC doesn't claim that every heuristic argument is correct all the time, we are saying that when they fail, there is a more sophisticated argument that would let us understand the failure. ↩︎ The Kolmogorov complexity of a string is the length of the shortest program that outputs on a fixed universal Turing machine. For instance, a string of all zeroes has very low Kolmogorov Complexity, whereas a random string will have Kolmogorov complexity equal to its length. K-complexity captures incompressibility: a string is "random" if . But Kolmogorov complexity alone does not capture "structure". A uniformly random string has near-maximal , yet we can describe everything important about it in one sentence: "it's random." This motivated the definition of sophistication . Informally, the sophistication of is the length of the shortest program that specifies a set containing , where is small enough that knowing it captures most of the structural information about , and the remaining bits needed to pin down within look random. (For a thorough treatment, see Antunes and Fortnow ; for an accessible discussion connecting sophistication to thermodynamics and dynamical systems, see Aaronson's blog posts on complextropy and Kolmogorov complexity .) Both and sophistication are uncomputable as stated, since they quantify over all programs with no runtime bound. They also don't always correspond to our intuitions about how complicated/structured a system is. Resource-bounded variants restrict to programs halting within some time bound . The -bounded Kolmogorov complexity is the length of the shortest program outputting in at most steps; one defines -bounded sophistication analogously. These variants are computable and connect to pseudorandomness: a string has low bounded complexity iff it can be efficiently generated, and low bounded sophistication iff its structure can be efficiently described. Aaronson's complextropy is essentially a resource-bounded sophistication measure. For a recent formal development of computation-bounded sophistication, see Finzi, Qiu, Jiang, Izmailov, Kolter, and Wilson . ↩︎ The rough idea is that any time the model learns a new fact, that fact is also incorporated into the advice string of our estimator. So if the model's evil plan involves some scientific fact that humans don't know but the model does, the estimate of catastrophe will catch it. This is true both for scientific facts (e.g. that these chemical compounds will make refrigerators work better but ruin the ozone layer) and for mathematical rules of thumb (e.g. when you're playing Go and the board looks like this, you should focus on moves that look like that). Arguments from cryptography suggest it would be completely infeasible to extract all these facts from the finished model, but it seems likely you can extract them all from the training process to create an "epistemically competitive" overseer. Some people ask whether we are essentially training a second model which we know to be inherently honest. I think this is very loosely correct (though training a transformer/neural network/some architecture that was chosen for performance with no particular thought towards anything like interpretability/alignment is a necessary step in the process, and for economic reasons what we will probably actually do is use the mechanistic model to align the neural network, and then deploy the neural network). ↩︎ Modulo fears of data poisoning. ↩︎ One reason MAD would be valuable, if we got it working, is exactly that it might let us distinguish "the universe looks good because it is good" from "the universe looks good because the cameras were hacked" — i.e., it pushes against the failure mode where the AI influences the measurement rather than the underlying reality. ↩︎ It might also help with finding a good thing to align to. We want the model to produce universes that look good because they are good, not because the cameras were hacked to show lots of smiling people. ↩︎ Discuss
Score: 29🌐 MovesJun 9, 2026https://www.alignmentforum.org/posts/M2tD23bvQLBqsEpqu/a-mike-s-eye-view-of-arc-s-research - Top 8 open source STT options for voice applications in 2026
Top open source STT options for voice applications
Score: 29🌐 MovesJun 9, 2026https://assemblyai.com/blog/top-open-source-stt-options-for-voice-applications - EY's AI leader says cost-cutting represents the 'smallest' ROI on AI for companies
EY's AI leader says cost-cutting represents the 'smallest' ROI on AI for companies Business Insider
Score: 29🌐 MovesJun 9, 2026https://www.businessinsider.com/ey-ai-leader-says-companies-too-focused-on-cost-cutting-2026-5 - Demostack Launches Demo Partner and Builder Agent, Bringing Agentic AI to Every Step of the Product Simulation Lifecycle
Demostack Launches Demo Partner and Builder Agent, Bringing Agentic AI to Every Step of the Product Simulation Lifecycle markets.businessinsider.com
- Floatboat Launches "Proactive Agent OS" That Works From Your Calendar
Floatboat, an AI startup backed by Sequoia, launches a “proactive agent operating system” that uses your calendar to automatically trigger work—meeting briefs, follow-ups, document gathering, and recurring workflows. Its built-in FloatIM interface treats AI agents like team members in a group chat, enabling multiple agents to collaborate autonomously. The platform supports 3,500+ apps, integrates with Lark and WeChat, and runs on models including DeepSeek and Kimi.
Score: 29🌐 MovesJun 9, 2026https://pandaily.com/floatboat-launches-proactive-agent-os-that-works-from-your-calendar - Why FinOps teams are turning to FOCUS for AI cost accountability
As artificial intelligence drives a new wave of token-based spend across cloud and on-premises environments, enterprises are discovering that understanding what they actually owe — and to whom — is harder than ever. The FinOps FOCUS specification is emerging as the open standard that provides practitioners with a shared data language to cut through multi-provider […] The post Why FinOps teams are turning to FOCUS for AI cost accountability appeared first on SiliconANGLE .
Score: 29🌐 MovesJun 9, 2026https://siliconangle.com/2026/06/08/focus-specification-ai-cost-accountability-finopsx/ - Has India lost the AI race? Not entirely
Also in this newsletter, what’s going on with Rajesh Exports?
- India GCCs Lead AI and Cloud Hiring as IT Services Trail
India’s Global Capability Centers are pulling ahead of IT services firms in AI and cloud hiring, as multinationals bring more sensitive, high-value technology work in-house. The post India GCCs Lead AI and Cloud Hiring as IT Services Trail appeared first on TechRepublic .
Score: 28🌐 MovesJun 9, 2026https://www.techrepublic.com/article/news-gcc-ai-cloud-hiring-apac-india/ - There is no AI boom without these workers. Meta just proved it.
There is no AI boom without these workers. Meta just proved it. Business Insider
Score: 28🌐 MovesJun 9, 2026https://www.businessinsider.com/meta-launches-construction-data-center-jobs-program-2026-6