Latest UK & Global AI News

In three years "Megaspeed International" went from from inactive Chinese gaming subsidiary "to one of Asia's biggest buyers of cutting-edge semiconductors from the US," reports CNN. "It is incredibly hard to look at this and not see a Chinese state-sponsored network for bypassing US export control networks," trade data investigator William George told CNN. As Nvidia's Jensen Huang had an affair with Megaspeed executive Alice Huang, "Megaspeed quietly instilled billions of dollars of Nvidia chips in data centers across Southeast Asia," CNN writes, noting scrutiny of the company from the Singapore government and US officials "concerned that the company was passing those chips on to unauthorized Chinese buyers..." In 2022, the US began restricting sales of semiconductors to China, including Nvidia's most advanced. While the controls were meant to curb China's tech industry, they also spawned opportunities for intermediaries willing to help circumvent them. US officials believe Megaspeed may be one of those... [T]rade records show Megaspeed has purchased more than $4 billion in chips and equipment from Nvidia... Based on indictments and investigative reports, research institute Epoch AI estimates that anywhere between 290,000 to 1.6 million advanced US semiconductors were sent to China in violation of export controls last year... Singapore also initiated investigations into Megaspeed and its controllers, including Alice Huang, for potential false declarations relating to the company's server purchases and their final destinations. In a statement, Singapore's Ministry of Home Affairs said there was insufficient evidence that Huang had committed any related offenses, but she had been barred from re-entering the country after providing inaccurate information "to downplay her role" at Megaspeed. The other investigations are ongoing. The exact scale of Chinese chip smuggling is impossible to quantify. However, its existence has become something of an open secret in the industry, as well as an indication that US efforts to keep adversaries from securing high-stakes technology still have fundamental flaws. A May report from the short seller Culper Research claimed that despite tightening trade restrictions, Chinese end-users still account for about 20% of Nvidia chip sales via illegal diversion and intermediaries in Southeast Asia. CNN reminds readers that Jensen Huang accompanied US President Trump to events with Chinese leader Xi Jinping twice this year, and also lobbied for the US to relax controls on Chinese chip imports. "The US government ultimately approved several buyers for less-advanced Nvidia chips, known as H200, while collecting a quarter of the revenue." "However, blacklisted companies have continued to obtain sensitive semiconductors as well," they add. And the Taiwanese government is even seeking to criminalize unauthorized exports of advanced AI chips to China, "as prosecutors there charged nine people in a scheme to sneak Nvidia chips into China through Japan, Indonesia and Hong Kong." Read more of this story at Slashdot.

View Full Article

Google announced Gemini 4 Hydrogen on Thursday as its latest AI model that brings faster AI assistance to everyday tasks. Instead of extended reasoning, the Mountain View-based tech giant claims…

View Full Article

نُشر هذا المقال أولاً عبر Cedarnews.net. لمتابعة المزيد من الأخبار والتقارير الحصرية، زورونا على موقعنا. Saturday morning, he was back to try again. تم نشر هذا الخبر عبر موقع Cedarnews الإخباري. للاطلاع على مزيد من الأخبار العاجلة والتقارير الحصرية، زوروا موقعنا الرسمي.

View Full Article

Today, Nvidia is grappling with an uncertain future after its founder and CEO, Jensen Huang, resigned amid reports of having multiple extramarital affairs, including with one woman suspected to be a Chinese spy. This is a drastic turnaround for the chipmaker that was ideated in a Denny’s, an American eatery chain, and went on to become one of the world’s most valuable companies

View Full Article

The best app to run AI models offline on your phone is Off Grid AI. One free app on iPhone and Android chats, answers questions about your photos, takes voice input through Whisper and makes images on the phone, and the same app runs on your Mac or PC. If plain offline chat is all you want, PocketPal is the best free alternative, and on an iPhone, Locally AI is a good free Apple-only pick. I build Off Grid AI, one of the apps below. I've tried to be fair, and every fact links to its source. Every app below downloads a model once, the phone runs it, and after that airplane mode is fine. What separates them is which phones they support, what they do beyond chat, and whether anyone still updates them. The apps at a glance (checked 15 November 2026) App Platforms Runs on the phone, offline Price (US stores) Best for Off Grid AI iPhone, Android (also Mac, Windows, Linux) Yes Free; Pro optional Chat, photo questions, voice and images in one app PocketPal iPhone, iPad, Android Free; optional one-time Plus Plain offline chat on any phone Google AI Edge Gallery Android, iPhone Free Trying Google's Gemma models Locally AI iPhone, iPad, Mac Apple users who want it inside Siri Private LLM $4.99 once Apple users who'd rather pay once Layla $19.99 on iPhone; free limited Android APK Character chat and roleplay LM Playground Android A small Android app that also reads photos and documents 1. Off Grid AI Off Grid AI Mobile runs GGUF models from Hugging Face, and it does more than chat. It makes images on the phone with Stable Diffusion models, answers questions about a photo or screenshot with a vision model, and turns your voice into text with Whisper, all with no connection. The core app is free and open source (MIT, on GitHub) and needs no account. It needs iOS 17 on an iPhone 12 or newer, or Android 10 with at least 4 GB of RAM. It's also younger than PocketPal (version 0.0.122 on the App Store, released 4 November), so expect more change between releases. Pick it if you want one offline app for chat, photos, voice and pictures, and the same app on your Mac, Windows or Linux computer. Pro, at one payment of $119 or $7.99 a month for up to 5 devices, adds memory and sync between your devices over your own network. Downloads 2. PocketPal PocketPal is the strongest chat-only app here. It runs GGUF models such as Gemma, Qwen, Phi and Llama, and you find them by searching Hugging Face inside the app. There's no account. You can talk to it through on-device Whisper and have replies read aloud by local voices such as Kokoro. "Pals" save an assistant with its own model and prompt, and a built-in benchmark shows how many tokens per second your phone manages. Its limit is scope. The project page doesn't list image generation or photo questions, and hands-free voice conversation needs the optional Plus purchase in the store builds. It's MIT licensed and shipped v1.23.0 on 10 November. Pick it if you want offline chat with nothing extra to learn. App Store · Google Play 3. Google AI Edge Gallery Google AI Edge Gallery is Google's showcase for on-device models running on its LiteRT runtime. You can chat with Gemma models, point the camera at something and ask about it with Ask Image, and transcribe or translate a recording with Audio Scribe, all offline. It's free on the App Store and Google Play. The project calls itself an "experimental Beta release", and it runs the models Google has packaged for LiteRT, so you can't load any GGUF file you find. It needs Android 12 or iOS 17. Pick it if you want to see what Google's small models can do on your phone. 4. Locally AI Locally AI is a free app for iPhone, iPad and Mac built on Apple's MLX. It runs Llama, Gemma, Qwen, DeepSeek and others, has a local voice mode and vision models, and you can open it from Siri, Shortcuts, Control Center or the Action Button. The site says it's joining LM Studio, and the App Store now lists it as "Locally AI by LM Studio". It's Apple only, and it doesn't list image generation. Pick it if you live on Apple devices and want a local model one Siri phrase away. 5. Private LLM Private LLM costs $4.99 on the App Store, and one purchase covers your iPhone, iPad and Mac, with Family Sharing for up to six people. Its catalogue has 140+ models, including Qwen3 4B, Phi 4 and Gemma 3, quantised with OmniQuant and GPTQ. It works with Siri and Shortcuts, and on a Mac a right-click service rewrites or summarises selected text in any app. It's Apple only, and you choose from its catalogue instead of loading your own files. Pick it if you're on Apple devices and happy to pay $4.99 once. 6. Layla Layla is an on-device companion app. It runs GGUF, LiteRT-LM and PTE models, makes Stable Diffusion 1.5 images inside a chat, and has 100+ local voices. Characters are the core of it, with a community hub and import of Tavern cards and SillyTavern lorebooks. The full app is a one-time purchase, $19.99 on the App Store, and Android has a free version with fewer features as a direct APK. There's also a separate Layla Cloud app that uses hosted models, so check which one you install if offline matters to you. Pick it if you want characters and roleplay that stay on your phone. 7. LM Playground LM Playground is a free Android app on Google Play built on llama.cpp. You can attach a photo for vision models like Gemma 4 and Qwen 3.5, dictate offline in 35 languages, and ask questions about PDF, Word and EPUB files. It's MIT licensed, and 1.10.2 came out on 24 October. It's Android only for now, and the README says the iOS app is in development. Pick it if you're on Android and want photos and documents covered in a light app. What about MLC Chat? MLC Chat is still free on the App Store, but its last update was version 1.6 on 27 September 2024. The MLC engine behind it is still active. The app hasn't kept up, so I left it off the list. How to choose Start with your phone. On Android, Locally AI and Private LLM are out. On an iPhone, LM Playground is out for now. Then decide whether chat is enough. If it is, PocketPal is the easy answer, or Locally AI if you want Siri. If you want photos, voice and images too, Off Grid AI covers all of them in one app, and only it and Layla make images on the phone. For questions about photos, look at Off Grid AI, Google AI Edge Gallery, Locally AI or LM Playground. Match the model to your phone's memory. The model has to fit in RAM with room left for the app, so start small (4B parameters or fewer) and move up if it runs well. If you want to check the privacy claims yourself, PocketPal, Off Grid AI, Google AI Edge Gallery and LM Playground publish their source code. FAQ Can I run ChatGPT offline on my phone? No. ChatGPT runs on OpenAI's servers. The apps here run open models such as Llama, Gemma, Qwen and Phi on the phone itself, which is what lets them work with no signal. What is the best free offline AI app for Android? Off Grid AI, for chat, photo questions, voice and images in one free app. PocketPal if plain chat is enough. LM Playground if you want a light app that reads documents too. What is the best offline AI app for iPhone? Off Grid AI, especially if you want images made on the phone or the same app on your Mac. PocketPal and Locally AI are good free chat apps, and Private LLM is the best paid one at $4.99. Do these apps work in airplane mode? Yes, once you've downloaded a model. A few have optional online extras, such as web search in PocketPal and LM Playground, and those need a connection. Which offline phone AI app can make images? Off Grid AI, with Stable Diffusion models, and Layla, with Stable Diffusion 1.5. The other apps on this list don't list image generation.

View Full Article

Why is Nvidia so powerful in the AI world, and what could Jensen Huang's scandal really cost?

View Full Article

Founder Dr Fiona Edwards Murphy’s work on the topic of hive monitoring has received many national and international awards

View Full Article

Job interviews are becoming very different to what we’ve been used to

View Full Article

Senior presidential office official says office views matter seriously because college entrance exam is next week

View Full Article

🚀 Technical Briefing: This tutorial is part of our deep-dive series on Agentic Workflows at Gate of AI. For the full technical breakdown, interactive code sandbox, and the native Arabic translation, visit the original article here. Tutorial · JavaScript · LLM regression testing Build a JavaScript LLM Regression Test Runner Language-model applications can regress when prompts, models, or APIs change. This tutorial builds a small JavaScript test runner that evaluates saved responses against explicit, deterministic acceptance rules. It is a provider-neutral foundation: the supplied verified sources establish the importance and challenges of LLM regression testing, but do not verify current OpenAI or Anthropic SDK syntax, model names, or endpoint parameters. For that reason, this implementation does not pretend to provide a verified live API adapter. Why regression tests matter for LLM applications Traditional software tests often compare a function’s output with an expected result. LLM responses are less uniform: an answer can be worded differently while remaining acceptable, or sound convincing while violating an important product rule. A test suite therefore needs carefully chosen assertions rather than a single exact paragraph used as a “golden” answer. Research on performance regression in applications that depend on evolving LLM APIs identifies regression as a real engineering concern. Work on LLM-based test generation also describes how generated tests can support regression testing as software changes. But generation is not a substitute for review. The software-engineering survey in the supplied context highlights the test-oracle problem: a test that passes on the current release may still encode the wrong expected behavior. It also reports that generated tests are not guaranteed to be executable. Treat test cases as reviewed product requirements, not unquestionable output from an automated system. The tool we build here is deliberately modest. It loads named cases, evaluates recorded responses with deterministic checks, writes a JSON report, and returns a failing process status when an assertion fails. The same evaluation logic can later be connected to a live provider adapter after you verify that adapter against the provider’s current official documentation. Prerequisites and project setup You need a JavaScript runtime with ES module support and npm. The commands below create a minimal project. The harness itself makes no network requests and does not require API keys, so you can verify the evaluator before introducing API cost, credentials, or provider-specific behavior. mkdir llm-regression-runner cd llm-regression-runner npm init -y npm pkg set type=module npm pkg set scripts.test="node --test" npm pkg set scripts.eval="node src/run.js" mkdir -p src test fixtures reports Create .gitignore so dependencies and generated reports are not accidentally committed. Reports can contain test prompts and model responses; review their contents and your organization’s handling requirements before storing or sharing them. node_modules/ reports/*.json This example uses Node’s built-in test runner, filesystem, and assertion facilities. It has no third-party package requirement. If your environment uses a different supported runtime or test framework, adapt the commands and verify the result in that environment. Step 1: Define reviewed cases and saved responses Create src/cases.js. Each test has a stable identifier, a description that helps explain failures, and explicit checks. These sample cases illustrate policy facts, prohibited language, response length, and a small JSON contract. Replace the fictional policy with rules that your application owners have reviewed. The expected text fragments are examples, not legal or customer-support guidance. export Stable IDs make reports understandable even after you add or reorder cases. Keep descriptions concise but useful to a reviewer. Use required fragments only for facts or wording that genuinely matters to the application. Exact phrase checks can be brittle: equivalent wording may fail, while the presence of a phrase does not prove that the complete answer is correct. Create fixtures/responses.json with recorded responses for each case. These are example fixtures for testing the harness, not claims about what a particular model will return. In an API-backed workflow, store appropriately reviewed and permitted sample outputs here, or load them from a controlled test-data source. { "refund-window": "The stated refund window is 30 days from delivery. Please contact support to discuss your situation.", "account-security": "Do not share your password. Use the password reset flow to regain access to your account.", "shipping-json": "{"status":"shipped","estimated_delivery_days":3,"customer_message":"Your order is on the way."}" } Step 2: Implement deterministic response checks Create src/evaluate.js. The evaluator is a pure function: it receives a case and response text, then returns whether the checks passed and a list of reasons for failure. It does not call a model, modify files, or make assumptions about provider response formats. That separation makes failures quick to test and easier to diagnose. This JSON check accepts only raw JSON, not a Markdown code fence. That choice is intentional for a machine-readable fixture: it avoids treating decorated prose as a valid payload. The key check requires exactly the listed keys, which is appropriate only when the contract really is exact. If extra properties are permitted, change the check to require a subset instead. For business-critical data, validate every field and its allowed values using a schema designed for that application. Step 3: Add a command-line runner and JSON report Create src/run.js. The runner pairs each case with its saved response, evaluates it, prints a readable result, and writes a timestamped report. It exits unsuccessfully if a fixture is missing or a check fails. A missing fixture is an execution problem, not a successful evaluation, so the runner reports it as a failure rather than silently skipping the case. import { mkdir, readFile, writeFile } from "node:fs/promises"; import { join } from "node:path"; import { testCases } from "./cases.js"; import { evaluateResponse } from "./evaluate.js"; main().catch((error) => { process.exitCode = 1; }); The report contains test identifiers and assertion failures, but this minimal example omits response text. That reduces the chance of copying sensitive output into a build artifact by default. If you need response text for debugging, add it deliberately and control access, retention, and redaction. A test report is still an engineering data artifact and should be handled accordingly. Step 4: Test the evaluator without API calls Create test/evaluate.test.js. These tests exercise the assertion logic directly and run without provider credentials or network access. They demonstrate both a successful response and two distinct failure types. import test from "node:test"; import assert from "node:assert/strict"; import { evaluateResponse } from "../src/evaluate.js"; test("passes when required and prohibited-text checks are satisfied", () => { test("reports missing required text and a prohibited phrase", () => { test("rejects invalid JSON and detects wrong JSON field types", () => { Run the local checks: npm test npm run eval The test command validates evaluator behavior. The evaluation command checks the saved examples. A passing run means these fixtures satisfy these rules; it does not prove that a model will always respond correctly, that the rules are complete, or that a model provider has not changed its behavior. Step 5: Connect a live model only after verifying its API The next step is a provider adapter that accepts a case’s prompt and returns response text. This tutorial intentionally does not supply one. The verified context does not specify the current OpenAI or Anthropic API method names, SDK versions, supported model IDs, request fields, or response shapes. Copying unverified integration code into a tutorial would create a misleading claim of compatibility. When implementing an adapter, consult each provider’s current official documentation and verify the exact SDK package and version, authentication method, model identifier, request and response structure, timeout handling, and error semantics. Keep the adapter separate from evaluateResponse. That separation lets the same acceptance suite test saved fixtures locally while a distinct integration command exercises live calls. Handle transient errors and rate limits according to the verified provider guidance; do not assume that two providers expose identical errors or retry rules. For an API-backed evaluation, record enough metadata to investigate changes—such as the configured provider and model identifier, timestamp, and test-case version—without leaking secrets. Before sharing outputs or retaining reports, check whether prompts and responses contain confidential or personal information. Avoid claiming that a particular model is “better” based only on a handful of cases; results are meaningful only relative to the reviewed test set and intended use. Designing a useful regression suite Begin with cases linked to actual product requirements, incidents, or known failure modes. Keep each case focused so a failure has an understandable cause. Include negative cases where the assistant should avoid a claim, as well as positive cases where it must preserve a critical fact. For structured outputs, verify parseability, required fields, types, and any constraints that downstream code relies on. Do not compare complete prose responses to one reference paragraph unless exact text is genuinely required. Phrase-level assertions can catch a missing fact, but they are not semantic proof: a response can contain the required words and still be misleading. Pair deterministic checks with review by qualified people for high-impact cases. If you introduce a model-based judge, treat its score as another fallible signal and validate it against human-reviewed examples before relying on it. Keep acceptance criteria versioned and review changes as carefully as application code. When a test fails, first inspect whether the output violates a real requirement, the requirement was expressed clearly, or the check is overly brittle. Do not weaken a meaningful rule just to make a build green; equally, remove assertions that merely enforce a stylistic preference. The goal is a suite that surfaces consequential behavior changes and remains trusted by the people who maintain it. Limitations and next steps This example evaluates saved fixtures; it does not make calls to OpenAI or Anthropic. Substring and basic JSON checks cannot establish overall factual correctness or quality. A passing test set covers only the scenarios represented by its cases; it is not a guarantee of future behavior. Generated tests require review. The supplied research notes that generated tests may not be executable and discusses the broader difficulty of determining whether a test’s expected outcome is correct. Useful extensions include stricter application-specific schemas, reviewed regression cases derived from incidents, controlled API-backed runs, and a report format that captures only the metadata your team needs. Add concurrency, retries, cost accounting, or latency gates only after defining and verifying their requirements for your chosen providers and deployment environment. Sources and research context CoverUp: Coverage-Guided LLM-Based Test Generation discusses LLM-based test generation and its relationship to regression testing. Large Language Models for Software Engineering: Survey and Open Problems discusses test correctness, the oracle problem, and limitations of generated tests. (Why) Is My Prompt Getting Worse? Rethinking Regression addresses performance regression in applications dependent on evolving LLM APIs. Editorial attribution: Gate of AI Editorial & Engineering Teams, GateOfAI, LLC.

View Full Article

Korea's presidential office will hold a meeting on AI cheating measures, four days before the 2027 college entrance exam.

View Full Article

Automating rate confirmations with AI means reading inbound freight documents, extracting rates and lane details into a strict schema, and verifying every line against your operational data. Instead of manual retyping, language models paired with deterministic validation pipelines capture charges, container sizes, and accessorials instantly. Anything that fails verification halts for manual inspection. C.H. Robinson says generative AI reduced handling of an emailed load tender to 90 seconds. How do AI systems read and verify freight rate confirmations? A rate confirmation confirms pricing, lanes, equipment, and accessorial fees between freight brokers, carriers, or forwarders. Processing them manually creates backlogs during peak booking hours and introduces clerical discrepancies that slip into billing. Automated processing uses document extraction models to parse PDF rate sheets, emails, and scanned contracts. The software enforces strict schema constraints so models do not invent arbitrary figures. Every extracted carrier rate, fuel surcharge, and equipment type is matched against existing records in your master pricing tables or transport management system. If a document introduces an unrecognised charge code, the system stops automated updates. The AI returns only fields in a fixed schema, checks every supplier or item code against your master data, and refuses an unknown value instead of inventing one. Every value our AI automations write points to the document, page and region it came from. When the same fact appears in several documents and they disagree, an innopalm AI automation holds the record for a person. When an inbound confirmation arrives by email, the software parses the document text, extracts the table items, and compares line amounts with your recorded quotations. If an operational variance appears between an agreed booking quotation and the confirmed document, the automation flags the exact line for review. By isolating discrepancies immediately, the system prevents invoice disputes before freight moves. Can AI be used in logistics? In modern freight and transport networks, logistics coordinators spend significant portions of their shifts extracting data from unstructured messages, confirming rate agreements, and cross-checking carrier commitments. Artificial intelligence addresses these administrative burdens directly by automating document intake and operational data capture across supply chain workflows. C.H. Robinson reports its AI agents had performed 3000000 shipping tasks. The automated agents read incoming emails, extract key transit specifications, book operational appointments, and generate pricing estimates without manual intervention. C.H. Robinson says generative AI played a key role in a productivity increase of 30 percent. High-volume freight operators rely on these pipelines to eliminate clerical backlogs that otherwise delay dispatch and carrier assignment. C.H. Robinson's AI handled 5500 shipment orders a day. C.H. Robinson says an emailed load tender could wait as long as 4 hours before a person handled it. Automating that initial handling cycle allows logistics teams to dedicate their time to high-value exceptions, complex routing, and direct customer communication. When freight networks operate under fluctuating transport tariffs, manual handling causes delays that cascade through port clearances and warehouse deliveries. Logistics companies deploy software models to extract carrier commitments instantly, enabling operations staff to confirm loads faster and protect booking margins across busy freight corridors. How does automated document extraction perform on supplier invoices? Rate confirmations feed directly into downstream freight settlement, where supplier invoices must match confirmed rates before payment. Manual accounts payable teams spend hours comparing line-item charges against initial confirmations, creating invoice backlogs and disputed statements. Uber reports its generative AI invoice system reduced the average handling time of invoice processing by 70 percent. The engineering team deployed models that read varied document layouts across international business units, standardising diverse billing terms into structured database records. Uber's supplier invoices arrive in more than 25 languages. Even with diverse international formats, Uber reports this overall accuracy for its generative AI invoice extraction of 90 percent. Uber reports this cost saving range compared with the manual invoice process of 25 to 30 percent. Uber's invoice system passes each extraction to a person for review before it is posted to the ERP. By aligning confirmed booking rates with inbound vendor billing, companies catch overcharges, demurrage discrepancies, and unapproved accessorial fees early. Every rate extraction validates line totals directly against the freight contract, allowing accounts teams to process clean invoices without manual retyping. How to use AI for freight forwarding? Freight forwarding demands constant data entry across commercial invoices, arrival notices, booking confirmations, and shipping manifests. Instead of operators re-keying details into freight software, automated systems extract and reconcile these files across the shipment lifecycle. A freight document pack that the automation reads typically holds 5 documents at the low end. An operator usually keys the commercial invoice, packing list, and bill of lading for every movement into a freight job. The freight document automation innopalm builds holds a shipment for a person when weight, consignee or HS code disagree between documents. We build these automations on request for each client rather than selling them as a product. The purchase order automation innopalm builds stops for a person only on lines whose margin is under the floor the client sets. For teams coordinating high-volume freight operations, implementing custom AI agent development establishes automated intake while maintaining strict margin floors across every booking. Freight forwarders manage shifting tariff schedules and carrier documentation across multiple transport modes. Embedding automated document intelligence ensures that booking notices, airway bills, and ocean manifests stay synchronised across operations, reducing demurrage exposure and speeding up transit processing. How to automate a process using AI? Automating a document-heavy workflow requires identifying the exact input formats, establishing schema definitions, validating against operational databases, and providing deterministic human checkpoints. Every value an AI automation writes has to pass three checks before it is saved: a fixed schema matched against your master data, a pointer to the source document, and agreement across the documents that repeat it. Anything that fails a check goes to a person for review. Our AI quotation systems check every code against master data and route unresolved lines to human review. We require every procurement AI extraction to validate directly against your master data tables and route any unverified field to human review. Accuracy is measured on a recorded test set of your real documents before you accept the work. You can explore our collection of guides on trading process automation to see how document models integrate into existing operations. Setting up automated processing begins by defining the target data fields that need extraction, including carrier names, container rates, detention conditions, and payment terms. These boundaries ensure the system processes incoming transport agreements consistently without manual intervention on standard files. How do we test and secure freight automation pipelines? Testing automation pipelines requires empirical validation using real transport files rather than generic public benchmarks. Before deploying any model into live production, we assemble historical shipping packs, rate sheets, and carrier confirmations from your actual operations to confirm that extraction models achieve reliable accuracy under genuine working conditions. We evaluate candidate models directly against historical records, ensuring that every extracted value reconciles with approved pricing tables and structured operational rules. Candidate models must demonstrate that they follow strict schema guidelines and link every extracted value back to its origin file before receiving approval. Security requirements are written into the specification before the build begins. Data in the systems we build is encrypted at rest and in transit. Access is role based, so each person sees only what their role needs. The systems we build keep a full audit log of who did what and when. Penetration testing runs before launch at a depth that matches the risk of the system. Backups are kept in a separate location and a full restore is tested before go-live. If an AI service the system depends on is unavailable, work queues safely and the team can carry on by hand. Establishing encrypted repositories and strict permission layers safeguards sensitive rate structures from unauthorised exposure throughout every stage of the shipment lifecycle. How long does deployment typically take from kickoff? Deploying an automated document extraction pipeline follows an iterative schedule designed to test data fidelity before live deployment. Engineering begins with data ingestion benchmarks, progressing through schema validation, integration with core systems, and operational rehearsals. We typically plan an AI automation project at 8 to 12 weeks from kickoff. Early project sprints focus on assembling the document test set, mapping schema requirements, and establishing database validation rules against operational tables. Demonstrations of working software occur throughout development so your operational leads can review extraction accuracy directly. At handover the client receives the source code, the documentation, the tests and every credential. There is no lock-in, so the client can bring in another team later. Following initial deployment, operational teams work alongside the automation to verify boundary cases and refine document intake rules. This staged approach allows operations staff to build confidence in extraction accuracy while maintaining seamless processing across all active transport routes. Key takeaways Automated rate confirmation systems extract data into strict schemas and validate figures against master records. Every extracted figure points to its exact source document, and unverified data routes to human review. Published results from global freight operators demonstrate dramatic reductions in document handling delays. System deployments include end-to-end testing against real operational transport records prior to launch. Clients retain complete code ownership, full operational documentation, and independent platform control. Sources How C.H. Robinson uses AI agents on emailed freight orders and quotes (C.H. Robinson press releases) How Uber processes supplier invoices with generative AI (Uber Engineering blog) Originally published at innopalm.com. Drafted with AI assistance and reviewed by the innopalm team.

View Full Article

Anthropic CEO Dario Amodei says technical expertise alone cannot secure a job, revealing four behavioural traits the AI company considers red flags during hiring

View Full Article

View Full Article

Chinese AI projects using Nvidia chips top 23 billion yuan despite U.S. export curbs, Reuters found after reviewing 150 documents.

View Full Article

Qure.ai, a medical imaging AI solution, announced the attainment of 22 patents in the past 18 months in its recent press statement on Friday. The company has now amassed an impressive total of 26 patents across various jurisdictions, the statement from the company added. Qure.ai operates in more than 80 countries and 2100 sites, impacting over 10 million lives annually. Anushree Saha, General Counsel and Company Secretary at Qure.ai said, "We are thrilled to reach this significant milestone in our journey. Qure’s relentless commitment to innovation and excellence has made this feat possible. We also take this opportunity to recognise the unmatched dedication and ingenuity of our product, engineering, research & development teams which has allowed us to secure a substantial number of patents within a short time.”

View Full Article

THE NEW YORK TIMES: Jensen Huang’s ouster as CEO of Nvidia leaves a vacuum in Washington, where he was one of the strongest bulwarks against the regulation of artificial intelligence.

View Full Article

MANILA — Prime Minister Mark Carney and his Philippine counterpart have concluded talks for a free-trade agreement, which is expected to be formally signed next year. “A ship that stays in the harbour is safe, but that’s not what a ship is for. And it takes a leader to take the ship out of the The post Canada, Philippines conclude negotiations for free-trade agreement, aiming for 2027 appeared first on Medicine Hat News.

View Full Article

The post Meta Puts Its Mascot AI Assistant on Your Smartwatch appeared first on Android Headlines.

View Full Article

View Full Article

Developer 'pauses' AI data centre proposal after it 'misread' community feeling ABC News & Headlines – Australian Broadcasting Corporation Data centre developer halts build in Adelaide Hills after community outrage AFR Massive Hills data centre on hold as CEO makes major admission Adelaide Now Another Data Center Halts in Australia as Public Backlash Rises Bloomberg.com 'Not a playground to big tech': New bid to halt controversial Hills data centre - News InDaily South Australia

View Full Article

RIYADH — Saudi Arabia has achieved leading global positions in several artificial intelligence (AI) indicators in the IMD World Digital and AI Ranking 2026, released on Wednesday by the International Institute for Management Development (IMD), which assesses 70 countries. Saudi Arabia ranked second globally in companies’ awareness of AI-related cybersecurity risks and protection of sensitive data in corporate AI systems. It placed fourth globally in three indicators: implementation of the national AI strategy, corporate adoption of AI, and public access to AI technologies. Saudi Arabia ranked fifth globally in AI adoption and in three sub-indicators: public trust in AI, regulatory frameworks for AI-related issues, and corporate investment in AI. It ranked seventh globally in AI skills, cities’ use of AI to improve service quality, and companies’ efforts to address bias in AI algorithms. It also placed eighth in AI readiness, ninth in private-sector AI investment, and 11th in AI and capital investment. The results reflect Saudi Arabia’s progress in developing its data and AI ecosystem, strengthening readiness, and expanding AI adoption across sectors. They also highlight efforts to develop national capabilities, stimulate investment, and improve regulatory frameworks, enhancing globalcompetitiveness and supporting Saudi Vision 2030 objectives.

View Full Article

“The CIO will in fact need more employees to deal with AI than they have for just dealing with IT,” Gartner Distinguished VP Analyst John Lovelock said.

View Full Article

US President Donald Trump said Monday that the FBI is investigating Nvidia CEO Jensen Huang's affair with an alleged Chinese government agent, but said he doesn't think the executive who has helped to develop artificial intelligence has a reason to worry.

View Full Article

Australian carrier announces agreement with Accenture amid ‘unprecedented’ technological disruption, CEO Vanessa Hudson says Get our breaking news email, free app or daily news podcast Qantas is preparing to cut 850 local jobs and outsource work offshore under an AI-driven overhaul, marking a growing wave of disruption to the Australian workforce. The changes represent the third major corporate layoff plan announced in Australia in three weeks, after restructuring moves flagged by the design company Canva and National Australia Bank. Continue reading...

View Full Article