Football matches, nuclear power generation, and hurricane prediction – AI keeps being an exciting topic to cover and talk about.
This edition also covers how AI models help students cheat and how China actually found a sturdy and effective way to tackle this challenge.
2 June 2025 Meta will launch AI ads on Facebook and Instagram
Meta will let advertisers use AI to create and target ads on Facebook and Instagram by 2026. The tool will generate images, video, text and targeting from an image and budget, bypassing advertisement agencies.
More can be found in The Guardian.
3 June 2025 Meta signs Nuclear Power deal to secure energy for AI
Meta signed a 20-year agreement with Constellation Energy to buy power from the Clinton nuclear plant in Illinois, starting when state subsidies end in 2027. The deal funds the plant’s relicensing and a 30 MW capacity increase, keeping one reactor in operation. It secures electricity for Meta’s AI and datacenters and could guide more tech firms toward similar arrangements.
More can be found in The Guardian.
4 June 2025 Reddit sues Anthropic for unauthorized content scraping
Reddit sued Anthropic of San Francisco, saying the company’s bots accessed Reddit over 100,000 times since July 2024 after pledging to stop in May 2024. Reddit argues the data use could be worth billions and labels it exploitation of site content. Anthropic disputes the claims and plans to defend itself. The case joins several copyright suits facing Anthropic and other AI companies.
More can be found in The Verge.
5 June 2025 Amazon tests humanoid robots in delivery
Amazon is writing software to allow humanoid robots to handle package delivery. It is testing them at a San Francisco “humanoid park,” where robots ride in a Rivian van and exit to drop off parcels while the driver serves another stop. Amazon supplies the control code and relies on external suppliers for robot hardware. Field trials on real routes will follow the indoor tests, expanding the work they began with warehouse robots and drones.
More can be found in The Guardian.
6 June 2025 Gemini can now schedule tasks
Google added “scheduled actions” to Gemini for AI Pro and AI Ultra subscribers. Users can set tasks for fixed times, such as a calendar recap each day or blog-idea prompts every Monday.
Gemini can also run single tasks, like summarizing an event the day after it ends. Tasks are listed in a “scheduled actions” page inside the app. The update positions Gemini as a more agent-like service, mirroring the reminder tool available in ChatGPT.
More can be found in The Verge.
9 June 2025 Chinese tech companies block AI during school exams
Chinese tech companies shut off some AI features during the four-day gaokao exams to curb cheating. More than 13.3 million students sit the tests, which decide university placement.
Apps from ByteDance, DeepSeek, Tencent, Alibaba and Moonshot halted image recognition and question answering services during exam hours, telling users the service was suspended. Authorities also employ AI surveillance, biometric checks, device screening and signal blockers, and cities adjust traffic and events to support exam logistics.
More can be found in The Guardian.
10 June 2025 Reddit takes Anthropic to court for unauthorized content scraping
Reddit sued Anthropic of California, alleging over 100,000 data requests after Anthropic said it stopped scraping. Reddit claims Anthropic bypassed robots.txt and terms, copying public and deleted posts to train Claude without a license. The platform notes it licenses data to OpenAI and Google and now seeks damages and a ban on Anthropic’s use. Anthropic denies wrongdoing.
More can be found in The Verge.
11 June 2025 Mistral builds European AI cloud
Mistral AI launched Mistral Compute with Nvidia, giving Europe a cloud option outside of US providers. It released two Magistral reasoning models, one open and one proprietary, that reveal their chain of thought in the user’s language. The models kept image reasoning and function calling after text-only training and answer queries within seconds. A new training system updates weights across GPUs quickly and is run on 18,000 Nvidia Grace Blackwell chips in France. The platform keeps data inside the EU and plans wider rollout on low-carbon power.
More can be found in VentureBeat.
12 June 2025 Deepmind improves hurricane prediction with AI
Google DeepMind and Google Research launched Weather Lab to share AI cyclone forecasts and assist the U.S. National Hurricane Center. The model predicts formation, possible tracks, intensity, size and shape up to 15 days beforehand, generating 50 scenarios. Tests show its 5-day track error is 140 km below ECMWF ENS and its intensity error beats NOAA HAFS. The training combines reanalysis data with a 45-year cyclone record, letting one system cover both predicted track and intensity. NHC forecasters now view these outputs with other models, and Weather Lab stores forecast archives for study while noting they are not active warnings.
More can be found on DeepMind’s blog.
15 June 2025 Thousands of UK students caught cheating with AI
Guardian data show nearly 7,000 confirmed AI-related cheating cases at UK universities in 2023-24, up from about 2,200 the previous year. That equals roughly 5.1 incidents per 1,000 students, with early 2024-25 figures indicating a rise to about 7.5. Traditional plagiarism cases fell sharply over the same period. Many universities still do not log AI misuse separately, and researchers say most cheating likely goes undetected. Studies suggest AI-generated work often evades detection tools, forcing universities to rethink assessment methods. Experts urge focus on skills beyond text production and warn blanket in-person exams are unrealistic.
More can be found in The Guardian.
17 June 2025 OpenAI wins $200 million contract with Department of Defense
The U.S. Department of Defense gave OpenAI a $200 million contract to build AI tools for administration, data handling, and cyber defense, with work due by July 2026. OpenAI says this is its first government partnership under a plan to supply agencies with models that follow its usage rules. The deal follows OpenAI’s 2024 link with Anduril and the firm’s removal of a ban on military use. OpenAI says the systems will speed health-care access for service members, improve program data review, and support cyber protection.
More can be found in The Verge.
19 June 2025 Midjourney releases its video model
Users can now animate images into five-second clips and extend them to 21 seconds. A prompt directs motion, with options for subject or camera movement. Access is on the web and Discord, priced at about one image credit per video second. Disney and Universal are suing Midjourney over alleged copyright infringement tied to the feature.
More can be found in The Verge.
20 June 2025 People adopt the AI language
Researchers report that after ChatGPT’s release, words it favors, such as “delve,” appear far more often in human speech. Speakers have begun to adopt this language without noticing, and their tone is becoming longer and more structured. Smart-reply studies show AI can boost positive wording but also triggers distrust when its use is suspected, weakening signals of authenticity and effort.
Models prefer Standard American English, reducing dialect diversity and erasing cues that mark vulnerability. Experts say communication may split between templated, AI-shaped forms and personal, messy expression, so active choices are needed to keep human quirks.
More can be found in The Verge.
23 June 2025 Salesforce introduces Agentforce 3
Salesforce released Agentforce 3, adding a Command Center that tracks AI agent activity.
The platform now supports the Model Context Protocol, so agents link to external tools through AgentExchange. Latency dropped 50 percent; reply streaming, model failover, and in-house Anthropic models serve regulated sectors. Firms such as PepsiCo report shorter case times and high autonomous resolution. More than 200 prebuilt actions speed deployment, and pricing scales with completed actions.
More can be found in official announcement.
24 June 2025 Verizon adopts Gemini to handle complicated customer service cases
Verizon added a Gemini-based assistant to the My Verizon app. The chatbot handles upgrades, new lines, billing questions and other tasks. If it cannot answer or act on a request, it connects the user to a human agent. Verizon trained small in-house models on its data and reports over 90 percent accuracy. Customer Champion teams will also use Gemini models and give 24/7 chat support.
More can be found in TheVerge.
24 June 2025 Anthropic wins copyright case
Judge William Alsup ruled that Anthropic’s training of AI models on digitized copies of books it bought is fair use. The ruling applies only to those physical books converted in-house. Alsup set a separate trial over Anthropic’s alleged use of pirated books and any related damages. He did not decide whether AI output infringes on copyrights.
More can be found in The Verge.
26 June 2025 Anthropic introduces building apps in Claude AI chatbot
Anthropic’s Claude chatbot now offers a beta tool that creates apps from user prompts. Users have built games, learning tools, data analysis apps, writing assistants and agent workflows. Apps call Claude through an API, and any shared app uses the viewer’s subscription credits. The feature is available on the Free, Pro and Max plans.
More can be found in The Verge.
27 June 2025 Google’s emissions rise with AI demand
Google’s carbon emissions have risen 51% since 2019, driven mainly by datacentre power for AI. Deployment of carbon-free energy sources such as small modular reactors is behind schedule. Google has signed contracts for 22 GW of clean energy and met its goal of plastic-free product packaging. The firm argues AI can still help others cut 1Gt of emissions each year by 2030.
More can be found in The Guardian.
30 June 2025 China autonomous robots football contest
Beijing hosted a three-a-side football contest between four teams of humanoid robots, billed as the first autonomous AI match. Videos showed robots falling, mis-kicking and needing stretcher help when they could not stand. Booster Robotics supplied the hardware; university teams wrote the control code. Tsinghua University’s squad beat China Agricultural University 5–3 in the final. Organizers say sport is a good test bed and hope robots can one day play safely with humans.
