Albania appoints a "corruption-proof" AI administer; Claude's memory expands
Get bigger weekly updates! Free subscribers receive the top stories each week, while Paid subscribers will get a few extra stories. All support for Handy AI directly helps me maintain the newsletter and keep the information flowing.
last week’s top stories
🤖 Albania appoints an AI ‘minister’ to curb corruption. In a world first, Albania’s Prime Minister Edi Rama introduced an AI-powered virtual “minister” named Diella to oversee public procurement. The AI bot will autonomously manage all government tenders with the goal of eliminating bribes and favoritism, essentially acting as a corruption-proof administrator. Diella originated as a digital assistant on Albania’s e-government platform and now joins the cabinet virtually. Read more
🧠 Claude AI gets a long-term memory upgrade. Anthropic rolled out a persistent “Memory” feature for its Claude AI assistant, allowing it to remember a team's past conversations, projects and preferences across chats. Available to business users, the opt-in memory means you don’t have to re-explain context, you can even ask Claude questions like “What were we working on last week?” and get an answer drawn from prior discussions. Users have full control over what Claude retains, with settings to edit or delete the stored summary and an Incognito mode for chats that shouldn’t be saved. Read more
⚖️ Rolling Stone owner sues Google over AI-generated summaries. Penske Media, publisher of Rolling Stone, Billboard, and Variety, filed a lawsuit accusing Google of using its content without permission to power the new AI “overview” answers in Google Search. The suit claims Google’s AI snippets pull directly from Penske’s articles and give users the info up top, reducing clicks through to the actual websites (Penske says its affiliate traffic is down by more than one-third since these AI summaries launched). Read more
🎵 Stable Audio 2.5 speeds up AI music generation for enterprise. Stability AI launched Stable Audio 2.5, a generative model that produces up to three-minute music tracks in mere seconds. Built for enterprise use, it uses a new 8-step generation method to drastically speed up output while improving quality. The model adds features like audio inpainting and better prompt control, enabling brands to quickly create custom, high-quality soundtracks at scale. Read more
🎙️ AI startup churns out 3,000 podcast episodes per week. Inception Point AI, led by former Wondery COO Jeanine Wright, is using generative AI to produce about 3,000 new podcast episodes every week. The company’s automated pipeline creates each one-hour show in roughly an hour at a cost of $1, featuring content voiced by a stable of ~50 AI “personalities” (like a faux nature host or a food expert). With over 10 million downloads since last year, the model is profitable with as few as 20 listeners per episode, though critics deride the content as “AI slop,” a label the CEO dismisses while predicting that soon “half the people on the planet will be AI.” Read more
💻 Replit launches Agent 3 after $250M boost. Coding platform Replit closed a $250 million funding round (tripling its valuation) and simultaneously launched Agent 3, its most autonomous AI coding assistant yet. The new Agent can run up to 200 minutes continuously (10x longer than before) debugging and improving its own code, and even building other software agents or automations from natural language prompts. Read more
🎧 AlterEgo debuts a 'telepathic' wearable for silent speech. Boston-based AlterEgo, a startup out of MIT, unveiled a headset device that lets users interact with computers or AI assistants using “silent speech” (essentially reading your mind’s intent to speak). The around-the-ear wearable uses tiny sensors and AI to detect subtle muscle signals in your jaw and throat when you think of words, translating those into text or commands in real time without you saying a word. In demos, AlterEgo’s CEO silently texted, coded, and even conversed with another person wearing the device, showcasing a potential future where we control tech nearly at the speed of thought. Read more
📂 Claude AI can now directly create Office files. Anthropic’s Claude assistant gained the ability to generate and edit actual files (including Excel spreadsheets, Word docs, PowerPoints, and PDFs) based on natural language instructions. You can ask Claude to build a financial model or turn meeting notes into a formatted document, and it will output a ready-to-use file with formulas, charts, and text as requested. Read more
🎬 OpenAI backs 'Critterz', a feature-length AI-made film. OpenAI is co-producing a fully AI-generated animated movie called Critterz, aiming to premiere at Cannes in 2026. The film, about a group of talking woodland animals, will use OpenAI’s latest models (GPT-4/5 for scripts and image models for animation) to create nearly all the visuals, potentially cutting production time to 9 months and budget to under $30 million. A team of around 30 humans (including writers from the Paddington franchise) is guiding the AI. Read more
📉 Musk’s xAI lays off hundreds of data annotators. Elon Musk’s AI startup xAI has reportedly laid off around 500 workers from its data annotation team, essentially its entire crew of “AI tutors” who were training the company’s Grok chatbot. Employees were notified by email of the abrupt downsizing, even as xAI claims it will hire more specialized AI trainers to replace the general labelers. This data annotation unit was xAI’s largest team, tasked with feeding and tuning Grok on raw data; their removal raises questions about xAI’s strategy (the company says it’s shifting to a smaller group of expert contractors as it refines the model). Read more
🧪 AI Research of the Week
Combinatorial prediction of therapeutic perturbations using causally inspired neural networks
From Harvard Medical School
Jake’s Take: The core idea is simple: teach an AI to pick small sets of genes to target so a sick cell’s activity pattern shifts toward healthy. The model, PDGrapher, learns from before-and-after experiments and uses a map of how genes influence one another to suggest combos worth testing.
This gives lab teams shorter, smarter lists for combination therapies in areas like cancer or neurodegeneration, and the tool is free to try. Hopefully we see a fast follow-up with CRISPR tests in real patient cells to check which picks translate.
and then, even more news…
💰 OpenAI shakes up its Microsoft deal to keep more revenue. According to The Information, OpenAI is renegotiating its partnership with Microsoft so that by 2030 it would only share about 8% of its revenue with Microsoft (and other partners), down from roughly 20% today. This shift could mean tens of billions more in OpenAI’s coffers over time, as the AI firm seeks a $500 billion valuation and transitions to a more profit-focused model. As part of the talks, Microsoft and OpenAI signed a non-binding term sheet to allow OpenAI’s nonprofit arm to restructure, and they’re also discussing how much OpenAI should pay Microsoft for the Azure cloud servers that power its models. Read more
🦾 Salesforce launches Agentforce for autonomous AI agents. Salesforce unveiled Agentforce, a new low-code platform for companies to build AI agents that not only chat but act autonomously in business workflows. These agents can carry out sales, service or marketing tasks end-to-end (updating records, sending emails, making decisions) using a built-in reasoning engine called Atlas to break down high-level requests into step-by-step plans. Read more