Claude 4.5 Sonnet arrives while OpenAI preps compute-intensive offerings
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
🧑💻 Anthropic releases Claude 4.5 Sonnet for long-running agents. Anthropic shipped Claude 4.5 Sonnet yesterday with gains in coding, agentic workflows, and computer use. The model sustained ~30-hour autonomous sessions in evaluations (a jump from ~7 hours in prior versions) and introduces finer control over “thinking” duration alongside upgrades to Claude Code for IDE and terminal use. Read more
🤖 OpenAI tests high‑compute ChatGPT features for pros. OpenAI’s Sam Altman revealed the company will roll out new “compute-intensive” ChatGPT features that will initially be exclusive to Pro subscribers due to their high processing costs. Some advanced tools might carry extra fees, like the rumored upgrade to their Sora video generation model. The goal is to push AI limits by throwing massive compute at tasks (even as OpenAI builds out huge GPU clusters) while promising to eventually drive down costs for consumers. Read more
🎬 AI actress Tilly Norwood nears Hollywood deal. An AI-generated “actress” named Tilly Norwood is close to securing representation with a major Hollywood talent agency. Creator Eline Van der Velde of Xicoia said at the Zurich Summit that they’ll soon announce which agency will rep the virtual actress, potentially a first in the industry. The move has sparked backlash from human actors, who called it “gross” and worry about AI replacements, but Norwood’s team insists she’s meant to complement, not replace, real performers. Read more
📹 Meta launches ‘Vibes’ AI video feed, teams with Midjourney. Meta introduced Vibes, a TikTok-style feed in its Meta AI app that shows only AI-generated short videos. Users can create or remix quirky AI videos (like fuzzy creatures or an ancient selfie) and share them across Instagram and Facebook. For Vibes’ early version, Meta partnered with Midjourney’s image-gen tech to boost visual quality while it develops its own models, part of Meta’s push to catch up in generative AI. Early user reactions were skeptical of the “AI slop” flood, but Meta is betting on Vibes to spur creative engagement. Read more
💻 Google finds 90% of developers now use AI tools. Google’s latest DORA 2025 report reveals AI adoption among software professionals has surged to 90%, up 14% from last year. Developers are integrating AI assistants into coding workflows for tasks like writing new code, fixing bugs, generating tests, and documentation. On average they spend 2 hours a day with AI help, and over 80% report higher productivity and code quality. Read more
🏥 Trump shares (then deletes) fake ‘MedBed’ AI video. Former President Donald Trump briefly posted a bizarre video – apparently AI-generated – that showed him announcing “MedBed” hospitals offering miracle cures to every American. The fake clip, styled like a Fox News segment with Lara Trump as host, had an AI Trump promising everyone a “medbed card” for free access to futuristic healing beds. In reality, “medbeds” are a conspiracy theory (think alien technology that cures all ailments) with zero evidence. Read more
✈️ Lufthansa to cut 4,000 jobs as it pivots to AI. Europe’s largest airline, Lufthansa, announced plans to eliminate 4,000 administrative jobs by 2030 as it accelerates automation and AI to boost efficiency. The cuts (about 20% of non-ops staff) will be mainly in Germany and come as the carrier struggles with high costs and seeks a higher profit margin. Lufthansa’s CEO said digitalization and AI-driven processes will help streamline operations, but unions are criticizing the move. Read more
🗞️ OpenAI’s ‘Pulse’ gives you an AI-curated morning briefing. OpenAI launched ChatGPT Pulse, a new feature that proactively generates personalized daily updates for users. Instead of waiting for queries, Pulse works overnight to deliver 5–10 briefings when you wake up – like your own AI news digest, task list, or creative prompts tailored from your chats and interests. Currently it’s only available to $200/month ChatGPT Pro subscribers (appearing as a new tab in the app) because it’s so compute-intensive. Read more
🧠 AI tool distinguishes deadly brain tumors in seconds. Researchers led by Harvard Medical School developed an AI system (called PICTURE) that helps neurosurgeons instantly tell apart look-alike brain cancers during surgery. Glioblastoma – an aggressive brain tumor – often resembles primary lymphoma in the brain under a microscope, leading to misdiagnosis. The AI was trained on pathology images and can identify glioblastomas vs. lymphomas with 98%+ accuracy in real time. In tests across five hospitals, it even outperformed human pathologists and other AIs. Importantly, the model has a built-in “uncertainty” flag to punt any unusual cases for human review. Read more
🧪 AI Research of the Week
Structural constraint integration in a generative model for the discovery of quantum materials
From MIT, Oak Ridge National Laboratory, others
Jake’s Take: The team taught a generative AI (called SCIGEN) to follow simple crystal geometry rules that tend to produce quantum effects, then let it propose millions of candidate materials. They ran physics checks on a computer to filter the list, synthesized two picks in the lab, and measured the magnetic behavior the AI predicted.
In practice: tell the AI the structure you want, screen for stability, then build. This should speed up materials discovery for superconductors and spin electronics (though the bar remains proof of strong performance at workable temperatures and scalable synthesis).
and then, even more news…
🖥️ Anthropic teases Claude AI that builds its own UI. A sneak peek from Anthropic suggests their Claude AI assistant might soon do more than chat – it could generate entire mini-applications with a GUI. A leaked demo called “Imagine with Claude” shows an interface where users ask Claude to “build” something, and the AI spawns a desktop-like window with icons, buttons, and interactive elements. In this prototype, Claude can effectively write and arrange its own interface on the fly (for example, creating a game or a dashboard app inside a window) rather than just returning text. Anthropic indicates this Imagine feature is an experiment for now (possibly a limited-time demo for Claude Max users) as they explore letting AI agents manipulate UIs directly. Read more