Ive Wants AI To Make Us Happy; Google For The Unimaginative; 40 Models, One Stop

Jony Ive Envisions ‘Peaceful’ AI Hardware with OpenAI

What’s new: Jony Ive and OpenAI are exploring 15–20 hardware concepts for a “family of devices” meant to fix our “uncomfortable” relationship with tech.

Ive argues it’s absurd to ship AI through legacy products; Sam Altman says a “totally new way to use a computer” will take time and must be justified. The emphasis is emotional well-being: tools that help people feel calmer, less anxious and more connected.

This is all well and good, but a far cry from the ambition shown when the two announced the marriage - suggesting shipping product in '26. Many reports of late point out this all may be harder than thought

🧩 Jargon buster: Agentic = AI that can take actions (not just chat), like clicking buttons, sending messages or running workflows.

Read more: The Rundown AI

Google’s Opal Turns “Blank Page Syndrome” Into Brainstorm Fuel

What’s new: Google Opal (no-code builder) lets you create a content brainstormer app that takes a topic and platform, pulls recent trends, and outputs hooks, outlines, hashtags - and even exports to Docs. It’s a friendly on-ramp for non-technical teams to spin up mini apps for LinkedIn, X or Instagram content.

🧩 Jargon buster: No-code = build apps via chat/drag-and-drop, not programming.

Read more: The Rundown AI

SoftBank Goes All-In on “Physical AI”

What’s new: SoftBank is reportedly buying ABB’s robotics division for $5.4B, pulling 7,000 engineers into its orbit to accelerate embodied AI - from chips to models to robots - which it's now calling "Physical AI."

The play: SoftBank CEO Masayoshi Son seems to want control of the hardware layer so AI can perceive, decide, and act in warehouses, retail, and logistics.

🧩 Jargon buster: Embodied AI = AI inside machines (robots, vehicles) interacting with the physical world.

Read more: AI Secret

Samsung’s Tiny Model Out-Reasons Giants on Puzzle Tasks

What’s new: Samsung’s Tiny Recursion Model (TRM) (≈ 7M parameters) reportedly beats DeepSeek R1 and Gemini 2.5 Pro on tricky logic puzzles, by drafting an answer and iteratively refining it through internal “rethinking” loops. It’s a sign that clever architecture can sometimes trump sheer size.

The model achieved 45% on ARC-AGI-1 and 8% on ARC-AGI-2, outperforming models thousands of times its size.

Rather than producing answers token by token, TRM drafts full solutions and then iteratively rethinks them - running up to 16 internal reasoning cycles to refine its logic. During each cycle, it uses a scratchpad system to critique and revise its reasoning about six times before finalizing an updated answer.

While these results are impressive for structured puzzle-style reasoning, they don’t yet generalize across broader cognitive tasks or open-ended problem domains.

🧩 Jargon buster: Parameters = the learned “knobs” in a model; fewer usually means lighter, cheaper, but not always worse.

Read more: The Rundown AI

IKEA Buys Locus To Own Its Last-Mile Algorithm

What’s new: IKEA/Ingka Group acquired Locus (routing optimization) to save ~$100M/year and give shoppers more delivery flexibility. Beyond speed, IKEA is pulling delivery intelligence in-house ... every van and route becomes data for pricing, inventory and customer models.

Says AI Secret, "retailers still outsourcing their last mile are not just losing efficiency, they’re surrendering the operating system of modern commerce, one delivery at a time."

🧩 Jargon buster: Last-mile = the final delivery leg to the customer - often the most expensive/complex.

Read more: AI Secret

Qualtrics Turns Patient Experience Into Intelligence

What’s new: Qualtrics is buying Press Ganey/Forsta for 6.75B to make a major healthcare push by pairing its AI sentiment tools with clinical benchmarking.

The goal? Turn patient feedback and clinician notes into actionable ops - reducing churn, improving satisfaction and tightening regulatory compliance.

🧩 Jargon buster: Experience data (X-data) = signals about how people feel (not just what they did).

Read more: AI Secret

Anthropic Researcher Cites “Adversarial Nation” Policy, Moves to Google

What’s new: Physicist-turned-AI researcher Yao Shunyu left Anthropic and joined Google DeepMind, pointing to Anthropic’s policy restricting access for subsidiaries in “adversarial nations like China” as a key reason.

He contributed to Claude 3.7 Sonnet and Claude 4 before departing, and will work on Gemini at Google.

🧩 Jargon buster: Foundation model = a large, general-purpose AI model (like Gemini or GPT) that can be fine-tuned for many tasks.

Read more: The Rundown AI

Tools & Launches

• Gemini 2.5 Computer Use - Google’s browser-control model that clicks, types, and navigates UIs on its own ... useful for booking, form-filling, and repetitive web workflows.
• Grok Imagine v0.9 - xAI’s upgraded image/video generator with synced audio and tighter motion consistency.
• ChatPlayground AI - Compare 40+ models in one place to see which fits your use case.
• QA.tech - Deploy a fleet of QA agents that automate software testing, deliver debugging context into pull requests, and improve coverage and reliability without manual review.
• VoiceType.com - Write hands-free at up to 9× faster using accurate, privacy-minded voice capture.
• The Brief - An AI agency in your browser that manages the entire marketing campaign lifecycle, from insights to optimization, without leaving your workspace.
• Retool AI AppGen - Generate secure, production-ready apps directly from natural language prompts that use your company’s data.
• ProhostAI - An AI assistant for vacation rental hosts that automates guest communication, cleaning schedules, and upsell offers—a full-stack co-host.
• Smart Clerk - Turns messy bank statements into bookkeeper-ready reports with auto-categorization.
• PromptSignal - Track how your brand ranks across ChatGPT, Claude, Perplexity, Gemini answers; monitor visibility & sentiment.

Today’s Sources: AI Secret, The Rundown AI, There’s An AI For That