AI’s recent physics discovery

Good morning, { AI enthusiasts }. OpenAI’s GPT-5.2 just discovered that a widely accepted answer in particle physics was improper, proposed the right one, and autonomously wrote the formal proof in 12 hours.

The “can AI actually think?” debate from skeptics is not going away, but the true conversation is shifting from if AI can contribute to science to how briskly it rewrites what we thought we already knew.

In today’s AI rundown:

LATEST DEVELOPMENTS

The Rundown: OpenAI just published a brand new research preprint where GPT-5.2 independently discovered a mathematical formula and formally proved it was correct, marking what the corporate calls AI’s first original contribution to theoretical physics.

The small print:

Why it matters: There’ll still be debate from skeptics over whether AI is really able to ‘recent’ ideas, but the outcomes are getting harder to argue with. AI being pointed at and difficult long-held beliefs in humanity’s most significant scientific fields is beginning to feel less like sci-fi and more just like the very near future.

The Rundown: AWS Marketplace’s recent eBook breaks down how organizations are using agentic AI to streamline operations, with actionable guidance on constructing, buying, and deploying AI agents at scale.

Contained in the eBook, you will find:

Download the free eBook to kickstart your agentic AI journey.

The Rundown: The Rundown Roundtable is a weekly feature where we poll members of The Rundown staff about how we use AI in our work and each day lives.

Rishi, Growth: I connected Claude Code to Apify’s API to scrape high-performing content from IG/TikTok for ad creative inspiration, and to ElevenLabs’ API to mechanically transcribe videos — so I can analyze not only visuals, but the precise hooks, pacing, and language top creators are using. From there, I developed a scriptwriting system that takes rough thoughts on a hook and angle and turns them into ad scripts, drawing inspiration from winners and applying proven copywriting principles.

After generating, it also grades itself against a 12-point rubric. If it doesn’t rating at the least 90%, it rewrites until it does. I then feed it back in and say, “That is the ultimate version.” It analyzes the changes I made and updates its understanding of my style.

Nate, University Educator: I proceed to seek out that Claude Artifacts (with its front-end design) is delightfully useful — and use it several times a day to learn something recent, or make amends for a news story, and switch anything right into a custom webpage right inside chat.

Attach surveys, a protracted article, a spreadsheet, etc., and tell Claude to show it into an interactive page displaying key insights. The design is impressive, and it takes just minutes. Next time, do that to quickly share findings together with your team.

The Rundown: On this guide, you’ll construct an AI agent that makes real sales calls in your behalf. You’ll learn to create the agent, get a $1 phone number, and upload an inventory of contacts for it to call — with the entire setup taking about quarter-hour.

Step-by-step:

Pro tip: Toggle on “Transfer to Number” in your agent’s Tools to have it patch hot leads through to your phone.

The Rundown: Today’s buyers use AI day-after-day to reply their questions, and don’t have any patience for a scavenger hunt in your website. Concierge is a custom Perplexity-style answer engine, trained on your organization’s brand & content, that delivers accurate, personalized responses to any questions your website visitors have.

Modern B2B brands use Concierge to:

Use Concierge to show every query into a chance.

The Rundown: ByteDance released Seed 2.0, a brand new family of AI models that match or beat GPT-5.2 and Gemini 3 Pro across dozens of benchmarks at nearly 1/10 of the value — capping every week that also saw its Seedance 2.0 model spark a Hollywood firestorm.

The small print:

Why it matters: Move over, DeepSeek… ByteDance is the one rattling the Western AI landscape now. With Seed 2.0 now surpassing the Nov-Dec releases from top labs at bargain prices, the pressure on Western labs is just going one direction — and the Seedance IP drama shows China’s powerhouse isn’t slowing all the way down to ask permission.

QUICK HITS

OpenClaw creator Peter Steinberger is joining OpenAI, with Sam Altman posting that he’ll help “drive the subsequent generation of private agents”.

The Pentagon is considering cutting off Anthropic’s $200M defense deal over the refusal to let the military use Claude for “all lawful purposes.”

Anthropic’s Claude was reportedly used via a Pentagon-linked Palantir deployment to support the U.S. military operation that captured Venezuela’s Nicolás Maduro.

Spotify CEO Gustav Soderstrom revealed that the corporate’s top devs haven’t written a single line of code this yr, saying they’re “all in” on the transition to AI.

Alpha School shared recent test results showing its 2-hour, AI-first academic model has students scoring within the 99th percentile across virtually every grade and subject.

Simile raised $100M to construct AI simulations of human behavior, with agents modeled on real people to assist firms predict customer decisions.

COMMUNITY

Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.

Today’s workflow comes from reader Peter S. in Hollis, NH:

“We recently had a water pipe break in our basement, causing damage to the items we store there. The insurance company required us to itemize, take pictures, and supply an estimated alternative cost for every damaged item.

As a substitute of searching web sites, I uploaded the photos in Copilot with descriptions to acquire a superb estimate and a number of other links to where we should purchase the items. What would have been days to finish the inventory list become a few hours.”

How do you employ AI? Tell us here.

See you soon,