Good morning! Here's what's happening in AI today:
xAI releases Grok for Excel for free spreadsheet building
Google Vids adds Gemini Omni for sketch to video generation
Mathematician claims Fable helped disprove the Jacobian conjecture
How to actually use Kimi K3 instead of just screenshotting the chart
And 4 new AI tools worth trying today
AI PRODUCTIVITY
Grok for Excel launched as a free add in that lets you build financial models, analyze data, and generate charts using plain language inside Excel itself.
Ask Grok to build a full financial model from a plain language description and it fills in the formulas, labels, and structure on its own, so you go from a rough idea to a working sheet without touching a single cell manually.
Grok can read market data you already have sitting in a sheet and summarize the patterns in plain language, catching trends you would normally have to build a pivot table or a separate chart just to see.
It works as a native add in inside the Excel interface you already use every day, so there is no new app to install, no separate login, and nothing to relearn to start using it.
Most people who want AI help with a spreadsheet have had to leave Excel to get it, copying data back and forth between tabs and tools. Grok for Excel puts that help directly inside the tool, right next to the data itself. For anyone who builds models or reports regularly, this cuts out a step that used to eat real time every single week.
AI VIDEO
Google Vids added Gemini Omni, letting you drop in reference photos or drawings as ingredients to generate video clips just by typing a prompt.
Drop in a rough sketch, a moodboard, or a reference photo as an ingredient and Gemini Omni uses it to shape the look, color, and style of the video clip it builds, so your own visual reference actually directs the output.
You describe the scene you want in a plain text prompt and the tool combines that description with your uploaded images to generate a finished, thematic clip in seconds, skipping the usual back and forth of a normal edit.
It is available right now on desktop through Google Vids with no separate app, plan, or account setup required, so anyone already using Google's tools can try it today.
Turning a sketch or a moodboard into an actual video used to require real editing skill and hours of work. Gemini Omni removes that barrier by letting your own reference images do the directing instead of you learning software. For anyone making quick video content on a regular basis, this shortens a process that used to take real production time down to minutes.
AI BREAKTHROUGH
Mathematician Levent Alpoge posted a single explicit polynomial map that he says Claude Fable helped find, claiming it disproves the 85 year old Jacobian conjecture.
The Jacobian conjecture asks whether any polynomial map that never locally collapses space, meaning it never squeezes the space it touches down to fewer dimensions, must always be reversible, and it has stood unproven since 1939 despite decades of attempts by serious mathematicians.
Alpoge posted one specific three variable map with a constant Jacobian determinant of negative two, along with three separate input points that all land on the exact same output, which would break the conjecture outright if the example survives scrutiny.
The post credits Fable with finding the example during a single working session while a friend asked about the problem during the World Cup final, adding a concrete AI plus math angle to a story that would already be notable on its own.
This is not a confirmed result yet. It is a viral, unreviewed claim from one mathematician's post, and math history is full of counterexamples that looked airtight at first glance and did not survive closer scrutiny. What makes it worth watching regardless is the AI angle, a frontier model reportedly helping surface a concrete challenge to a problem that has resisted formal proof for nearly a century.
HOW TO AI
A benchmark chart goes viral. Everyone screenshots it. A tiny fraction of the people who saw that chart ever actually route a real task to the model.
Kimi K3 just landed in exactly that spot, 2.8 trillion parameters, a 1 million token context window, and a Frontend Arena score that beats Claude Fable 5 and GPT-5.6 Sol. Most people will stop at the screenshot.
What it actually is
K3 is a Mixture-of-Experts model from Moonshot AI, a Beijing lab backed by Alibaba. It routes each request to 16 of 896 available experts rather than using the full 2.8 trillion parameters every time, which is why a model this large still runs at a usable speed.
For API usage, pricing is $3 per million input tokens, $15 per million output, with cache hits at $0.30 per million, ten times cheaper if you structure your prompts right.
The trap most people fall into
K3 currently only has one reasoning mode, max effort. There's no fast option yet. Every single request, even a one-line question, burns full thinking output at the higher price. Independent testers clocked over 13,000 reasoning tokens just to generate a simple icon. Don't route trivial tasks here, save K3 for work that actually justifies the cost.
Try it free first
Go to kimi.com, confirm K3 is selected before sending anything, and test it with something real, not a one-line question. K3 is available under standard rate limits with no API key needed.
One limitation worth knowing: the full 1 million token context window is gated to paid membership tiers, the free plan gives you the model, not the full window, which is fine for judging quality but not for genuine long-context work.
Test it with something real
Ask K3 something more substantial than a one-line question, an explanation with tradeoffs, or a small script it can write and explain.
The money-saving trick
Keep your stable context, your project docs, your codebase, identical across every request, with the part that changes coming last. Identical prefixes hit the cache and cost ten times less. Shift that stable context even slightly and you lose the discount and pay full price again.
See what it can actually build
Ask it for something small and self-contained, like a single HTML file with a working animation, no external libraries. Save the code it gives you and open it in a browser to confirm it actually runs.
When to route to K3, and when not to
Long-context reading, repo-wide analysis, and frontend-heavy coding are where K3 earns its cost, this is genuinely where it currently leads. For the hardest multi-step reasoning or anything security-sensitive, stay on Fable 5 or Opus 4.8 for now, K3's hallucination rate climbed alongside its accuracy gains, a real tradeoff worth knowing before pointing it at anything factual and unsupervised. For quick edits and everyday small tasks, use something cheaper entirely, K3's max-only reasoning makes it the wrong tool for anything trivial.
P.S. You can access all my prompts, workflows and AI trainings if you upgrade here
Linear introduced Loops, recurring workflows that Linear Agent can run for your team on its own.
Gemini Notebook rolled out Collections, a flexible new way to organize your notebooks without rigid folders.
Reve introduced Templates, letting you swap in your product, copy, or brand onto a ready made layout in seconds.
🎮 Unity CLI: Connect coding agents and pipelines to your Unity projects straight from the terminal.
🎥 Reve Video: Bring your images to life with Kling 3 and Seedance 2 built right in.
🦥 Unsloth for AMD: Train and run open models like Qwen and Gemma on your own AMD hardware.
🧭 Ramp Router: One endpoint that automatically picks the right model for every request.
THAT’S IT FOR TODAY
Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!
See you tomorrow :)
- Dr. Alvaro Cintas
✓ Full archive of premium guides with ready-to-use prompts
✓ Structured AI courses (step-by-step, start-to-finish)
✓ Every upcoming premium tutorial