Thu July, 2026
3 Mins read
Case Studies
Founder's Desk
Physics & Cross-Domain Discovery
World models are giving AI a body. Engineering R&D will never be the same.
9 Mins read
I Asked GPT-4, Claude & Gemini to Design a Battery. All Violated Physics.
1 Mins read
Embracing Nature’s Wisdom: Resolving the Office Temperature Debate through Biomimicry
4 Mins read
Physics has never needed help. It’s been running the same governance model since long before anyone built a model to argue with it: obey, or don’t exist. No appeals process. No RLHF for reality.
Every few weeks there’s a new paper about teaching a language model to “understand” thermodynamics — like entropy is a foreign language it just needs more flashcards for. Wrong metaphor. Thermodynamics isn’t a dialect you get fluent in. It’s a bill that comes due whether or not you can conjugate it correctly.
The question was never whether an LLM can reason about physics. It’s why we keep asking it to.
Physics was never the thing missing AI. Engineering was — the part where someone has to decide, with real money and real deadlines, whether an idea deserves eighteen months of a team’s life.
When an engineering team looks at a new concept, two very different problems are happening at once, and the industry keeps treating them like one.
The first is discovery.
What’s been tried? What hasn’t? Which patent, which paper, which completely unrelated industry already solved a version of this thirty years ago under a different name? This is the part AI is genuinely, almost unfairly good at — finding the connection a human would need a decade of reading to stumble into.
The second is a much ruder question: can this thing actually exist? Not “does it sound plausible.” Not “would a model choose it.” Does it survive conservation of energy. Does it clear a stress limit. Nature doesn’t do plausible. Nature does yes or no.
A recent piece by Ebrahimi in Towards AI makes the same point from the control side: a delivery drone hits a wind gust, the flight controller handles the physics, the LLM decides whether to abort or replan. His line: “None of these components do the other’s job.” Correct.
I’d just push that boundary earlier. His system protects a drone already built. Nobody’s LLM is asked whether the drone should exist at all.
That’s the question actually worth the AI’s time. Should this concept move forward at all — before the eighteen months, before the $8M prototype, before anyone’s calendar clears for a simulation campaign that was doomed the day someone approved the direction.
We built KRAFT around one non-negotiable line. Every physics computation carries it.
llm_involvement: NONE
Not a guideline. Not a softly-worded best practice somebody can override under deadline pressure. The model discovers, connects, explains. It does not get a vote on whether something obeys the second law. It never has.
Physics was never waiting for AI to understand it. Engineering was — not for another chatbot, but because we’ve gotten very good at validating ideas and stayed strangely bad at deciding which ones earned the validation in the first place. That’s not a physics problem. It’s a discipline problem. And the next decade of engineering AI won’t belong to whoever builds the biggest model. It’ll belong to whoever draws the boundary first.
Fri April, 2026
9 Mins read
Thought Leadership · April 2026 We spent thirty years teaching AI to read. Then to write. Now we're teaching it… See more
Tue March, 2026
1 Mins read
I tested GPT-4, Claude, and Gemini on the same battery design challenge. All three violated the second law of thermodynamics.… See more
Sridhar DP
Founder's Desk
Wed January, 2026
7 Mins read
The $50M Trust Problem: Why Fortune 500 Companies Won't Bet on AI Predictions Every VP of Engineering knows the prototype… See more
Tue August, 2025
1 Mins read
Collection series - 2021 Want more innovation? Sridhar, Founder & Chief Integrator of Innovation Enabler, discusses creativity and innovation. He… See more
Sridhar DP
Founder's Desk
Quantum leaps - : Latest innovation breakthroughs and stratagies , delivered to your inbox .
Kreat.ai uses cookies to offer you a better experience. See Cookie Notice for details
Kreat.ai uses cookies to offer you a better experience. See Cookie Notice for details.