AIToday
Large Language ModelsZenn AI/MLPublished: Oct 13, 2026, 22:00 JST

USPTO criteria made AI produce 6 inventions in 30 minutes

USPTO criteria made AI produce 6 inventions in 30 minutes

3 Key Points

  1. What happened

    The author fed AI a team roadmap, survey results on world models and medical AI, one constraint — cross the two areas — and USPTO novelty (§102) and non-obviousness (§103) criteria, and got six proposals in about 30 minutes.

  2. Why it matters

    The author reports the tweak that changed the output was the evaluation criteria — supplying the patent review standard pushed the AI to explain why each idea would not be obvious, and the six proposals ranged from an operating-room physics simulator to a cell-state model that optimizes drug dosing order.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The writeup is built around two free-to-try tools and one experiment. World Labs' Marble, released in February 2026, takes a single photo and produces a persistent 3D space that can be exported to Unity, Unreal, or Blender; it has a free tier of 4 generations and paid plans from $20 per month. DeepMind's AlphaGenome reads up to 1 million base pairs and predicts thousands of properties across the 98% of the genome that does not code for proteins, with source code and weights released for non-commercial use on January 28, 2026, alongside a Nature paper; DeepMind reports it outperformed existing models on 25 of 26 evaluations.

The author then describes feeding AI a team roadmap, survey results on world models and medical AI, a single constraint to cross the two areas, and USPTO novelty and non-obviousness criteria. Six proposals came back in about 30 minutes. The author singles out one — a system that generates verifiable simulation evidence for regulatory submissions by detecting uncovered scenario space and auto-generating runs to fill it — as the most promising, and notes the practical patent lesson that claims are better written as a control or simulation device than as a diagnostic method.

FAQ
What tools does the writeup cover?
It covers World Labs' Marble, which turns a single photo into a walkable 3D space, and DeepMind's AlphaGenome, which reads up to 1 million base pairs and predicts thousands of properties in non-coding regions.
How can I try these tools?
Marble has a free tier that allows 4 generations without a credit card. AlphaGenome is free for non-commercial use; you get an API key and run pip install alphagenome.
What did the AI invent?
It generated six proposals, including a surgical simulator with physics-conservation gating, a system that separates robot failures from biological signals, and a verified simulation-evidence generator for regulatory submissions.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleMicroLLM lab runs seven tiny AI models in your browser