Google's Deep Review wants to be the second pair of eyes on every long document 🔍

AI has made writing fast. Checking has not kept up. That gap is exactly what Google Research is going after with Deep Review, which went public on 14 October. It's an agentic system that reads a long technical document and tries to find what's wrong with it: broken proofs, logic gaps, unsupported claims. For anyone working on legal and document AI, like me, this is a launch worth a close look.

From conference pilot to public tool

Deep Review is the grown-up version of Google's Paper Assistant Tool (PAT). Over the past year, authors of more than 16,000 manuscripts at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027 opted in for private feedback before the submission deadline, fully separate from official peer review. The need is real: Google counts over 74,000 submissions to ICLR, NeurIPS and ICML this year, up from about 23,000 in 2023.

Now anyone can use it. Every user gets two free reviews per week, and you can connect a Gemini API key for higher limits and programmatic access through the Deep Review API. Google says uploaded documents are never used to train its models and are stored for one year, only so you can see them in the interface.

Why the design is interesting

No single giant prompt. One LLM call on a long document tends to run out of reasoning budget, miss links between sections, or accept subtle claims too easily. Deep Review uses a four-stage pipeline instead. A segmenter agent splits the document into logical (possibly overlapping) parts. Compute is allocated by information density, so a dense proof gets more than an introduction. Specialised agents review each segment in parallel while still seeing the full document. Finally a synthesis agent merges the reports, re-runs consistency checks and does external grounding against hallucinations.

Encouraging numbers. In the June preprint, on a subset of the SPOT benchmark (26 papers with 29 known math and proof errors), PAT found 89.7% of the errors, against 55.2% for a single zero-shot Gemini 3.1 Pro call and 21.1% for the original SPOT state of the art. In the ICML author survey, 92.1% of 733 respondents would use it again, and 31% said they ran entirely new experiments because of its feedback.

Beyond research papers. Google explicitly lists design documents, presentations, and legal and financial documents, where it checks consistency across clauses and data tables and flags ambiguous terms. That is the part that excites me most: contract review is exactly this kind of careful cross-referencing. Antigravity users also get a "Teamwork Deep Document Review Pattern" through the /teamwork-preview command.

Keeping it honest

This is not a replacement for a human reviewer, and Google doesn't pretend it is: the aim is to take on the most tiring checks "while leaving higher-level judgements to humans". The preprint lists the most common pilot complaints: date hallucinations, PDF parsing issues, and falsely claiming a correct proof is wrong. Only 55.8% of STOC and 64.8% of ICML respondents found the feedback mostly or all grounded. Good for an AI tool, but you still verify every critique. The legal and finance use cases are also newer than the pilots, so I'd love to see published results there. At launch it takes PDFs and plain text, with multi-file input "coming soon".

Still, I really like the direction. If AI helps us write faster, it should also help us check harder. Two free reviews a week is plenty to try it on your next draft, and I'm curious how it handles a real contract.

Sources

Reacties

Een reactie posten

Populaire posts van deze blog

SpaceX just asked to fly Starship hundreds of times a year 🚀

Last week the U.S. Federal Aviation Administration took a quiet but important step: it opened a formal review of SpaceX's request to launch Starship from Starbase in South Texas up to 456 times per year . Today the site is cleared for only 25 flights a year. That jump is the clearest signal yet that Starship is meant to become routine hardware, not a rare test rocket. What the FAA is reviewing On 14 October 2026 the FAA published a Notice of Intent to prepare a Programmatic Environmental Impact Statement (PEIS) for SpaceX's Starbase plans. A PEIS looks at a whole program of operations rather than a single launch, which is exactly what you need if you want a reusable rocket to fly the way airlines fly planes. SpaceX's proposal comes in two phases, according to the Federal Register notice: Phase 1: up to 100 launches and 200 landings per year (100 Ship, 100 Super Heavy), any day or time of week. Phase 2: up to 456 launches and 912 landings per year at Starbase. La...
Meer lezen

AI that can actually "see" a crystal 🔬

Most of the solid stuff around us, from battery materials to chips, is made of crystals: atoms stacked in neat, repeating 3D patterns. Large language models are great at reading papers about those materials, but they have never really looked at the structures themselves. A team at MIT just showed a clever way around that, and the trick is one I keep coming back to: good synthetic data. What they built The model is called CLaC, short for Contrastive Language-Crystal model, and it is described in a paper in Nature Communications by Yang Jeong Park, Mayank Kumaran, Chia-Wei Hsu, Elsa Olivetti and Ju Li. MIT News covered it on 18 October 2026. If you know CLIP, the model that taught image generators to connect pictures and words, you already get the idea. CLaC has two branches: a graph neural network that reads the 3D crystal structure, and a text encoder that reads a description. Contrastive training pulls matching structure-text pairs together in one shared space and pushes mismatche...
Meer lezen