@RefineDotInk

AI feedback for research. Refine devotes hours of compute to help you find and fix issues that matter to readers and reviewers. Try it at https://xitter.cf/t.co/5I1Ggidlzo

refine.ink
Joined October 2025
  • Tweets270
  • Following377
  • Followers4.4K
11Media
Refine is a diligence tool for science and any other area that demands the same high standards. You process a document. Refine studies it and flags anything that threatens correctness, consistency, or clarity. A Refine report saves hours or days of expert work. 1/
3
7
4
61
47,391
This is awesome. At Refine, we know deep AI reviews are an indispensable tool for the working scientist. Good system outperforms humans in many important dimensions. We also think the insight of top human researchers on what's important is still way beyond the models.
Brendan Nyhan (@BrendanNyhan on 🟦☁️)@BrendanNyhanOct 23
"No matter how we look at the data, AI wins. It wins in 100% of the pairs overall when judged by another AI model...and 87% overall when judged by human journal editors... [H]umans agreed as much or more with AI’s judgment than they did with other humans"
1
2
16
1,984
Refine retweeted
Ray Fisman@RFismanOct 23
Replying to @emollick
This is why we use @RefineDotInk as an "extra" reviewer at AEA journals. But its role is very different from human referees.
1
1
12
1,259
Replying to @evavivalt @paulnovosad
They could have at least paid for @RefineDotInk. :P
1
1
733
Did this referee write their own review and then just “clean it up” with AI? 1. I doubt it 2. Even if they did, this practice is destroying trust AI writing in ref reports should be banned and sanctioned
Eva Vivalt
@evavivaltSep 29
Very frustrating when a paper is rejected on the basis of a negative review that is not just largely AI-generated (per Pangram) but materially incorrect.
6
4
2
77
15,093
Refine retweeted
Refine
@RefineDotInkSep 28
From economics to medicine: see how Refine 5 performed in our 108-paper review benchmark. Find errors before you submit. Try a free preview—no credit card required. New customers get 80% off their first purchase.

2
16
154,218
Refine retweeted
Ben Golub
@ben_golubSep 23
One thing that's implicit in this release is that we're finally doing vesioning! Refine has been updated multiple times every month behind the scenes, but versions give us a clear way to communicate when and how much the technology is improving.
Ben Golub
@ben_golubSep 23
Refine 5, a strong new reviewing tool, is live today. It beats the amazing new frontier models at verifying technical work. Free with every Refine review through 10/15. We benchmarked on 108 papers including bio, engineering, physics, environmental science, econ. 🧵 1/
Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

ALT Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

5
23
4,141
Each Refine 5 review uses enough frontier tokens that we're flirting with losing money on it. To celebrate the launch, we decided to get crazier. Is your research outside social science? Reply "Try Refine 5 on [your field]" for free or dirt-cheap reviews. Rules below 👇
Ben Golub
@ben_golubSep 23
Refine 5, a strong new reviewing tool, is live today. It beats the amazing new frontier models at verifying technical work. Free with every Refine review through 10/15. We benchmarked on 108 papers including bio, engineering, physics, environmental science, econ. 🧵 1/
Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

ALT Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

41
11
1
33
21,175
🥇 First 5 replies: 5 full reviews, FREE 🥈 Next 25: reviews at 95% off 🥉 Next 25: reviews at 55% off A review checks every detail like an obsessive but friendly referee or coauthor, minus the 3-month wait. Follow us (just for a bit) and we'll DM you.
6
5
1,276
A major upgrade to the reviewer behind the scenes.
Ben Golub
@ben_golubSep 23
Refine 5, a strong new reviewing tool, is live today. It beats the amazing new frontier models at verifying technical work. Free with every Refine review through 10/15. We benchmarked on 108 papers including bio, engineering, physics, environmental science, econ. 🧵 1/
Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

ALT Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

1
4
1,012
Refine retweeted
Ben Golub
@ben_golubSep 23
Refine 5, a strong new reviewing tool, is live today. It beats the amazing new frontier models at verifying technical work. Free with every Refine review through 10/15. We benchmarked on 108 papers including bio, engineering, physics, environmental science, econ. 🧵 1/
Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

ALT Refine 5's wins in each research area, against GPT-6 Astra and Claude Fable 5.1 combined. Economics: macro 14 of 18, econometrics 18 of 18, applied micro 15 of 18, theory 14 of 18. STEM: clinical epidemiology and biostatistics 24 of 24, computational genomics 24 of 24, control theory and dynamical systems 24 of 24, environmental and ecological modelling 23 of 23, machine learning and statistical learning 22 of 24, statistical physics and soft matter 19 of 24.

5
21
8
130
33,925
Refine retweeted
Ben Golub
@ben_golubSep 22
ICLR is ... not in good shape in terms of harnessing AI for academic review Good day to check out @RefineDotInk
XY Han
@XYHan_Sep 16
Just learned that the Paper Assistant Tool (PAT) that ICLR uses for auto-generating llm-reviews is powered by Gemini 2.5 Deep Think. To which I respond: What makes Gemini 2.5 DT believe it's qualified to judge the paper I wrote with Fable 5.1 Max and GPT 6 Astra Ultra?
2
2
34
6,048
Refine retweeted
Ben Golub
@ben_golubSep 20
we're cooking at Refine over the weekend. teaser of what's coming over next 2 weeks: new pipeline is set for release (already running for many) we obsess over delivering value over any alternative -- here's an eval comparing review quality to a great frontier prompt 1/
6
7
94
14,292
Today @Google, we published work led by @m_codreanu on AI in Science, drawing on 15 million Gemini interactions, 2,600 specialised AI models & a 600-scientist survey. What did we find about how AI is changing science? Read on...
28
274
28
1,072
167,204
Replying to @NorthwesternU
@NorthwesternU Economics department dropping some hard truths about predocs (good!)
22
69
41
647
211,314
I wonder to which extent the important economic theory papers of the past, published in good economics journals, have been checked by AI for errors. It seems this would be a worthwhile project.
7
1
14
3,754
Replying to @RemyLevin @ben_golub
On one of my papers, @RefineDotInk rated as one of the priority comments the slightly unclear variable description in the table notes in Appendix C... I took that to mean the paper was in pretty good shape and I should probably submit already 🙃
1
1
5
771
Refine retweeted
Max Spero
@max_spero_Sep 3
Talking with journalists is cool because you have to make sure you never say anything that could be twisted or taken out of context. But if you take a couple seconds to think about a question, they can just write about you as if you're the most awkward person alive
WIRED
@WIREDSep 2
Meet the AI police who can make or break careers—in publishing and beyond. wired.com/story/pangram-has-…

Pangram Has Emerged as the Gold Standard of AI Detection. Should You Trust It?

Meet the AI police who can make or break careers—in publishing and beyond.

wired.com
309
228
121
9,528
1,025,484
Refine retweeted
Ben Golub
@ben_golubAug 31
If you like Refine (our tool for deep technical reviews of work in STEM and quantitative social science): A chance to stock up on credits for the start of the academic year: code BACKTOSCHOOL at checkout gives you 35% off the 3-review or 10-review package.
2
7
1
98
13,345
Code here for a big discount on reviews - Expires tomorrow, limited number of redemptions :)
Ben Golub
@ben_golubAug 31
If you like Refine (our tool for deep technical reviews of work in STEM and quantitative social science): A chance to stock up on credits for the start of the academic year: code BACKTOSCHOOL at checkout gives you 35% off the 3-review or 10-review package.
5
2,070
This is genuine intellectual humility: placing truth above pride.
Ben Golub
@ben_golubAug 28
It’s scary to put all your past research under a microscope. Every article at the frontier of math and science has mistakes, and you don't know whether they unravel work you are proud of until you look. @littmath had the courage to run the experiment with @RefineDotInk 1/
2
13
2,809
Load more