Paul Esau

93 posts

Paul Esau

@ptesau

Spring 2023 @ScovillePF at @WisconsinProj. Military export policy writer/researcher. 2022 @Laurier and @StudyofCanada alum.

Washington, D.C Katılım Ekim 2014

105 Takip Edilen71 Takipçiler

Paul Esau@ptesau·2d

IMO, writers and editors should both be using detectors in their workflows, especially if they have also integrated other AI research/editing tools. AI can liberate people from the "hardness" of writing (empowering for many!), but only by introducing other risks. (2/2)

English

Paul Esau@ptesau·2d

I agree with @alexcdot that, while ragebait gets clicks, AI detection is fundamentally about transparency and trust. I'm not in favor of "gotcha" moments unless factual errors (bad stats, vibe citations, metacommentary, etc.) indicate no human checked the work. (1/2)

Alex Cui@alexcdot

WSJ reports @pangramlabs on falsely accusing people of AI. wsj.com/opinion/the-ai… The ai detection industry needs to rise above accusations as a way to drive use (which people on twitter love, I get it, ragebait sells) If we really want to promote human-written content, we should help people establish that their writing is human. Give people the ability to replay their process. Trust comes from transparency. So what did we do about this at @GPTZeroAI ?

English

Paul Esau@ptesau·31 Mar

I've loved working on this project (and with this team). AI. It's a little terrifying, but it's also SO COOL to have an editor available at any hour (and trained by a cohort of diverse experts).

Edward Tian@edward_the6

Today, we’re launching AI Reviewer Expert Feedback. We partnered with Emmy-winning writers to train AIs to give expert writing feedback through GPTZero.

English

Paul Esau@ptesau·28 Mar

@shadihamid @matthewkaemingk Thanks for the rec! And thanks to both of you again for doing what you do! (P.S. Shadi: Not only is it a signed copy, but it's lived lived rent-free in my head for the last two years — I keep referencing it in convos about US FP.)

English

Shadi Hamid@shadihamid·28 Mar

@ptesau @matthewkaemingk Christian Hospitality and Muslim Immigration in an Age of Fear is amazing! Also thanks for buying The Problem of Democracy :)

English

Shadi Hamid@shadihamid·25 Mar

New season of ZEALOTS AT THE GATE is about to drop. The only Muslim-Evangelical podcast around to my knowledge.

Comment@commentmag

Zealots at the Gate has a new home at @Georgetown and a new season coming soon. Subscribe wherever you stream podcasts!

English

Paul Esau@ptesau·27 Mar

@shadihamid @matthewkaemingk Although it reminds me that I bought Shadi's The Problem of Democracy in DC a few years back, but never picked up one of Matthew's books. Seems unfair. Which one would you recommend, @shadihamid ? 😀

English

Shadi Hamid@shadihamid·27 Mar

@ptesau Thank you! Means a lot to hear that cc @matthewkaemingk

English

Paul Esau@ptesau·25 Mar

Now that's a crazy coincidence! Great pull!

(((James Acton)))@james_acton32

The first attack on a nuclear facility occurred 36 years BEFORE Iran's attack on Osirak. In 1944, by pure luck, Japan attacked the Hanford plant in Washington that was producing the plutonium that would be used in the nuclear weapon dropped on Nagasaki. (1/n)

English

Paul Esau@ptesau·18 Mar

@dex_eve So the NYT could be correct in calling out Cooper's reluctance to apply Article 51 fully, instead of conveniently narrowing the definition of "indiscriminate" to point b?

English

Paul Esau@ptesau·18 Mar

@dex_eve Doesn't this last concession invalidate your original criticism of the NYT, tho? The 1% failure rate is a hypothetical ideal for future US-manufactured cluster munitions — the cluster munitions transferred to Ukraine in 2023 from stockpiles had a much higher failure rate (5-10%).

English

Decker Eveleth@dex_eve·18 Mar

This is a disappointing article from the NYT. It does not seem to understand the substance of Adm. Cooper's comment, and thus ends up with an argument that is mostly irrelevant to said comments. A short thread.

John Ismay@johnismay

CENTCOM is unhappy with this story, but unable to argue with the facts it presents. This morning I offered an on-record interview with Adm. Cooper in case he wanted to clarify his remarks, but that was rejected out of hand by a staffer nytimes.com/2026/03/17/us/…

English

423

121.7K

Paul Esau@ptesau·11 Mar

I'm glad someone is writing about the FCPA and the consequences of the Trumpian pause. I mean, I hate the emotional fallout of reading pieces like this, but I'm glad they exist.

Richard Nephew@RichardMNephew

Latest from me on anti-corruption issues (yes, that is still a problem too) A Year Later – What Did the Pause on FCPA Enforcement Do? at justsecurity.org/133481/year-la…

English

Paul Esau@ptesau·26 Şub

Tried it this morning on X and LinkedIn. A very cool tool that's going to generate a ton of embarrassing conversations (and hopefully force some accountability).

Edward Tian@edward_the6

Today, we're launching AI Vision. The first AI slop detector that exposes content as you scroll.

English

Paul Esau@ptesau·19 Şub

@Er_Woods @edward_the6 Considering your level of expertise, being a dinosaur is a solid strategic choice. Dataset coding checks out, but I'm surprised it's been so useful for language learning. Chinese, right? And you don't use any tools for baseline research or comms clarity/structure suggestions?

English

Eric Woods@Er_Woods·19 Şub

@ptesau @edward_the6 It’s very good at languages (shocking!) so I’ve also found use in assisting with language learning. Namely as a grammar reference and tool to break down complex sentences and vocab into their component parts

English

Edward Tian@edward_the6·18 Şub

When was the last you didn’t use AI to write?

English

968

Paul Esau@ptesau·19 Şub

@Er_Woods @edward_the6 I'm actually pretty curious what benefits you've seen from AI tools given the technical nature of your work, the multiple languages involved, and the consequences of hallucinations. Have you experienced any significant increases in productivity, or just lower noise/signal ratios?

English

Eric Woods@Er_Woods·18 Şub

@edward_the6 right now

English

154

Paul Esau@ptesau·16 Şub

I agree. The Venn diagram of people who claim hallucinations are no longer a problem and people who never check the footnotes is a circle.

Edward Tian@edward_the6

Just tested chatgpt 5.2 for hallucinations. EVERYONES saying it's no longer a problem in 2026. Well guess what... It hallucinated over 10/40 citations on this prompt.

English

Paul Esau retweetledi

Scoville Fellowship@ScovillePF·2 Şub

Interested in working for a think tank or advocacy organization in DC on international peace and security for 6 to 9 months? Join us on Wednesday, February 4 at 3:00 PM EST to hear from Scoville Fellows and staff to learn more about the fellowship. scoville.org/overview/infor…

English

202

Paul Esau@ptesau·21 Oca

@alexcdot "FirstName Lastname" in a published paper gets me every time. "And Others" is just the cherry on that slop cupcake!

English

320

Alex Cui@alexcdot·21 Oca

Another example of "vibe citing". Authors non-existent and publication dates are off by years. These papers had to beat out 15,000 others, which got rejected. How does this happen??

English

479

56.9K

Alex Cui@alexcdot·21 Oca

Okay so, we just found that over 50 papers published at @Neurips 2025 have AI hallucinations I don't think people realize how bad the slop is right now It's not just that researchers from @GoogleDeepMind, @Meta, @MIT, @Cambridge_Uni are using AI - they allowed LLMs to generate hallucinations in their papers and didn't notice at all. It's insane that these made it through peer review👇

English

280

1.4K

6.3K

996.3K

Paul Esau@ptesau·21 Oca

I know every generation of academics moans about the diminishing quality of publications, but this crisis (caused by the intersection of AI tools and huge increases in submissions) just feels different. I'm happy to be building solutions with @alexcdot and the GPTZero team.

Alex Cui@alexcdot

English

415

Paul Esau retweetledi

Edward Tian@edward_the6·13 Oca

Hey Alex, huge fans, and kudos for an awesome study on AI detection reliability in the ecosystem HUGE CLARIFICATION we were benchmarked in this study without accounting for our three-output structure, excluding our third 'mixed' class it's like having a two party election, forgetting there's a smaller third independent party; we care about the third option because more writing on the internet is now humanized, modified, lightly edited with ai, or mixed and combined. accounting for this, our Recall is 99.3% and FPR is 0.5%, outperforming Pangram here.

English

3.2K

Paul Esau retweetledi

Alex Cui@alexcdot·15 Ara

Millions of people are now seeing "verified by GPTZero" badge - starting today. @hackernoon , one of world's most popular tech blogs, is officially verifying their 100,000s of articles and 50,000+ tech writers with @GPTZeroAI . And this is just the beginning...

English

2.5K

Paul Esau@ptesau·12 Ara

@alexcdot Another example of the "slop flood" @AishaKDown ? NeurIPS may be almost over, but ICLR is at a critical point in its submission review pipeline for 2026.

English

165

Alex Cui@alexcdot·6 Ara

There are a LOT more papers that need to be desk rejected at ICLR. Somehow, this hallucination wasn't caught. So, I went on a crazy rabbit hole and found 50 more (many are just as funny). We're approaching crisis levels Paper titles included below 👇 gptzero.me/news/iclr-2026/

ICLR 2026@iclr_conf

This paper has been desk rejected. LLM-generated papers that hallucinate references and do not report LLM usage will be desk rejected per ICLR policy (blog.iclr.cc/2025/08/26/pol…) Reviewers of other versions of this submission have been notified.

English

408

178.3K

Keşfet

@alexcdot @shadihamid @matthewkaemingk @dex_eve @Er_Woods @edward_the6 @elonmusk @BarackObama