Lightnews — Scholar-powered news

Mehmet Mars Seven ♞

@mehmetmars7.bsky.social

2.6K followers 2K following 240 posts

Lecturer @kcl-spe.bsky.social @kingscollegelondon.bsky.social Game Theory, Econ & CS, Pol-Econ, Sport Chess ♟️ Game Theory Corner at Norway Chess Studied in Istanbul -> Paris -> Bielefeld -> Maastricht https://linktr.ee/drmehmetismail Views are my own

linktr.ee

Posts Media Videos Starter Packs

Pinned

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Apr 8

I'm sure many names are missing, but even putting this list together took some time. Please feel free to suggest more names from any discipline, economics, computer science, political science, biology, mathematics, and so on, either via DM or by simply tagging them below

go.bsky.app/RjAhP7Z

1 3 21

Reposted by Mehmet Mars Seven ♞

UNET @unet.bsky.social · 8h

The UNET seminars continue!
📅 Tuesday, Oct 14, 2–3 pm
📍 Room C43 + online
Our next speaker is Mehmet Mars Seven (@mehmetmars7.bsky.social, King’s), presenting “Elo performance in sports.”
🔗 kcl.ac.uk/people/mehmet-ismail

2 2

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 2d

AI getting better at math won't displace mathematicians, it will amplify the foxes (the wonderers like Terry Tao) more than the hedgehogs (the hyper-specialized)

Reposted by Mehmet Mars Seven ♞

Hadi Hosseini @hadihoss.bsky.social · 4d

When we ask large language models to make or recommend decisions, who gets resources, opportunities, or aid, whose values are they representing?

A short🧵on our new #NeurIPS2025 paper: “Distributive Fairness in Large Language Models: Evaluating Alignment with Human Values.”

1 2 3

Reposted by Mehmet Mars Seven ♞

The Guardian @theguardian.com · Aug 26

The secrets of lost luggage auctions: I bought four bags for £100. What would I find inside?

Unclaimed suitcases were once destined for landfill. Now people are ‘suitcase gambling’ – bidding for bags and their unknown contents, and diving deep into strangers’ lives

www.theguardian.com

3 6 47

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 4d

this link works: www.nber.org/papers/w31411

Praying for Rain

Founded in 1920, the NBER is a private, non-profit, non-partisan organization dedicated to conducting economic research and to disseminating research findings among academics, public policy makers, an...

www.nber.org

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 4d

Reminded me of the perfect coincidence in Turkey this summer: a call for rain prayers after a long drought was made just as the Meteorology Service warned of three days of heavy rain across the country

www.nber.org/system/files...
h/t Arpit Gupta

1 1 6

Reposted by Mehmet Mars Seven ♞

Miranda Yaver @mirandayaver.bsky.social · 5d

We have officially entered the psychogenic fever stage of the academic year and overcommitment.

1 12

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 5d

Leaving the working here for now (results sound intuitive at first glance), but definitely looking forward to exploring it further

Arkady Konovalov @arkadykonovalov.bsky.social · 5d

📢 New (economics) preprint 📢

(with Jiaxin Yu)

Excited to share our study on the cooperation in the infinitely repeated Prisoner's Dilemma, where we use payoff manipulations and all kinds of methods to disentangle strategies, beliefs, and preferences: papers.ssrn.com/sol3/papers....

Reposted by Mehmet Mars Seven ♞

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 6d

We're finally out of stealth: percepta.ai
We're a research / engineering team working together in industries like health and logistics to ship ML tools that drastically improve productivity. If you're interested in ML and RL work that matters, come join us 😀

Percepta | A General Catalyst Transformation Company

Transforming critical institutions using applied AI. Let's harness the frontier.

percepta.ai

7 18 120

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 8d

Are top chess players good at math?

via Esports World Cup

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 9d

This stuff is real. A couple of years ago, Maastricht University got hacked, and the students and staff couldn't access the systems for some time, creating a mini chaos which was later resolved
www.bbc.co.uk/news/article...

'You'll never need to work again': Criminals offer reporter money to hack BBC

Reporter Joe Tidy was offered money if he would help cyber criminals access BBC systems.

www.bbc.co.uk

Reposted by Mehmet Mars Seven ♞

Vince Crawford @vincecrawford.bsky.social · 13d

I am pleased to announce the publication of "Expectations-based Reference-Dependence and Labor Supply: Eliciting Cabdrivers’ Expectations in the Field" (with Miao Jin, Juanjuan Meng, and Lan Yao), JEBO special issue in honor of Gary Charness. Open access here www.sciencedirect.com/science/arti...

ScienceDirect.com | Science, health and medical journals, full text articles and books.

urldefense.com

1 2 3

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 15d

Perhaps half of the papers I was asked to referee this year wouldn’t have passed the desk before LLMs, but now they look OK at first. It takes much more time to tell whether they really should pass the desk. I think this makes the editors’ job harder.

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 17d

Best of luck with the new challenge!

1 2

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 17d

Somewhat funny: before ChatGPT, I had never seen a completely made-up reference. It is the kind of error only an LLM could make.

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 19d

Colleagues in higher ed: how is your institution responding to AI use?
Are there university-wide initiatives for in-person exams, oral exams, and/or AI detection policies?
Is the pace of AI so fast that no single uni can realistically keep up?
Do you know if universities coordinating their policies?

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · 19d

The owners probably thought "nobody will read this"

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 22

4/ GPT 5 Pro: chatgpt.com/share/68a812...
Deep Think: g.co/gemini/share...

‎Gemini - Nash Equilibrium vs. Maximin Payoffs

Created with Gemini

g.co

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 22

3/ both models confused some basic game theory facts mixed with some correct statements.

1 1

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 22

2/ Result? Both models failed. Spectacularly.
GPT-5 Pro: gave a mathy polished answer… but wrong.
Gemini Deep Think: less formal… but also wrong. 😅

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 22

🧵GPT-5 Pro vs Gemini Deep Think on #gametheory round 2:

1/ After GPT pro performed better on a novel question, I tried something simpler: essentially a “database-style” lookup query. The answer is known and exists in the literature: no reasoning required.

1 1

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 21

11/ Here's the Gemini conversation that includes the first prompt
g.co/gemini/share...
The original paper: Super-Nash Performance:
doi.org/10.1111/iere...

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 21

10/ With more complex game-theory problems, both models fail. I think that the overall progress is visible, so we may be inching toward a functional mathematics / game theory "engine" but nothing close to AGI or whatever.

1 3

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 21

9/ Reliability: Neither model is reliable enough to trust without an existing solution. Everything must be checked manually. That verification price often eats up the productivity gains. (I tested many more things.)

1 3

Mehmet Mars Seven ♞ @mehmetmars7.bsky.social · Aug 21

8/ Instruction-following: GPT-5 Pro: generally decent; kept formats, echoed assumptions, flagged inconsistencies. Deep Think: drifted, forgot promised items, sometimes delivered nothing after long thinks. Once returned nothing after 1.5 days!