Physicist | Postdoc at MPI for Mathematics in Sciences, AI for Mathematics
Joined May 2019
- Tweets 999
- Following 4,579
- Followers 859
- Likes 5,319
49 Photos and videos
Pinned Tweet
Jan 30
1/ Are frontier LLMs the only path to AI math breakthroughs? I think not!
We introduce FlowBoost, A lightweight RL Flow-Matching framework that discovers new extremal geometric structures, beating AlphaEvolve with 100–1000× less compute with zero-shot geometry-aware & reward-guided generation. 🚀
arXiv: arxiv.org/pdf/2601.18005
1
13
60
16,469
Baran Hashemi retweeted
May 12
Life doesn’t wait for Earth to become “easy.”
Our new study shows that ancient life used rare Earth metals (molybdenum and tungsten) 3 billion years ago, reshaping how we think about metabolism, evolution, and the search for life beyond Earth.
nature.com/articles/s41467-0… @NASA
3
24
120
13,036
Baran Hashemi retweeted
May 8
Attenborough at 100 — A Sting Cut
We know what humans think of David Attenborough: the greatest broadcaster in TV history. But what do the animals think?
93
1,860
6,987
336,635
Baran Hashemi retweeted
May 8
Wishing a Happy Birthday to Sir David Attenborough. Thank you for the knowledge, passion, and hope you’ve passed on to all of us.
Celebrate 100 years of Sir David Attenborough with Ocean with David Attenborough on @DisneyPlus and @hulu
316
4,892
21,634
567,063
May 6
واقعا صحبت کردن با جامعه Abelian Varieties آدمو فروتن میکنه. تو یه ورکشاپ راجع به Moduli Spaces, یکی که کاندید فیلدز در آینده احتمالا باشه (از نظر سنی) داشت کارشو لکچر میداد، زمانی که فکر میکردم دیگه کارش از این خفن تر و پیچیده تر نمیتونه باشه، یهو وصلش کرد به واریته های شیمورا و من دیگه رفتم….
2
167
Baran Hashemi retweeted
Baran Hashemi: FlowBoost Reveals Phase Transitions and Spectral Structure in Finite Free Information Inequalities arxiv.org/abs/2604.11922 arxiv.org/pdf/2604.11922 arxiv.org/html/2604.11922
1
5
238
Feb 16
WTH
Pierre Lochak, Hiroaki Nakamura, Leila Schneps: The Grothendieck-Teichm\"uller group $\widehat{GT}$ acts on the genus $g$ mapping class group with $0$ or $1$ marked point arxiv.org/abs/2602.12462 arxiv.org/pdf/2602.12462 arxiv.org/html/2602.12462
1
4
365
Baran Hashemi retweeted
تاریخ یک کلانروایت خطی و غایتمند نیست. فهم این نکته ما را از چارچوبهایی که تحت تأثیر باورهای جزماندیشانه پیرامون «مسیرِ محتومِ تاریخ» شکل گرفتهاند، رها میکند. علم جادهای یکطرفه نیست که اگر مانعی چون «غزالی» نباشد، لزوماً به سرمنزلِ «گالیله» و «نیوتن» ختم شود.
امیرمحمد گمینی در کتاب «ما چگونه ما نشدیم» بهدرستی خاطرنشان میکند که نقد غزالی به طبیعیات ارسطویی نه تنها مانع علم نشد، بلکه محرک مکتب مراغه در نقد نظام بطلمیوسی بود. علم در جهان اسلام نه با «تهافت»، که در بسترِ دیالوگ با آن به تکامل رسید.
4
2
43
2,389
Baran Hashemi retweeted
Flow-based extremal mathematical structure discovery. ~ Gergely Bérczi, Baran Hashemi, Jonas Klüver. arxiv.org/abs/2601.18005v1 #AI4Math
1
2
370
Baran Hashemi retweeted
Jan 30
1/ Are frontier LLMs the only path to AI math breakthroughs? I think not!
We introduce FlowBoost, A lightweight RL Flow-Matching framework that discovers new extremal geometric structures, beating AlphaEvolve with 100–1000× less compute with zero-shot geometry-aware & reward-guided generation. 🚀
arXiv: arxiv.org/pdf/2601.18005
1
13
60
16,469
Feb 1
Here, we not surrogating the simulation, no!
We are pushing the simulation towards extremal regions of the moduli of solutions via an on-policy reward-guided fine-tuning with self-distillation.
Jan 30
1/ Are frontier LLMs the only path to AI math breakthroughs? I think not!
We introduce FlowBoost, A lightweight RL Flow-Matching framework that discovers new extremal geometric structures, beating AlphaEvolve with 100–1000× less compute with zero-shot geometry-aware & reward-guided generation. 🚀
arXiv: arxiv.org/pdf/2601.18005
2
245
Jan 31
Jan 30
1/ Are frontier LLMs the only path to AI math breakthroughs? I think not!
We introduce FlowBoost, A lightweight RL Flow-Matching framework that discovers new extremal geometric structures, beating AlphaEvolve with 100–1000× less compute with zero-shot geometry-aware & reward-guided generation. 🚀
arXiv: arxiv.org/pdf/2601.18005
218
Jan 30
This work is inspired by PatternBoost by @f_charton at @axiommathai
Jan 30
1/ Are frontier LLMs the only path to AI math breakthroughs? I think not!
We introduce FlowBoost, A lightweight RL Flow-Matching framework that discovers new extremal geometric structures, beating AlphaEvolve with 100–1000× less compute with zero-shot geometry-aware & reward-guided generation. 🚀
arXiv: arxiv.org/pdf/2601.18005
3
14
4,415
Jan 30
1/ Are frontier LLMs the only path to AI math breakthroughs? I think not!
We introduce FlowBoost, A lightweight RL Flow-Matching framework that discovers new extremal geometric structures, beating AlphaEvolve with 100–1000× less compute with zero-shot geometry-aware & reward-guided generation. 🚀
arXiv: arxiv.org/pdf/2601.18005
1
13
60
16,469
Jan 30
6/
📄 Paper: arxiv.org/pdf/2601.18005
💻 Code: github.com/berczig/FlowBoost
Built at @AarhusUni and @mpiMathSci, and @TU_Muenchen
Happy to discuss methodology, results, or potential applications!
#AI4Mathematics #AI4Science #FlowMatching
1
1
6
621
Jan 30
1
2
545