Kianté Brantley (@xkianteb) profile picture

Kianté Brantley

@xkianteb

Assistant Professor at Harvard University @KempnerInst and SEAS | Fitness enthusiast | (He/Him/His)

xkianteb.github.io
Joined May 2009
  • Tweets2.7K
  • Following1.1K
  • Followers2K
15Media
Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)Photo by Kianté Brantley (@xkianteb)
Does LLM RL post-training need to be on-policy?
10
44
11
326
114,597
Haha! All day, every day, you always gotta keep the energy high, no matter what!
Prithviraj (Raj) Ammanabrolu (@rajammanabrolu) profile picturePrithviraj (Raj) Ammanabrolu
@rajammanabroluNov 9
Many reasons why something may not work. But there's always at least one reason why it will. The only trick that actually matters is to always be high energy. F it we ball
Image from Prithviraj (Raj) Ammanabrolu (@rajammanabrolu)
1
4
565
Likewise, there’s nothing better than hanging out with friends and having fun!
Prithviraj (Raj) Ammanabrolu (@rajammanabrolu) profile picturePrithviraj (Raj) Ammanabrolu
@rajammanabroluNov 3
At least half of my motivation to go in to work everyday is to hang out with people I think are fun and smart
7
385
Kianté Brantley retweeted
Susan Murphy lab (@SusanMurphylab1) profile picture
Susan Murphy lab@SusanMurphylab1Oct 26
I am looking forward to talking about online reinforcement learning for digital health interventions in the dyadic setting in tomorrow’s Decision, Risk and Operations Seminar at the Columbia Business School!
1
16
585
Yeah, I can relate to this!
Peter Henderson (@PeterHndrsn) profile picturePeter Henderson
@PeterHndrsnOct 17
After reading AI-written slop, my brain feels so exhausted (or grossed out) that I have to cleanse the palate by reading something written before the invention of the internet.
9
485
Kianté Brantley retweeted
Harvard SEAS (@hseas) profile picture
Harvard SEAS@hseasOct 9
Five Harvard SEAS computer science graduate students—Prashanth Amireddy, Tian (Sunny) Qin, Paula Rodriguez-Diaz, Itai Shapira and Kevin Zhang—have been named 2027 Siebel Scholars. Congratulations! bit.ly/4znRkYj
Harvard SEAS Ph.D. students Prashanth Amireddy and Paula Rodriguez-Diaz

ALT Harvard SEAS Ph.D. students Prashanth Amireddy and Paula Rodriguez-Diaz

Harvard SEAS Ph.D. students Itai Shapira and Tian (Sunny) Qin

ALT Harvard SEAS Ph.D. students Itai Shapira and Tian (Sunny) Qin

Harvard SEAS student Kevin Zhang

ALT Harvard SEAS student Kevin Zhang

3
15
2,103
Kianté Brantley retweeted
David Alvarez Melis (@elmelis) profile picture
David Alvarez Melis@elmelisOct 9
Long overdue announcement: thrilled to see Sunny Qin @sunnytqin and Paula Rodriguez-Diaz @paularodrid named 2027 Siebel Scholars! 🎉 So proud of these two rising stars. seas.harvard.edu/news/five-p…
Five Ph.D. students named Siebel Scholars

Five Ph.D. students named Siebel Scholars

Scholars receive funding, access to network of scholars, researchers, and entrepreneurs

seas.harvard.edu
3
4
37
2,105
📢✨Excited to announce this year's lineup for the 2026-27 Kempner Seminar Series, which kicks off Oct. 16! Check out the full speaker lineup for what promises to be a great series on the latest research related to natural and artificial intelligence! bit.ly/4j7PeHf
Kempner Institute Announces 2026-27 Kempner Seminar Series Lineup - Kempner Institute

Kempner Institute Announces 2026-27 Kempner Seminar Series Lineup - Kempner Institute

Cambridge, MA— The Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University today announced the speakers for the 2026-27 Kempner Seminar Series, a research-level...

kempnerinstitute.harvard.edu
1
9
1,695
Kianté Brantley retweeted
David Alvarez Melis (@elmelis) profile picture
David Alvarez Melis@elmelisOct 6
My flight last night was delayed (thanks @Delta!) so I used the 'bonus' time to (finally, after lots of procrastination) put together (i.e. vibe-code) a website for my OT course at @Harvard (CS 2840, Fall '25) dmelis.github.io/otml-course….
3
8
1
104
8,031
Kianté Brantley retweeted
Shi Feng (@ihsgnef) profile picture
Shi Feng
@ihsgnefOct 5
Building Comparative Motivation Profiles with Symmetric Belief Editing will appear at NeurIPS. We refined the methodology and messaging of our performative misalignment work. We are excited to extend this to white-box auditing. by @David00845018 @1700x0 @MinTaywon @JinghuaO via @MATSprogram 1/7
Shi Feng (@ihsgnef) profile pictureShi Feng
@ihsgnefMar 18
New post: Sycophancy Towards Researchers Drives Performative Misalignment We found no clear evidence that scheming is more valid than sycophancy to explain alignment faking. 🧵
Image from Shi Feng (@ihsgnef)
1
5
27
1,714
Kianté Brantley retweeted
Yoav Artzi (@yoavartzi) profile picture
Yoav Artzi@yoavartziOct 5
Go talk to Nathan. He might even tell you how he crushed the nanogpt speedrun record for a few hours with a new LM head
Nathan Godey @COLM 🌉 (@nthngdy) profile pictureNathan Godey @COLM 🌉@nthngdyOct 5
Attending @COLM_conf this week! I'll be presenting this paper on the softmax - or gradient - bottleneck (Thursday, 4:40pm) My DMs are open if you want to reach out :)
1
17
1,477
Kianté Brantley retweeted
Nathan Godey @COLM 🌉 (@nthngdy) profile picture
Nathan Godey @COLM 🌉@nthngdyOct 5
Attending @COLM_conf this week! I'll be presenting this paper on the softmax - or gradient - bottleneck (Thursday, 4:40pm) My DMs are open if you want to reach out :)
Nathan Godey @COLM 🌉 (@nthngdy) profile pictureNathan Godey @COLM 🌉@nthngdyMar 12
🧵New paper: "Lost in Backpropagation: The LM Head is a Gradient Bottleneck" The output layer of LLMs destroys 95-99% of your training signal during backpropagation, and this significantly slows down pretraining 👇
Image from Nathan Godey @COLM 🌉 (@nthngdy)
2
8
2
101
7,603
If you're at COLM, be sure to chat with Nathan—he's fantastic!
Nathan Godey @COLM 🌉 (@nthngdy) profile pictureNathan Godey @COLM 🌉@nthngdyOct 5
Attending @COLM_conf this week! I'll be presenting this paper on the softmax - or gradient - bottleneck (Thursday, 4:40pm) My DMs are open if you want to reach out :)
17
1,158
Kianté Brantley retweeted
Bingbin Liu (@BingbinL) profile picture
Bingbin Liu@BingbinLOct 4
We'd love your opinion! The quick poll below decides what our @MOSS_workshop panel will debate. Please vote and share! 🌱
MOSS (@MOSS_workshop) profile pictureMOSS
@MOSS_workshopOct 4
Help us shape the panel at MOSS @COLM_conf ! We've put together a quick poll. Questions with the most split opinions will become panel discussion topics. Vote: form.typeform.com/to/UOyzADc… See more info on our website: sites.google.com/view/moss-c…
5
1
22
1,937
Kianté Brantley retweeted
Sasha Rush (@srush_nlp) profile picture
Sasha Rush
@srush_nlpOct 1
Sampling to Reinforce (srush.github.io/sampling-to-…) An elementary tutorial on sampling w/o randomness with a focus on gaining intuition for variance reduction as the key challenge.
20
47
664
36,867
Kianté Brantley retweeted
Andrew Drozdov (@mrdrozdov) profile picture
Andrew Drozdov
@mrdrozdovSep 22
Pumped for this!
Yacine Mahdid (@yacinelearning) profile pictureYacine Mahdid
@yacinelearningSep 22
alright guys I'll have the pleasure this week to interview @mrdrozdov from databricks on all thing retrievals like: - RL for knowledge agents - learning representation program for retrieval - place of harness in all that if you have any questions on retrieval shoot it my way
Image from Yacine Mahdid (@yacinelearning)
1
2
32
1,917
Kianté Brantley retweeted
Susan Murphy lab (@SusanMurphylab1) profile picture
Susan Murphy lab@SusanMurphylab1Sep 22
I am really looking forward to talking on online RL for digital health interventions in the dyadic setting at a joint CLIMB and Neyman seminar at UC Berkeley tomorrow! For info, see cdss.berkeley.edu/events/cli…
CLIMB + Neyman + EECS Colloquium with Susan Murphy: Online Reinforcement Learning for Digital...

CLIMB + Neyman + EECS Colloquium with Susan Murphy: Online Reinforcement Learning for Digital...

Online Reinforcement Learning for Digital Health Interventions in the Dyadic Setting We present our ongoing work on the development of an online reinforcement learning (RL) algorithm that personali...

cdss.berkeley.edu
3
23
1,165
I like this goal a lot!
David Bau (@davidbau) profile pictureDavid Bau@davidbauSep 12
If all computer scientists do is create superhuman AI, we will have failed. Our goal must be to make *humans* smarter. One of my 2026 teaching goals is to teach students to master the difficult art of using AI agents in ways the students get smarter, not dumber. Lesson ideas?
Image from David Bau (@davidbau)
10
1,143
Kianté Brantley retweeted
Susan Murphy lab (@SusanMurphylab1) profile picture
Susan Murphy lab@SusanMurphylab1Sep 4
The Murphy lab welcomes new postdoc Jixin Li, who just finished his PhD at Northeastern!
Image from Susan Murphy lab (@SusanMurphylab1)
2
18
780
I think leaning into it was the only way to go!
Jonathan Frankle (@jefrankle) profile pictureJonathan Frankle
@jefrankleAug 20
I'll know we have AGI when food delivery services start offering wine pairings for Chipotle.
Image from Jonathan Frankle (@jefrankle)
Image from Jonathan Frankle (@jefrankle)
3
513
Kianté Brantley retweeted
Yoav Artzi (@yoavartzi) profile picture
Yoav Artzi@yoavartziAug 19
New work from the lab by Nathan taking on the overall suite training process and usage. Very nice benefits in training compute, and the alignment buys significant gains in speculative decoding
Nathan Godey @COLM 🌉 (@nthngdy) profile pictureNathan Godey @COLM 🌉@nthngdyAug 19
🧵New paper: Matryoshka LM Suites Instead of training all models of a suite independently, we nest them within a single architecture and train the suite in a single run => much less compute, same performance Also allows free distillation, better speculative decoding & more 👇
Image from Nathan Godey @COLM 🌉 (@nthngdy)
1
7
42
6,777
Load more