Hadi Khalaf (@hskhalaf) profile picture

phd agent @harvard, graduate fellow @kempnerinst, i work on ai alignment

🇺🇸🇱🇧
Joined February 2025
Photo by Hadi Khalaf (@hskhalaf)Photo by Hadi Khalaf (@hskhalaf)Photo by Hadi Khalaf (@hskhalaf)Photo by Hadi Khalaf (@hskhalaf)Photo by Hadi Khalaf (@hskhalaf)
Hadi Khalaf retweeted
When agreement with the user's framing is overrepresented among high-reward outputs relative to responses that correct the user's premise, optimization makes the model more agreeable. A (particular) annotator bias is enough for a Bradley-Terry reward to satisfy this condition arxiv.org/abs/2602.01002
1
1
1
88
Hadi Khalaf retweeted
On Wednesday morning (#4116) @hskhalaf and I are also presenting our work on AI evaluations using Maximal Lotteries! arxiv.org/abs/2602.21297
1
1
78
Spotlight at NeurIPS! See you in San Diego 😄 You can check out our paper here: arxiv.org/pdf/2506.19248
How can we improve LLMs without any additional training? 🤔 The standard playbook is using Best-of-N: generate N responses ➡️ use a reward model to score them ➡️ pick the best 🏆 More responses = better results... right? Well, not exactly. You might be reward hacking! Instead, you should hedge! 🎯
3
646
The "shady" privacy formula in question is differential privacy...
The 2020 Census was a fraud. The Biden admin used a shady “privacy” formula that scrambled the data and miscounted 14 states. It included illegal immigrants and handed Democrats extra seats. Americans deserve a fair count and I’m fighting to fix it.
322
why does gpt 5 sound like a tech bro
147
I judge llms by how bayesian they are #1 gemini 2.5 pro (channeling bayes himself) #2 gpt 5 #3 o3 #4 gpt 4o Please stop the bayesian propaganda
2
312
I wrote a fun little article about all the ways to dodge the need for real-world robot data. I think it has a cute title. sergeylevine.substack.com/p/…
430
How can we improve LLMs without any additional training? 🤔 The standard playbook is using Best-of-N: generate N responses ➡️ use a reward model to score them ➡️ pick the best 🏆 More responses = better results... right? Well, not exactly. You might be reward hacking! Instead, you should hedge! 🎯
1
1
2
9
1,653
Hedging significantly mitigates reward hacking and achieves superior performance with minimal computational overhead in our experiments!
1
3
235
Hadi Khalaf retweeted
It is critical for scientific integrity that we trust our measure of progress. The @lmarena_ai has become the go-to evaluation for AI progress. Our release today demonstrates the difficulty in maintaining fair evaluations on @lmarena_ai, despite best intentions.
26
135
40
717
284,347
Hadi Khalaf retweeted
Ever wonder how LLM developers choose their pretraining data? It’s not guesswork— all AI labs create small-scale models as experiments, but the models and their data are rarely shared. DataDecide opens up the process: 1,050 models, 30k checkpoints, 25 datasets & 10 benchmarks 🧵
11
105
10
646
74,081
Happy to share we received best paper at NENLP workshop at Yale 🥳🥳! tldr: Current alignment methods give excessive discretion to annotators in defining what good behavior means. This means we don't know what we are aligning to ‼️ We formalize discretion in alignment and propose mechanisms for data curators & model developers to monitor for it. Paper link below ⬇️
3
4
1
23
3,121
gemini is crazy at coding, insanely crazy good
125
Does anyone like arxiv html? I immediately switch to the pdf view
1
125
Yes 👍🏼
day in the life of an AI PhD in 2025 > wake up > new research idea (5 minutes) > kick off related work search w/ Deep Research (15 minutes) > set up 4 instances of Claude Code to start project (30 minutes) > get o1 started on proof for paper (5 minutes) > play tennis (6 hours)
246