Uljad Berdica
@uljadb99
AI @ Oxford. Stand-up Comedy. And Food I like.
Unlock real diversity in your LLM! 🚀 LLM outputs can be boring and repetitive. Today, we release Intent Factored Generation (IFG) to: - Sample conceptually diverse outputs💡 - Improve performance on math and code reasoning tasks🤔 - Get more engaging conversational agents 🤖
1/ 🕵️ Algorithm discovery could lead to huge AI breakthroughs! But what is the best way to learn or discover new algorithms? I'm so excited to share our brand new @rl_conference paper which takes a step towards answering this! 🧵
This paper scores very highly on the simplicity / ability Pareto frontier.
Unlock real diversity in your LLM! 🚀 LLM outputs can be boring and repetitive. Today, we release Intent Factored Generation (IFG) to: - Sample conceptually diverse outputs💡 - Improve performance on math and code reasoning tasks🤔 - Get more engaging conversational agents 🤖
Standard: Prompt -> response ✖️ (lacks diversity) Ours: Prompt -> intent, {prompt, intent} -> response ✔️ (high diversity and quality) Works everywhere we tried out of the box. That's it. Use it if you are not doing so already.
Unlock the Hidden Diversity in Your Language Model. In our new paper, Intent Factored Generation (IFG), we propose an inference time method to increase the diversity of generations from LLMs. IFG leads to improvements in searching for solutions to maths and code problems. (1/6)
It’s truly a privilege to be able to wake up every morning, see where the latest intelligence frontier is, and help push it a little further.
Last #runconference at #ICML2025 (at least for me). Was great running with new and old friends this week, and looking forward to another set of runs at @RL_Conference in August!
Third #runconference ! Last one (for me) tomorrow, join us at 7am in tourist info booth just outside convention center!
Really good point!
pass@1 becomes meaningless with composite systems or models like o1 that can perform pass@k internally: I think we should use or rather we'll end up with pass@flops of test-time compute or pass@time for time spent at test ☺️ this will allow for better comparisons
pass@1 becomes meaningless with composite systems or models like o1 that can perform pass@k internally: I think we should use or rather we'll end up with pass@flops of test-time compute or pass@time for time spent at test ☺️ this will allow for better comparisons
.@OpenAI o1 has started rolling out to the API!
Shame on you @NeurIPSConf, i've got tons of shaming emails. @COLM_conf was amazing, it used AI to improve reviewing and it was great! @ReviewAcl spits blood to improve quality and reduce load. Here is the tale of this round of neurips reviews, of automatic threats and disrespect
Exciting tool! Hyperparameters are there to be conquered.
🚀 Excited to announce Hyperoptax, a library for parallel hyperparameter tuning in JAX. Implements Grid, Random, and Bayesian search in pure JAX so that you can rapidly search across parameter configurations in parallel ‖. 📦 pip install hyperoptax github.com/TheodoreWolf/h…
SoReL and TOReL achieved: - Accurate regret estimation using only offline data 📊 - Competitive with the best online hyperparameter tuning methods 🏆 - All without requiring online interactions in the real environment 🛡️ Great work led by @ClarisseWibault and Mattie Fellows!
How can we bypass the need for online hyper-parameter tuning in offline RL? @FLAIR_Ox is introducing two fully offline algorithms: SOReL, for accurate offline regret approximation, and TOReL, for offline hyper-parameter tuning! arxiv.org/html/2505.2244…
How can we bypass the need for online hyper-parameter tuning in offline RL? @FLAIR_Ox is introducing two fully offline algorithms: SOReL, for accurate offline regret approximation, and TOReL, for offline hyper-parameter tuning! arxiv.org/html/2505.2244…