@Apple On a personal note, tomorrow will be my last day as an AIML resident. I’m truly grateful for the opportunity and the supportive environment that helped me grow and learn. I’ll be taking time off to reflect and spend with family.
Looking forward to the next chapter in my life!
🚨 New Preprint: IMPACT – Inflectional Morphology Probes Across Complex Typologies 🚨
How well do large language models (LLMs) handle morphologically rich languages? 🌍🤖
Check out our new preprint!
📰: https://t.co/utafBWgGRl
Think2SQL: Bridging the Reasoning Gap in Text-to-SQL for Small LLMs
Leveraging Reinforcement Learning with our reward mechanism, we pushed Qwen-Coder-2.5 7B to achieve performance on par with much larger LLMs (>400B params) on the BIRD dataset! 🤯
Model and paper below 👇
Given that Meta is also moving to @CommunityNotes fact checking, this work is relevant again! In our auditing of the early crowdsourcing fact checking program Birdwatch, we found both promising results and concerning manipulation risks
Merry Christmas from Saint Georges of Derdghaiya in South Lebanon.
This ancient Church, which has stood for over 300 years, was bombed and reduced to rubble by Israel.
9 innocent civilians, seeking refuge within its sacred walls, were killed in cold blood.
#Beirut now- Israel is inflicting gratuitous violence on an unarmed population. This is what the EU and the US are supporting, funding, arming, and calling self defense. No words.
Tabular DL success on benchmarks ≠ success in production. We know this first hand, trying to ship models. This motivated us to create TabReD - a new suite of 8 tabular datasets that capture real-world data characteristics overlooked by existing benchmarks. (1/N)
The devastation in Jabalia refugee camp is overwhelming. Relentless Israeli airstrikes have killed hundreds of Palestinians, including children, with entire families erased. Civilians, trapped under the rubble, have no safe place to go.