PhD CS @ Tulane | NLP LLMs Agents | Superman & Snyder Fan
New Orleans, LA
Joined October 2022
- Tweets360
- Following187
- Followers11
- Likes241
🎉 Excited to share our paper “Can LLMs Write Faithfully? An Agent-Based Evaluation of LLM-generated Islamic Content” - accepted at @NeurIPSConf (@MuslimsinML Workshop)!
We dive into how LLMs can be made more trustworthy & explainable.
🔗 arxiv.org/abs/2510.24438
#neurips2025
Even top models still fall short in faithful content generation.
Citation hallucinations, overconfidence, and contextual gaps remain real barriers to trustworthy AI.
This work shows how agent-based evaluation can make LLM behavior more explainable.
As LLMs enter high-stakes domains, from medicine to law to cultural knowledge, faithfulness and grounding will define their credibility.
This is a small step toward measurable, explainable, and trustworthy AI.
🔗 Paper: arxiv.org/abs/2510.24438
💻 Code: github.com/AbdullahMushtaq78…
Loved this blog! Crisp examples & solid results. Connectionism feels like the way to tackle non-determinism in LLMs.
But...if a model’s wrong, won’t connectionism just make it consistently wrong? 🔥 Curious how we solve that one! @thinkymachines
Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference”
We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to prompt engineering. Here we share what we are working on and connect with the research community frequently and openly.
The name Connectionism is a throwback to an earlier era of AI; it was the name of the subfield in the 1980s that studied neural networks and their similarity to biological brains.
thinkingmachines.ai/blog/def…
At the end of the day, we obv need both: determinism + accuracy. Stable and correct outputs. That’s the real challenge!
Loved this first blog by @thinkymachines
Detailed, fact-packed, and full of relevant examples. Definitely a worthwhile read.
If we push determinism mainstream, benchmarking itself may need a rethink.
A deterministic LLM might actually be highly capable, but look bad because it doesn’t get multiple chances (reasoning models!?). Could mean more effort/major revamp is needed to measure performance fairly.
Interesting angle: most benchmarks use best@k or avg@k Non-determinism can ‘get lucky’ and score higher if any run is right. Determinism kills that variance, so, if it’s wrong, it stays wrong, which could make results look worse.
Mind-blowing just mind-blowing 🤯🔥🔥
This is what happens when you let the guy have complete creative freedom💯
@ZackSnyder @netflix
#RebelMoon
#RebelMoonInCinemasWorldWide
Abdullah Mushtaq retweeted
Part One: A Child of Fire - December 22, 2023.
Part Two: The Scargiver - April 19, 2024.
On Netflix. @RebelMoon.
Abdullah Mushtaq retweeted
Zack Snyder in person at @Gamescom
Opening Live Night for the World Premiere of the REBEL MOON teaser trailer, plus a special announcement.
Stay tuned. 😎
Adding the final touches. The @RebelMoon teaser trailer will make its global debut live @Gamescom. See you there on Tuesday, August 22. #RebelMoon
Abdullah Mushtaq retweeted
Adding the final touches. The @RebelMoon teaser trailer will make its global debut live @Gamescom. See you there on Tuesday, August 22. #RebelMoon
