Mohd Sanad Zaki Rizvi

M.S. Thesis @ U. of Edinburgh | Previously: Google Research India, Microsoft Research India | #nlproc #NLU #AI | Thoughts my own

[Mohd Sanad Zaki Rizvi @sanad_maker]

Our paper - "GCM: A Toolkit for Generating Synthetic Code-mixed Text" has been accepted at the 16th Conference of European Chapter of the Association for Computational Linguistics @eaclmeeting in the demo paper track.

[Mohd Sanad Zaki Rizvi @sanad_maker]

@Im_IrushiK I am guessing here but AWS like IaaS would be existing then. And Perl and PHP has been used for making websites and servers even Facebook started with PHP. Why do you need Docker or Kubernetes when you can build your own IaaS? And migrations/deployment can be done manually.

[Mohd Sanad Zaki Rizvi @sanad_maker]

LLMs are fluent today—but only by conquering hallucinations can they become truly trustworthy.

In my latest blog, I explore why this challenge matters, the most promising solutions, and what’s next for LLMs.

https://t.co/GuHZQHoTBT

#AI #LLM #NLP #TrustworthyAI #ChatGPT

[Mohd Sanad Zaki Rizvi @sanad_maker]

Huge shout-out to my incredible collaborators Alex Murphy, Aden Haussmann, Ping Nie, Guifu Liu, @aryopg , & @PMinervini at @EdinburghNLP and @EdinburghUni! 🙌

This is a big step toward grounded, hallucination-free LLMs!

Check out our research & let’s discuss!

n\n

[Mohd Sanad Zaki Rizvi @sanad_maker]

🚀 New ArXiv paper alert!

By combining agentic frameworks (ReAct) with smart decoders (DeCoRe, DoLa, CAD), we boost factual accuracy in complex reasoning tasks —reducing those annoying hallucinations! 🔥

🔗 Paper: https://t.co/4spoFrSOXH

1\n

[Mohd Sanad Zaki Rizvi @sanad_maker]

🔍 Key takeaways:
✅ Faithful decoding + ReAct = better answer accuracy & recall across datasets.
✅ Thought-Action-Observation flows improve with these decoders.
✅ DoLa decoder on HotpotQA? Answer F1 ↑ 13.1 points! 📈
Better decoding = more reliable LLMs! 💡
2\n

[Mohd Sanad Zaki Rizvi @sanad_maker]

Definitely true!

[Andrej Karpathy @karpathy]

Probably not what you want to hear but docs 😅. Actual real life examples. Better and more comprehensive kwarg docs. More helpful links to actual code not just wrapper of wrapper of wrapper code. Example code of larger apps showing best practices (style of torch titan, nanoGPT or etc). Helpful historical context if any, possibly links to useful issues. In process of my zero to hero videos I think I’ve come by ~10 examples of bad, incomplete, unhelpful or misleading docs where you just kinda have to know somehow.

[Mohd Sanad Zaki Rizvi @sanad_maker]

@shaily99 @EdinburghNLP Thank you so much!

[Mohd Sanad Zaki Rizvi @sanad_maker]

I’m thrilled to share that I’ve graduated with Distinction from the M.S. Speech & Language Processing (NLP & AI) program at The University of Edinburgh @EdinburghNLP (1/n)

[Mohd Sanad Zaki Rizvi @sanad_maker]

@wrackspurt9 @EdinburghNLP Thanks Abhirut bhai!

[Mohd Sanad Zaki Rizvi @sanad_maker]

@somakaditya @Roprajo @EdinburghNLP Thanks!

[Mohd Sanad Zaki Rizvi @sanad_maker]

@Roprajo @EdinburghNLP Thank you so much!

[Mohd Sanad Zaki Rizvi @sanad_maker]

@tjkalyan7 @EdinburghNLP Thank you!

[Mohd Sanad Zaki Rizvi @sanad_maker]

@rudrasohan @EdinburghNLP Thank you!

[Mohd Sanad Zaki Rizvi @sanad_maker]

@PMinervini @partha_p_t @monojitchou @fooobar @lai_cat_lai @EdinburghNLP @JangraAnubhav @shivanikapania @deepaknathani11 @BigAmeya @shaily99 @JainUmangi @Roprajo Thanks @EdinburghUni @EdUniStudents for giving me this great opportunity. (n/n)

[Mohd Sanad Zaki Rizvi @sanad_maker]

@PMinervini @partha_p_t @monojitchou @fooobar @lai_cat_lai And all the amazing people @EdinburghNLP, my friends @JangraAnubhav @shivanikapania @deepaknathani11 @BigAmeya @shaily99 @JainUmangi @Roprajo and more friends for being my rock like support entire time. You all know who you are! (6/n)

[Mohd Sanad Zaki Rizvi @sanad_maker]

@BigAmeya @EdinburghNLP Thanks buddy!