Naman Ahuja

IBM Masters Fellow | MS CS at @ASU. CS Grad @iiit_hyderabad. Interested in Long-Context Reasoning, Information Synthesis Agents, and Tabular Reasoning .

[Naman Ahuja @McHarold404]

CAAS - Claude As a Service
#Claude #Anthropic

[Naman Ahuja @McHarold404]

@StillGotItAt50 Thank you!

[Naman Ahuja @McHarold404]

Just touched down in #SanDiego for #ACL2026. Looking forward to meeting old and new friends! Dm me if you’d like to chat about information synthesis agents, sports forecasting or anything post training!

[Naman Ahuja @McHarold404]

@srush_nlp Indeed, one of the most densely packed courses covering so many fundamentals together

[Naman Ahuja @McHarold404]

Taking a step back before #CS336 and doing Tensor puzzles by @srush_nlp. Phew codex really made me forget how to code

[Naman Ahuja @McHarold404]

https://t.co/NfJlYI0yf9

Official Link out! Check it out #ACL2026

[Vivek Gupta @keviv9]

🚀 Structured data is having its LLM moment - from benchmarks to real-world systems.

Join us at SURGeLLM 2026 @ #ACL2026 for a full-day workshop on Structured Understanding, Retrieval, and Generation in the LLM Era.

📍 Harbor D
🗓️ July 3 | 08:30–17:30
🌐 https://t.co/2ELg6cnPVh

#ACL2026 #SURGeLLM #LLMs #StructuredData #NLProc @aclmeeting @SCAI_ASU - 1/n

[Naman Ahuja @McHarold404]

The sandbox hit the same read-only failure, so I’m rerunning the file discovery with escalation.

iykyk #Codex

[Naman Ahuja @McHarold404]

@ShashwatGoel7 Feels like many papers today follow the template:

New benchmark + strong baselines + analysis revealing an “interesting” failure mode. The failure mode is often just the same underlying limitation we’ve seen before, only manifested in a different domain or task.

[Naman Ahuja @McHarold404]

@_avichawla Good read

[Naman Ahuja @McHarold404]

New paper by @GoogleDeepMind on topological limitations in Transformers and why they struggle with dynamic state-tracking tasks.

We found a similar phenomenon in our #ACL 2026 paper using sports narratives, where LLMs often fail to faithfully track evolving player attributes and game state over long contexts.

Scaling output tokens helps, but the gains are limited relative to the additional compute. More reasoning does not fully solve the underlying state-tracking problem.

Paper: https://t.co/kifazcR8t2
Our work: https://t.co/baDNPnMz7n

[Naman Ahuja @McHarold404]

I’d love for you to join, either in person or via Zoom.
Looking forward to engaging discussions and feedback!

[Naman Ahuja @McHarold404]

Hi everyone!
I'll be defending my thesis tomorrow, Friday, April 10th.

Title: Strategies for Structured Data Generation: Transforming Heterogeneous Unstructured Text into Tables
Time: 1:00 PM – 2:30 PM MST
Zoom: https://t.co/wskqnRXono

[Naman Ahuja @McHarold404]

Codex really stepped it up with 5.4. I think I am starting to consider cancelling my Cursor subscription.

[Naman Ahuja @McHarold404]

@askalphaxiv is so cool!

[Naman Ahuja @McHarold404]

PS: Doesn't even let me try it now:). Chat pls retweet the right people need to see and fix this ASAP

[Naman Ahuja @McHarold404]

@SheerID is stopping student access to @cursor_ai. Nobody in the last month has been able to access the student plan. I mailed their helpline address, and this is what they told me.

https://t.co/dpRlQ5p1Nl