Aniket Singh @ ACL'26

🤔most of the time

[Aniket Singh @ ACL'26 @Enigma__37]

Evolution of LLM landscape 🧵:

> RNN:
- First in line for sequential data
- problem of vanishing gradients because of long temporal dependencies
> LSTMs:
- LSTMs mitigated vanishing gradients using gating mechanisms, cell states and hidden state

(1/n)

[Aniket Singh @ ACL'26 @Enigma__37]

Eval master note:
1. Bad search results -> retrieval failure
2. Right data, wrong conclusions -> reasoning error
3. Invented a number -> hallucinations
4. Answered outside its remit -> scope violations

Credits: @seldo @cohere

[Aniket Singh @ ACL'26 @Enigma__37]

@Azaliamirh @StanfordAILab @achowdhery Thanks for creating this @Azaliamirh I have binged 3 lectures till now and completing all nine pretty soon. The ‘Large Language Monkey’ is pretty interesting!

[Aniket Singh @ ACL'26 @Enigma__37]

@random_walker I was honestly thinking about this last weekend. It could save not many tokens, universal tech stacks with reverse-interpreter for human understanding.

[Aniket Singh @ ACL'26 @Enigma__37]

@eyad_khrais Finally someone said it. Coming from software engineering background and working as Applied Engineer I couldn’t agree more, especially with: “software engineering trains you to think deterministically, while applied AI forces you to think probabilistically.”

[Aniket Singh @ ACL'26 @Enigma__37]

@h100envy Is this not just grid search + early stopping on the config space?

[Aniket Singh @ ACL'26 @Enigma__37]

I am 15mins walk away from the @OpenAI office. Can someone pretty please give me a office tour

[Aniket Singh @ ACL'26 @Enigma__37]

@deedydas @steipete Would love to join next time!

[Aniket Singh @ ACL'26 @Enigma__37]

@Loganullyott ✋🏻

[Aniket Singh @ ACL'26 @Enigma__37]

@benln Please organise one after 15th June as well. I really wanna attend this!

[Aniket Singh @ ACL'26 @Enigma__37]

+1. Will be working on agents eval harness. Would love to connect on post-training, RLHF and anything tech in general

[Zhuofeng Li @zhuofengli96475]

Bay Area this summer — happy to connect and chat about agentic systems, RL, and LLM post-training! ☀️

[Aniket Singh @ ACL'26 @Enigma__37]

@sama The sycophancy problem

[Aniket Singh @ ACL'26 @Enigma__37]

@PavanKumarNY Hey @PavanKumarNY, I am currently building my mind’s replica, which will help me remember everything(inspired by LLMWiki but a lot different).

Background: incoming AI research intern at Captial One, while pursuing masters from UCSD

[Aniket Singh @ ACL'26 @Enigma__37]

@tejas7c dm’ed!

[Zhijian Liu @zhijianliu_]

Super happy to see my UCSD colleague help port our DFlash to TPU. Big speedup too!! More to come.

[Aniket Singh @ ACL'26 @Enigma__37]

@endingwithali istg this has happened to me for 2 interviews. I didn’t get either of the offers but both of these HMs have pinged me on LinkedIn that they are trying to find a suitable openings because of my interview performance.

Shameless plug: actively looking for SDE/MLE summer intern

[Aniket Singh @ ACL'26 @Enigma__37]

@steipete I am tired of this. Can’t even run 3 VScode instances parallely

[Aniket Singh @ ACL'26 @Enigma__37]

@shamsharoonn did submit using https://t.co/cVQlAt96e6 as well

[Aniket Singh @ ACL'26 @Enigma__37]

@nikos1 @composio here is my attempt: https://t.co/sESUK9GKq8

[Aniket Singh @ ACL'26 @Enigma__37]

App link: https://t.co/5r5E3o4yHs
(its Groq BYOK)

[Aniket Singh @ ACL'26 @Enigma__37]

built flash mode in @TwinMind_AI's live suggestion feature, which feels like almost real-time. give it a try, link below

(ps: used a simple trick that Instagram uses for image upload)