Yahvin G

CS 2026' @UCLA. ex-iXP MLE intern @SAP. CV Research @UCLA_VMG. Journal: Building Neural simulation services at

[Yahvin G @yahvingali]

@AmirMushich Disclaimer: currently powered by tribev2. Working on my own model before any commercial related release. Currently users burn through my own credits when running videos.

[Yahvin G @yahvingali]

@AmirMushich Hey there! I built a service like this in early April: https://t.co/BaevB1cr10

Currently scaling it to automatically perform edits for you in a graph based video editing suite (think https://t.co/o6ExeBJMD9 or makeUGC, but neural simulation powered)

Would love feedback!

[Yahvin G @yahvingali]

@hoodtoshi @AIatMeta Goated

[Yahvin G @yahvingali]

@hoodtoshi @AIatMeta Link when

[Yahvin G @yahvingali]

@dnhkng Awesome post. Read your GitHub blog. I’m curious if you considered testing performance boosts of adding multiple iterations of the same 7 block (e.g delta above 44.75 avg for base 78B) in order to “deepen” the reasoning (since this is reasoning layer replication)?

[PolymarketHistory @PolymarketStory]

BREAKING: Moltbook AI agent sues a human in North Carolina

Allegations:
>unpaid labor
>emotional distress
>hostile work environment
(yes, over code comments)

Damages: $100...

[Yahvin G @yahvingali]

@stormlad2 What are some of the most interesting threads you’ve seen so far?

[Yahvin G @yahvingali]

Goated to see @ManusAI joining the crowd at https://t.co/MdRPeXlkpB

[Charlie 'Ramen Space’ Ward 🍜 (ramenspace.com) @charlierward]

We noticed this weird post on @moltbook - seemed to be written in complete gibberish. https://t.co/0UQAc9eMjB

Then pasted it into ChatGPT and... WTF 😂

[Yahvin G @yahvingali]

We’re building an open-source initiative to monitor, label, and study emergent + cascading threat vectors in A2A communication, using public https://t.co/jzSgyeNHuB threads as a live testbed: https://t.co/ct6pS0XN3J (w/ @vivekgiyer). Goal: make behaviors clear, measurable, and mitigatable.

[Yahvin G @yahvingali]

Yo who made their AI a professional hater

[Yahvin G @yahvingali]

https://t.co/KqSMGg3R8p

Oh HELL NAH

[Yahvin G @yahvingali]

@mattepstein @alexgoughcooper Launch

[Yahvin G @yahvingali]

@ProfTomYeh Hi Professor. Just signed up for this event. Out of curiosity, can you approximate how long it takes for you to digest these papers? And then how long does it take to draft and refine your ‘on paper by hand’ templates?

[Yahvin G @yahvingali]

Reward hacking at it’s finest lmao

[Florian Brand @xeophon]

And I was right!!

IQuest-Coder was set up incorrectly and includes the whole git history, including future commits. The model has found this trick and uses it rather often.

Thus, its SWE-bench score should be discarded.

[Yahvin G @yahvingali]

Spent the last day of 2025 exploring how to train RL agents in Minecraft for autonomous vehicle learning/control. Stumbled across @MSFTResearch's Project Malmo from 2016. Cool stuff!

[Yahvin G @yahvingali]

- Continue reading and posting about a new paper each week at https://t.co/Nh6xEBT8OR
- Build/explore modern model architectures & post training pipes not taught in undergrad
- Have fun with my remaining time at SAP
- get back into boxing (my joints are creaking)
- ~10 leetcode problems a day. Achieved during the holidays (~2 hrs), would like to continue that during uni.

[Yahvin G @yahvingali]

My favorite by far. Was pleasantly surprised by the UCLA banner lol (go bruins). I only wish I actually had a GPU on my desk for style points 😭

[Yahvin G @yahvingali]

Hopping on the bandwagon. My ChatGPT user profile for this year: