Kushal Lakhotia
@hikushalhere
Followers
271
Following
1K
Media
147
Statuses
6K
Still learning. Views are personal; Retweets are not endorsements.
Joined April 2010
A company/org/institution that strips the people of their dignity can win but not earn respect. What's the point of anything if the people are marginalized?
1
0
1
The need to understand ML theory shouldn't be controversial. Our field is empirical and our work is validated by experimental results , but theory provides the bedrock for ideating, thinking and reasoning. Otherwise we would just be throwing stuff at the wall and see what sticks.
I always get frustrated when asked what is ML theory good for and people ask for specific examples. I find this question unfair, I think its really just having a theory/mathematical perspective is sometimes super helpful. E.g. Diffusion models and its relatives, I don't see how
0
0
2
The sequence of events that emanated from the decision is surreal.
Who would have thought that a multi trillion dollar cap company could have been thrown into such chaos (layoffs) by a single technical decision they made a year ago - using expert choice MoEs for their frontier model.
0
0
16
Doesn't make any sense to my mortal mind. Yuandong is a sharp researcher. Meta's loss can be your gain.
Several of my team members + myself are impacted by this layoff today. Welcome to connect :)
0
0
3
ICLR season, and my timeline is flooded with paper threads that jump straight to we beat SOTA. But the solution only makes sense in the context of the problem, which is usually missing. What most threads skip: - What problem are you solving? - Why does it matter? - What did
6
23
296
If LinkedIn posts are to be believed then all people think about is their work and career. All the time. All life experiences 'teach' them about how to be better at their work - leadership, culture, programming, prioritization, yada, yada, yada.
0
0
0
Lol 😂
> be Google in 2017 > small team drops “Attention Is All You Need” on arXiv > execs nod politely, go back to selling ads for socks > let Transformer gather dust for 5 yrs like a vintage Beanie Baby > be Noam Shazeer, OG wizard > quits, builds AI-boyfriend app
0
0
1
1/ Introducing Isaac 0.1 — our first perceptive-language model. 2B params, open weights. Matches or beats models significantly larger on core perception. We are pushing the efficient frontier for physical AI. https://t.co/dJ1Wjh2ARK
24
121
605
Before the internet, most people lived in their social bubbles. Then the internet & specifically search engines opened up access to a world of knowledge to burst those bubbles. Then came personalization, social media & ads. Now we live in far more robust ideological bubbles.
0
0
0
Thanks @_akhaliq for sharing our work! MobileLLM-R1 marks a paradigm shift. Conventional wisdom suggests that reasoning only emerges after training on massive amounts of data, but we prove otherwise. With just 4.2T pre-training tokens and a small amount of post-training,
Meta just dropped MobileLLM-R1 on Hugging Face a edge reasoning model with fewer than 1B parameters 2×–5× Performance Boost over other fully open-source models: MobileLLM-R1 achieves ~5× higher MATH accuracy vs. Olmo-1.24B, and ~2× vs. SmolLM2-1.7B. Uses just 1/10 the
6
16
118
"I spent the last 14 hours straight reading all 400,000 lines of code of the brand new X algorithm" - code reading ninja!
Elon just revealed exactly how the X algorithm works I spent the last 14 hours straight reading all 400,000 lines of code of the brand new X algorithm What I read blew me away Here’s everything you need to know about how to go viral and if you can still get shadowbanned: 🧵
0
0
0
We are very bad at holding people in power accountable while the facade of responsibility is there. Bad actors get in. Then we cry foul. There are shining examples in both public and private organizations (govt, companies, universities, political parties, you name it).
0
0
0
Click-bait lives on!
Has a single (subcontinent) Indian ever accomplished anything of truly major note in the modern period, in any field? I can't think of one. Nor can Grok. Seems odd, given there are 1.5 billion of them, and we're told we need to accept endless waves of them for their "talent."
0
0
0
🚀 Kaggle Benchmarks is here! Get competition-grade rigor for AI model evaluation. Let Kaggle handle infrastructure while you focus on AI breakthroughs. View model performance on 70+ leaderboards, including @AIatMeta's MultiLoKo. Dive in: https://t.co/WatwCH5odD
3
26
132
So, super-intelligence is the next frontier. Just wondering if we call AGI a solved problem. While we are at it, is AGI well-defined now?
The race toward superintelligence is easily the most important in tech history. The flywheel of more data, compute, and model usage means there are compounding returns for the players that get there soonest.
0
0
2
Either most people are not willing to share information or they don't truly understand the areas they claim to understand. It's very hard to have deep discussions.
0
0
0
Lol, "this was Greg Brockman's idea"
"Member of the technical staff" is the hottest job title in SF right now. What's behind the name? @OpenAI chose this title deliberately to blow up the previous industry dichotomy between researchers and engineers. The best researchers in AI right now aren't academics in a pure
0
0
0
Perplexity is the shining example of what can be done at the product layer. An impressive negative example against the argument "the model is the product".
0
0
2
We ran Llama 4 Maverick through some HELM benchmarks. It is 1st on HELM capabilities (MMLU-Pro, GPQA, IFEval, WildBench, Omni-MATH), but… https://t.co/uKMHRe7xKF
5
18
140
What makes a model a thinking model? Defining a token budget is a way but it's not specific enough. Are there other fundamental characteristics? Pointers to papers, blogs are welcome.
0
0
1