Hrant Khachatrian @Hrant25

Training neural nets at @YerevaNN yerevann.com Yerevan, Armenia Joined November 2010

Tweets

436
Followers

453
Following

261
Likes

4K

Armen Aghajanyan @ArmenAgha

20 hours ago

I'll be giving a talk tomorrow at @aiDotEngineer (2:50pm/Room 2006), about deprecating the distinctions between VLM/VLA's and what it takes to build truly embodied foundation models.

3 2 26 3K 9

View Details

How to complain about neurips submissions I am supposed to review without violating policies? I'm losing track of the paper from the very first sentence of the abstract. For all papers I was assigned to review. #neurips #neurips2026

0 0 1 256 0

View Details

Hrant Khachatrian @Hrant25

a week ago

Baidu initializes from deepseek Just like we initialize from Qwen or Gemma

Susan Zhang @suchenzang

a week ago

this is what open-source looks like

13 93 2K 135K 561

0 0 0 145 1

View Details

Hrant Khachatrian @Hrant25

2 weeks ago

@mervenoyann @googlegemma @allen_ai what's going on with Qwen :'(

0 0 0 109 0

View Details

Hrant Khachatrian @Hrant25

2 weeks ago

Anti-oligopolist ideas from Satya Nadella, shared by Elon Musk "The last thing any of us want is a world where every company across every sector is ceding value to a few models that eat everything they see. If all the value is accrued by only a few models, the political economy will simply not tolerate it. There is no societal permission for an AI future that hollows out entire industries." "... This means the real opportunity is not in picking the best model but instead in building a learning loop on top of models where human capital and token capital compound. ..."

Elon Musk @elonmusk

2 weeks ago

Interesting

4K 6K 50K 31.7M 19K

0 0 1 107 0

View Details

Hrant Khachatrian @Hrant25

2 weeks ago

You could have names it Opus 4.9 Pro ("Pro" to justify the price) and no one would interfere. But you gave it mythological names, hyped it for months before releasing, and here we are

Anthropic @AnthropicAI

3 weeks ago

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees. The net effect of

13K 26K 88K 92.2M 24K

0 0 21 522 0

View Details

Hrant Khachatrian @Hrant25

3 weeks ago

@giffmana I'd compare with DINO and SigLIP

0 0 0 257 0

View Details

Hrant Khachatrian @Hrant25

3 weeks ago

@ClementDelangue yes, and let us know how small labs can contribute with ~100 GPUs

0 1 1 114 0

View Details

Hrant Khachatrian @Hrant25

3 weeks ago

This is so cool! Eventually all the tasks will be solved in agentic manner!

Armen Aghajanyan @ArmenAgha

3 weeks ago

1/ Today we're shipping Perceptron Agentic Detection. Describe what you want in natural language, or show one example crop, and an agent grounds it in the image. No fine-tuning, no fixed class list.

1 10 38 11K 16

0 0 0 88 0

View Details

Hrant Khachatrian @Hrant25

3 weeks ago

I still can't believe Andrej Karpathy works in this company :/

NomoreID @Hangsiin

3 weeks ago

When Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT. Anthropic estimated that this would affect approximately 0.03% of traffic.

97 157 1K 1.3M 538

3 1 80 2K 4

View Details

Hrant Khachatrian @Hrant25

a month ago

ahhhh, so cool

Suren Atoyan @suren_at

a month ago

weekend update: messing with real dogs 😀

13 42 318 29K 48

0 0 1 74 0

View Details

Hrant Khachatrian @Hrant25

a month ago

@giffmana @agarwl_ I think he means (human expert)-inspired inductive biases, not human data

0 0 1 61 0

View Details

Hrant Khachatrian @Hrant25

a month ago

Yes

Richard Sutton @RichardSSutton

a month ago

The bitter lesson in 26 words: Don’t be distracted by human knowledge, as AI has been historically. Instead focus on methods for creating knowledge that scale with computation, like search and learning.

138 982 7K 594K 3K

0 0 1 71 0

View Details

Armen Aghajanyan @ArmenAgha

2 months ago

I'm excited to finally release the fruit of the research we've been doing at Perceptron for the last 16 months: Perceptron Mk1. We've been developing multi-modal recipes from the ground up to build models that perform best in the physical world, from video understanding to embodied reasoning to robotics. Mk1 is our scaled up recipe.

Perceptron AI @perceptroninc

2 months ago

Today we're releasing Perceptron Mk1: frontier video and embodied reasoning.

29 72 434 1.3M 462

17 24 274 64K 97

View Details

Hrant Khachatrian @Hrant25

2 months ago

@yaroslavvb Human focus/attention ?!

0 0 1 64 0

View Details

Hrant Khachatrian @Hrant25

2 months ago

@HyperTechInvest That's not equivalent to using 60k GPUs

0 0 0 152 0

View Details

Hrant Khachatrian @Hrant25

2 months ago

@beffjezos basically they do that with Cursor

0 0 0 12 0

View Details

Hrant Khachatrian @Hrant25

3 months ago

@_InfoGram_ He is not a reporter. He is a congressman, Seth Moulton

0 0 1 2K 0

View Details

Hrant Khachatrian @Hrant25

3 months ago

If you are in Armenia, you can get H100 hours from YSU for participation in OpenAI Parameter Golf

YerevaNN @YerevaNN

3 months ago

Yerevan State University, YerevaNN and Eleveight AI invite AI engineers, students and researchers from Armenia to participate in OpenAI Parameter Golf! The goal is to train the best language model that fits in 16MB and runs in under 10 minutes on 8×H100 GPUs.