Ojas Shirekar

PhD CS @TUDelft | Advised by @chiragraman | Generative Multimodal ML | Free Energy | Cognition |

[Eric W. Tramel @fujikanaeda]

when you make typos in your prompts, you should fix them for the sake of the J-space. E.g.

[Garry Kasparov @Kasparov63]

Incredible Egyptian goal is disallowed because of a foul far away, then same situation a few minutes later and goal for Argentina not disallowed! No VAR, nothing? FIFA again looks like a corrupt joke, playing favorites for stars.

[Nikita Bier @nikitabier]

@MrBeast For the love of God, make a single piece of content without financial bait

[Paul Graham @paulg]

Man do I want Paraguay to lose. Not just because they deserve to, but because I don't want to have to watch any more of them.

[Lucas Beyer (bl16) @giffmana]

Quote-reply to Rohan because I think it can be interesting to many more.

So there are two things you're missing here:

1) You're only looking at one specific instantiation of the general JEPA idea. There are many different instantiations.

2) The core JEPA idea (Joint Embedding Prediction Architecture) is to embed two "views" and predict one from the other. The views can be different augmentations, different time-steps, etc.

Crucially, prediction happens in embedding space, which contrasts to predicting in data space as done by LLMs, diffusion models, MAEs, ...

At least from the vision community, the main reason it got quite a bit of flak is that... literally everyone who was doing some self/un-supervised learning there has shared this thought already. MANY people did such models in the peak self-supervised period, which was ca 2017-2021. Then in 2022 comes Yann, slaps a new names on it, a paper with just the idea and no experiments to show for it, and goes on PR tour. That's why many didn't take it well.

The core idea, almost everyone I know agrees is worth pursuing, especially since many already were doing so. It's very reminiscent of why Stanford got flak when they introduced and arguably tried to appropriate the "Foundation Model" term.

That being said, by now foundation model has stuck and detached from Stanford, it may end up going similarly for JEPA.

[cat @a_musingcat]

it’s crazy how students run distillation attacks on their teachers and nobody does anything to stop it

[Owen Carey @owenthcarey]

Practicing LeetCode at 32 years old.

[Slovak Football 🇸🇰 @slovak_football]

I have a feeling Rotterdam will witness scenes that it hasn't had since 1940

[ThePrimeagen @ThePrimeagen]

love the new ui work YouTube!

[Ojas Shirekar @ossjs]

@tudelft

[Karthik @_karthik]

due to the heatwave the Delft municipality opened the doors of Oude Kerk for people to come and work out of

for €1/day you can come here, work the whole day while being serenaded by Gregorian chants!

i love these little, thoughtful things the Delft municipality does

[Ojas Shirekar @ossjs]

Why does Claude keep going down so often? @AnthropicAI @ClaudeDevs

[Anubis @eduantao]

Acabei de sem querer fechar uma aba que eu pretendia ler há 2 anos

[Ram @ramprasad_c]

Yes, Britain famously transferred wealth to India. When British arrived in India its share of the world economy was 4%. When they left in 1947, they had taken it to 23%, roughly equal to all of Europe combined.
On a serious note.
The Indian railways were financed entirely by bonds sold on the London Stock Exchange. British investors were guaranteed a return of 5% per annum by the colonial government. A guaranteed return, in an era when no other safe investment in Britain offered anything close. And who guaranteed those returns? Indian taxpayers. Indians paid for the construction. Indians paid the guaranteed profits to British shareholders. Indians paid for the equipment, which was manufactured exclusively in Britain and shipped to India at inflated prices. One mile of Indian railway cost twice what the same mile cost to build in Canada or Australia, because the guaranteed return meant there was no incentive to control costs. The more it cost, the more British investors and suppliers earned.

And what were these railways designed to do? Move raw materials from India’s interior to ports. Cotton from the Deccan to Bombay. Jute from Bengal to Calcutta. Coal from Bihar to wherever the Empire needed it. Tea from Assam to London’s drawing rooms. The routes connected mines and plantations to harbors. Not cities to cities. Not people to opportunities. Raw materials to ships. The Indian public’s transportation needs were, as Shashi Tharoor put it, entirely incidental.

Oh, and the railways also moved troops. Very efficiently. So that when Indians protested being looted, the British could deploy soldiers to shoot them. That was the other “infrastructure investment.”

But wait, there is more. Before the railways, India had the world’s finest textile industry. The British smashed the looms, broke the weavers’ thumbs (this is not metaphor, this is documented history), imposed tariffs on Indian cloth, and shipped raw cotton to Manchester to be manufactured into garments that were then sold back to Indians. India went from being the world’s largest textile exporter to an importer of British cloth within a generation.

The Bengal Famine of 1943 killed an estimated 3 million people. Churchill diverted food supplies from Bengal to already well-supplied British troops and European stockpiles. When informed of the famine, his response, on the record, was to ask why Gandhi had not died yet. This is the “infrastructure investor” Musk is defending.

India contributed 2.5 million soldiers to fight in two World Wars on Britain’s behalf.

So let us summarize the colonial “investment” in India. They took a 23% global economy and left it at 4%. They destroyed the world’s finest textile industry. They built railways with Indian money, for Indian resources, generating British profits. They engineered famines that killed millions. They drained an estimated $45 trillion in today’s value over 200 years.
That’s some unprofitable adventure.

[Ideogram @ideogram_ai]

Introducing Ideogram 4.0: the best open image model in the world.

Think it. Make it. Own it.

Download the weights, fine-tune on your own data, and run it on your hardware. Live on every Ideogram plan and the API today.

[Chirag Raman @chiragraman]

Excited that SimPPL has been selected for @adaption_ai's inaugural research grant program!

I’ll be co-leading this with @swapneel_mehta, exploring adaptive representation techniques for investigative agents using Arbiter's real investigative traces as a grounded testbed.

The core question: how can agents move beyond treating documents, metadata, tool outputs, and analyst feedback as separate signals, and instead build adaptive representations they can reason over?

Grateful to @sarahookr, @sudip_r0y, and the Adaption team for supporting this work!

[Ojas Shirekar @ossjs]

@alexinexxx Haha how did you end up getting a Marcus ban?

[Mathieu @miniapeur]

Every 5 neurips/icml papers PhD student interviewing for the same internships at DeepMind like

[Ojas Shirekar @ossjs]

@softmaxedx Thanks ChatGPT

[Ojas Shirekar @ossjs]

@avt_im @miniapeur Would you say the same applies for industry positions?

[Jack Lu - ICML 🇰🇷 @Jacklu_me]

Would, but NeurIPS set an adversarial deadline date this year.