Dan Kondratyuk

Co-Founder at Prev. Research Scientist at @LumaLabsAI (Realtime Video World Models, Ray), @GoogleAI (VideoPoet). Let's automate research!

[Dan Kondratyuk @hyperparticle]

I've started a new company, https://t.co/Pkv6jWtamd, as a co-founder with my good friend Josh Dillon.

We're currently building teams of autonomous AI Scientists which are running experiments and discovering new knowledge.

[Dan Kondratyuk @hyperparticle]

Perhaps the marketing behind "Mythos is too powerful" was a little too convincing

[Anthropic @AnthropicAI]

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 this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.

Access to all other Claude models is not affected.

We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.

Read our full statement: https://t.co/bwn0sximKZ

[Dan Kondratyuk @hyperparticle]

@YGandelsman @RubenEVillegas @CVPR A lot of this is carried by momentum, so people will do what they are familiar with. Breaking that cycle will take a big push. As more parts of scientific engineering are automated with coding llms, maybe this can lead to more people being able to try new things.

[Dan Kondratyuk @hyperparticle]

@sumith1896 Had a great time chatting with you all

[Sumith Kulal @sumith1896]

Thanks all for coming and circling with us!

[Dan Kondratyuk @hyperparticle]

This is incredibly cool! An OS that is entirely simulated by a model. This shows glimpse of how world models might look like in the future. Any idea can transform into any app.
https://t.co/80HB8S4kE4

[Dan Kondratyuk @hyperparticle]

If we push this thought to the extreme, the ultimate form of a world model might resemble something like an operating system. It can simulate any type of computable program, running its own set of programmable neural processes to accurately represent any kind of environment.

[Fei-Fei Li @drfeifei]

https://t.co/Kt50ttQRMJ

[Dan Kondratyuk @hyperparticle]

@baaadas @LumaLabsAI It was really fun working with you Jiaming, we truly built some amazing things together. Wish you all the best on your next adventures!

[Dan Kondratyuk @hyperparticle]

Correction: Wednesday the 3rd

[Dan Kondratyuk @hyperparticle]

I'll be at @CVPR in Denver next week, presenting a talk on world models on Tuesday. If anyone wants to chat during the conference, just shoot me a message.

https://t.co/AAm96eLTjB

[Dan Kondratyuk @hyperparticle]

If you want to take part in using our systems or help us build them, consider reaching out!

[Dan Kondratyuk @hyperparticle]

I've started a new company, https://t.co/Pkv6jWtamd, as a co-founder with my good friend Josh Dillon.

We're currently building teams of autonomous AI Scientists which are running experiments and discovering new knowledge.

[Dan Kondratyuk @hyperparticle]

Our bet is simple: by 2030, most ML research won’t be done one hand-run experiment at a time, but researchers will direct large fleets of models to propose experiments, launch runs, debug failures, and come back with evidence. And that's what we're building at https://t.co/PTCss7hNEr.

[Dan Kondratyuk @hyperparticle]

The initial prototype discovered a new algorithm and wrote a paper, which you can read about in our blog: https://t.co/XBeZoU5mR5. We're now scaling this prototype up, testing it in our own research workflows to automate much of the grunt work. Even when we sleep, our system proposes new ideas, sets up experiments, compiles results, and discovers new insights which inform the next round of discovery.

[Dan Kondratyuk @hyperparticle]

Coding agents are great at writing new code, but pretty bad at deleting code. It's what inevitably leads to a lot of bloat over time. Deleting code is the halmark of a great senior engineer, i.e., one that can write the least amount of code to get the job done. In my mind that's what's missing to make them robust at building good software.

[Dan Kondratyuk @hyperparticle]

@alex_whedon Let me guess, you're already thinking of breaking the 100M token context barrier :)

[Dan Kondratyuk @hyperparticle]

@willccbb If you make compaction differentiable, maybe you can learn the optimal compaction strategy across most tasks you care about

[Dan Kondratyuk @hyperparticle]

@JiaweiYang118 Nice result. Makes me wonder what is the "ultimate form" of loss we should be optimizing

[Dan Kondratyuk @hyperparticle]

@vai_viswanathan I like to think of world models as simulations of some environment, able to represent what comes next.

Most commonly it's seen as something visual or tangible (video/3D/robotics/etc), but perhaps LLMs that simulate OS (e.g., shell envs) might also be considered as world models.

[Dan Kondratyuk @hyperparticle]

When I ask people what they mean when they are working on "World Models", I get a very different response every time. It's always fun trying to see all varied and different perspectives.

[Vai Viswanathan @vai_viswanathan]

https://t.co/pzlJpu7QyR

[Dan Kondratyuk @hyperparticle]

As with any new software, it's still going to have some rough edges.

But I put a lot of checks/manual reviews in place to make sure the code quality is to a good standard: 90+% test coverage, fully typed, docstrings that explain intent, and lots of examples.

[Dan Kondratyuk @hyperparticle]

I wanted to speedrun how fast I could OSS a complete Python package that solves a non-trivial, important job, and I managed to pull it off in about a day.

I've never felt so productive writing software, especially complete packages. The most joy I've felt in a long time.

[Dan Kondratyuk @hyperparticle]

Technically it did take another half day to polish things up, but the core implementation was fully operational in just a few hours.

Software package: https://t.co/A8siccO0Oe
Blog post: https://t.co/aRh4GwnzrU