Pranav Bharadwaj

@afterquery, prev umd

[Pranav Bharadwaj @pbokc_]

@siddhantpa1iwal @vedvedere me

[Pranav Bharadwaj @pbokc_]

@vedvedere what 😭

[Pranav Bharadwaj @pbokc_]

@tinkerapi please make context windows bigger 🙏

[Pranav Bharadwaj @pbokc_]

@_michellechoe @AfterQuery nailong plushie crucial

[Pranav Bharadwaj @pbokc_]

@AfterQuery @poolsideai 👀

[Pranav Bharadwaj @pbokc_]

@spencermateega @AfterQuery it's been fun

[AfterQuery @AfterQuery]

Congrats to the @thinkymachines team on the release of Inkling!

Inkling is a fully open-weight (Apache 2.0) 975B-A41B MoE model with native support for text, image, and audio inputs, featuring a context window of up to 1M tokens. It leads all U.S. open-weight models on @ArtificialAnlys’s Intelligence Index.

We’ve tested it internally and seen strong performance across reasoning, coding, and agentic tasks.

Inkling has day-one support on Tinker, and we're excited to start running experiments with it.

AfterQuery is proud to have supplied data that helped train the model.

[AfterQuery @AfterQuery]

Kimi K3 ranks #1 on @AfterQuery's SpreadsheetBench 2, surpassing Claude Fable 5.

An open weight model now outperforms all closed-sourced models. Read more in the Kimi K3 blog and SpreadsheetBench 2 paper linked below.

Congrats to the @kimi_moonshot team on the incredible model!

[Pranav Bharadwaj @pbokc_]

@AfterQuery @Kimi_Moonshot Crazy

[yudrew is recursing @AndrewDYu]

Excited to announce the launch of LMArcade! Play model-generated games, vote for the best, and save your favorites to play whenever.

In our private beta, we collected 33K ratings across 23 models and 7 harnesses. We also built v0 of our own harness - AQ-Gaming, which (currently) tops the model+harness leaderboard with GPT 5.5. Fable unsurprisingly tops the Model leaderboard, and the best open source model is Mimo V2.5 Pro, which tops Deepseek V4, Claude Sonnet 5, and even Gemini 3.1 Pro.

This is just the start of our work on games. We have many more extensions in store, including harness submissions, new prompts, and many more axes of game complexity. Can we find the principal components of fun? How close are we to generating good multiplayer games, or even massively multiplayer games? What will the best adaptive, generative games do? If questions like these excite you, hit us up at [email protected]

Had a blast working with Michael Evans, @n8thantran , and Andy Le to make this come to life. Come play some games with us!

[Pranav Bharadwaj @pbokc_]

@lillian_ma_ @MechanizeWork @tamaybes @MatthewJBar @EgeErdil2 @PrimeIntellect @vincentweisser @AfterQuery

[Pranav Bharadwaj @pbokc_]

@DhravyaShah Simple honeypot: put this in front of a page, if the agent passes, deny entry. Not smth I’d rely on per se, but quick and easy.

[Pranav Bharadwaj @pbokc_]

@nyrr please change your logo back

[Pranav Bharadwaj @pbokc_]

black box prompt-based testing is like unit testing with no knowledge of line coverage

[Pranav Bharadwaj @pbokc_]

A desirable interface for LLMs could knobs you can control post-training in deployment like creativity vs stability/correctness

[Pranav Bharadwaj @pbokc_]

LLMs are great at planning and bad at careful implementation - not sure why they're being used the opposite way

[Pranav Bharadwaj @pbokc_]

100% vibecoded a 3D canvas with tldraw vibes at Claude Create-a-thon @AnthropicAI

Claude 4 one-shotted a simple working version

[Pranav Bharadwaj @pbokc_]

Computer to human transition is accelerated by tech bros realizing AI will be a better computer than them

[Pranav Bharadwaj @pbokc_]

shoes should be narrow at the heel and wide at the toe box like a triangle

[Pranav Bharadwaj @pbokc_]

chatgpt just repeated the same thing like 50 times:
https://t.co/fD1daogHJW