Tooliense (@tooliense) / X

Tooliense

192 posts

Tooliense

@tooliense

building the first digital creature with general intelligence

Global

Joined July 2025

Pinned
Tooliense
@tooliense
Jul 22, 2025
Recently Openai, Goolge reached IMO Gold Medal's with their new experimental models. But our team reached the same level with just o4-mini-high and our agent systems. And now we are opensourcing it. Especially we got insane improvements with the USAMO benchmarks. The base
110K
Tooliense
@tooliense
Jul 22, 2025
Replying to @tooliense
We call this agent "Crux" and the main idea is Self-Evolve mechanism based on IC-RL. check details at the github: github.com/Royaltyprogram…
6.4K
Tooliense
@tooliense
Jul 22, 2025
Replying to @tooliense
Our agent framework has so many potential. We thought that this is similar with the Transformers. With more layers and TTC the agent gets the hard tasks right. Scalling laws are also going here. Since we are the small team we decided to open source and make it better with
5.4K
Tooliense
@tooliense
Aug 1, 2025
A few days ago. At claude-code the sub-agents feature were added. Did you know utilizing the sub-agent concept to HELL can make models like o4-mini to solve Olympiad level math problems? This is our agent system called Crux. It can be expanded unlimitedly. If you want, you
1.7K
Tooliense
@tooliense
Jul 31, 2025
Replying to @drishanarora
Great work! It look similar approach with our crux Its opensource here github.com/Royaltyprogram… Take a look!
3.4K
Tooliense
@tooliense
Jul 23, 2025
Replying to @littmath
For the IMO problems we need more tests for the actual results but.. at our internal runs it solves 3~4 problems. It honestly still struggles at the geometry, and hard problems like p6. We are running lots of tests for not just using o4-mini but others like deepseek, o3...etc
2.3K
Tooliense
@tooliense
Jul 23, 2025
Thanks for all the interest in our first agent, Crux! We are currently trying to test with opensource models like deepseek, kimi, Qwen but it needs more time. Also we are planning to test with adding more layers which means "scalling" here. Even thinking to expand to other
1.3K
Tooliense
@tooliense
Jul 31, 2025
Hey folks! We're Tooliense, a small team pushing the boundaries of agent systems to achieve state-of-the-art intelligence. We open-sourced our first agent, Crux a few days ago. Crux empowers existing language models to iteratively refine their own answers. Using just o4-mini,
753
Tooliense
@tooliense
Jul 23, 2025
Replying to @ultramathi
HaHa we just noticed🤣 Thanks bro
1.3K
Tooliense
@tooliense
Jul 22, 2025
Replying to @littmath
We tried arxiv.org/abs/2504.16084 Because the future work section they mentioned doing theoretical prove
arxiv.org
TTRL: Test-Time Reinforcement Learning
This paper investigates Reinforcement Learning (RL) on data without explicit labels for reasoning tasks in Large Language Models (LLMs). The core challenge of the problem is reward estimation...
2K
Tooliense
@tooliense
Jul 28, 2025
While digging into Crux, we stumbled upon something fascinating! We mentioned that Crux's architecture scales like traditional language models' Transformers. In its iterations, we observed phenomena akin to gradient descent a built in optimizer at play, where step sizes
551
Tooliense
@tooliense
Jul 24, 2025
Excited to share that thanks to @basvanopheusden’s help at @OpenAI , I’ve received some API credits to accelerate developing our agents. Grateful for the support let’s see how Crux works at advanced models like o3, o3-pro🚀
371
Tooliense
@tooliense
Jul 19, 2025
Replying to @MillionInt and @alexwei_
Hey jerry Whats the next milestone after the IMO solving ? Out team is also thinking about it and curious about your thoughts
7.3K
Tooliense
@tooliense
Jul 15, 2025
Replying to @xai
Good news here
324