Haowei Lin Profile
Haowei Lin

@AndyLin2001

Followers
161
Following
53
Media
27
Statuses
66

Undergrad and PhD student @PKU1898, Continual Learning | LLM Agent | AI4Science.

Beijing, China
Joined May 2023
Don't wanna be here? Send us removal request.
@AndyLin2001
Haowei Lin
1 month
Can LLMs discover scaling laws? The answer is YES. And they can do it better than humans. https://t.co/c0aJRv2k9I #AI #LLM #ScalingLaws #AgenticAI #ML
1
1
9
@AndyLin2001
Haowei Lin
1 month
SLDAgent is based on a simplified version of OpenEvolve @asankhaya ( https://t.co/w1rLcKLA00) with domain specific prompts, evaluator, and initial program. We also think SLDBench will be a good testbed for both evolving-based agents and general coding agents.
Tweet card summary image
github.com
Open-source implementation of AlphaEvolve. Contribute to algorithmicsuperintelligence/openevolve development by creating an account on GitHub.
0
0
3
@AndyLin2001
Haowei Lin
1 month
This shows OpenEvolve remains highly competitive, even with an OSS model. Are Shinka's results the current SOTA? I only run the experiments for 6 times and find this impressive one.
1
0
1
@AndyLin2001
Haowei Lin
1 month
Impressive results from our latest run: OpenEvolve + GPT-OSS-120B with single-phase evolution reached a score of 2.6359830849. This not only tops AlphaEvolve (2.635863) but is within 1e-8 of ShinkaEvolve resultsโ€”all without adding any new features. @SakanaAILabs @asankhaya
1
2
2
@AndyLin2001
Haowei Lin
1 month
An example of SLDAgent discovered law v.s. human law: conceptually sound and better fitness. This is really a newly discovered knowledge / science.
1
0
0
@AndyLin2001
Haowei Lin
1 month
This isn't just theoretical! We showed its practical utility in two key applications: 1. Hyperparameter Tuning: deriving near-optimal learning rates and batch sizes for pre-training. 2. Model Selection: Perfectly identifying the best pre-trained LLM for SFT across 3 datasets.
2
0
0
@AndyLin2001
Haowei Lin
1 month
The results are stunning. Paired with GPT-5, SLDAgent beats the human-expert-derived laws on all tasks in SLDBench. The discovered laws aren't just a better fit; they are more conceptually sound, capturing the underlying dynamics in a more principled way than the human versions.
1
0
0
@AndyLin2001
Haowei Lin
1 month
Our key innovation is SLDAgent, an evolution-based agent that mimics the scientific process. It starts with a simple guess, then iteratively mutates and refines both the mathematical formula and the method to fit it to data, constantly improving its "hypothesis."
1
0
0
@AndyLin2001
Haowei Lin
1 month
Our key innovation is SLDAgent, an evolution-based agent that mimics the scientific process. It starts with a simple guess, then iteratively mutates and refines both the mathematical formula and the method to fit it to data, constantly improving its "hypothesis."
0
0
0
@AndyLin2001
Haowei Lin
1 month
We built SLDBench, a testbed for this task, curated from over 5,000 real-world LLM experiments across 7 tasks This isn't just about rediscovering known formulas; it's an open-ended challenge where even human experts don't know the "perfect" answer. https://t.co/rkLSuLcygG
1
0
1
@AndyLin2001
Haowei Lin
1 month
For decades, finding scaling laws has been a slow, manual process for human experts. It's a major bottleneck for developing foundation models, requiring tons of trial-and-error to predict how a model will behave when you spend millions to train it.
2
0
0
@AndyLin2001
Haowei Lin
2 months
Check the original website here: https://t.co/NUhBavyGrb Thanks @ori_press and the team for the great work!
Tweet card summary image
algotune.io
Can Language Models Speed Up General-Purpose Numerical Programs?
0
0
3
@AndyLin2001
Haowei Lin
2 months
Excited to announce that I've adapted AlgoTune for both OpenHands and Terminal Bench! It's a fast, unbounded benchmark perfect for evaluating AI agents, offering a great alternative to slower suites like SWE/Kaggle tasks. Check it out: https://t.co/bmwOQUoTxa #Agent #Benchmark
1
1
8
@AndyLin2001
Haowei Lin
3 months
Recent updates: 1. EvoSLD is available on arXiv: https://t.co/Mx9r65erzB 2. We tested EvoSLD on StepFun Law ( https://t.co/aqW0io2SBj), predicting optimal learning rate (lr*) and batch size (bs*) using pre-training tokens (D) and model parameters (N). EvoSLD discovered a law:
Tweet card summary image
arxiv.org
Discovering scaling laws for predicting model performance at scale is a fundamental and open-ended challenge, mostly reliant on slow, case specific human experimentation. To investigate the...
@AndyLin2001
Haowei Lin
4 months
Thrilled to announce EvoSLD: A new framework for Automated Scaling Law Discovery! ๐Ÿš€ Scaling laws are critical for predicting AI model performance in research or industry, but finding them is a complex, manual process. EvoSLD uses Large Language Models in a novel evolutionary
0
2
12
@AndyLin2001
Haowei Lin
4 months
Thrilled to announce EvoSLD: A new framework for Automated Scaling Law Discovery! ๐Ÿš€ Scaling laws are critical for predicting AI model performance in research or industry, but finding them is a complex, manual process. EvoSLD uses Large Language Models in a novel evolutionary
1
4
16
@du_yilun
Yilun Du
6 months
Excited to share work on using classical search approaches to scale inference in diffusion models! We show how global graph search algorithms (BFS, DFS) and local search can be used to improve generation performance across domains such as image generation, planning, and RL!
7
59
500
@helloiamleonie
Leonie
6 months
Level 1: Keyword-based retrieval Level 2: Semantic-based retrieval Level 3: Reasoning-based retrieval ๐—ช๐—ต๐—ฎ๐˜ ๐—ถ๐˜€ ๐—ฅ๐—ฒ๐—ฎ๐˜€๐—ผ๐—ป๐—ถ๐—ป๐—ด ๐—œ๐—ป๐—ณ๐—ผ๐—ฟ๐—บ๐—ฎ๐˜๐—ถ๐—ผ๐—ป ๐—ฅ๐—ฒ๐˜๐—ฟ๐—ถ๐—ฒ๐˜ƒ๐—ฎ๐—น?I It is the ability to find relevant information that requires logical thinking to connect your query
21
292
2K
@asankhaya
Asankhaya Sharma
6 months
๐Ÿงต 1/9 ๐ŸŽ‰ Overwhelmed by the incredible response to OpenEvolve! What started as my attempt to replicate Google DeepMind's AlphaEvolve has become something far beyond my expectations. Our open-source repo is now trending on GitHub!
2
4
15
@AndyLin2001
Haowei Lin
6 months
๐Ÿš€ Thrilled to contribute to the #OpenSource community by implementing @GoogleDeepMind's AlphaEvolve! Applied it to symbolic regression tasks, showing how AI can *discover physical laws* from raw experimental data. โš›๏ธโœจ Check out the code & examples๐Ÿ‘‡ https://t.co/qfH6iqfQWz
0
5
15