Bruce Yang FinTech@AI4Finance

Bruce Yang FinTech | LLM, AIGC, AI Agent, DRL

Founder and President @AI4FinanceFound

Creator of FinRobot, FinGPT, FinRL

[Bruce Yang FinTech@AI4Finance @By_FinTech]

🚀 Exciting News from AI4Finance Foundation! 🚀
As Founder and President, I'm thrilled to announce the official launch of our latest venture—the FinRobot Project! Our Github Repo: https://t.co/zqyHu7Wea9
#AI4Finance #FinRobot #OpenSource

[Quantitative Finance Papers @QFinancePapers]

FinRL-X: An AI-Native Modular Infrastructure for Quantitative Trading

Hongyang Yang, Boyu Zhang, Yang She, Xinyu Liao, Xiaoli Zhang
https://t.co/j3Za0QV0Zh [𝚚-𝚏𝚒𝚗.𝚃𝚁 𝚌𝚜.𝙻𝙶 𝚚-𝚏𝚒𝚗.𝙲𝙿]

[Alex Nguyen @alexngsx]

🚨 Holy shit.. someone just open sourced what quant funds pay $300K+ per year to build. Full RL trading infrastructure from research to live deployment.

It's called FinRL.

Most retail algo traders hit the same wall: you can find tutorials on backtesting or RL theory, but never a framework that covers the entire pipeline from research to live trading. FinRL does.

Here's what it actually does:

→ Gym-style market environments for stocks, crypto, forex, and portfolio optimization — same interface as training a game-playing AI
→ Full train-test-trade pipeline: train on historical data, validate out-of-sample, deploy to paper or live trading
→ Multiple RL algorithms out of the box: DQN, PPO, SAC, TD3, A2C, swap without rewriting your environment
→ Connects to real data sources: Yahoo Finance, Alpaca, Binance, and others
→ Paper trading support so you validate strategies without real capital at risk
→ Portfolio optimization environments for multi-asset allocation strategies

Here's how it works:

You define a trading environment: which stocks, what time period, what transaction costs. The RL agent trains on that environment using historical price data, learning to maximize cumulative reward (your returns minus costs). Test it on unseen data. If it holds up, move to paper trading.

Here's the wildest part:

This isn't a weekend project. The AI4Finance Foundation has published peer-reviewed papers using FinRL. 3,200 forks. 228 contributors. Last commit: 3 days ago.

The research-grade infrastructure for reinforcement learning in finance is free. The only remaining moat for institutional quants is proprietary data and actual capital.

Free. MIT license. 14.3K stars.

[0x_Miko @Mikocrypto11]

一个清华学生，把 Anthropic 的 AI 用成了 Polymarket 上的提款机
$1,430 → $1,550,750

而且素材里给出的数据更夸张：
44,364 笔交易
100% 胜率
单笔最大盈利 $23,600
这个账号叫 k9Q2m

按这段素材的说法，他不是靠运气，也不是靠猜
而是把 6 套对冲基金常用公式 同时塞进 bot 里，每个 tick 都跑一遍
多数人还在判断
这个 bot 直接算

它跑的 6 个核心模块是：
1）LMSR Pricing
Polymarket 的价格沿对数曲线变化
bot 会提前算出自己的进场会带来多大价格冲击
比如市场给 BTC 5 分钟上涨 31¢，模型却判断这段曲线已经错价，于是先进去等修正

2）Kelly Criterion
每一笔都按最合适的仓位去下
不会大到把账户打爆，也不会小到没意义

3）EV Gap Detection
它一直在扫一个东西：
市场价格到底错了多少
比如市场给 30¢，真实概率被它算到 55¢，那 EV 就直接转正，触发进场

4）KL-Divergence
BTC 5 分钟 和 15 分钟 市场本来就有关联
一旦两边漂开，它就当成套利信号
当统计距离超过 0.2，就开始标记机会

5）Bayesian Updates
新区块确认
成交量异动
价格跳动
这些新信息一进来，它就立刻更新概率
先验是 54%，新数据进来后，后验可能直接跳到 71%

6）Stoikov Execution
不是看到机会就冲
它会继续算一个更合适的执行价格
只在风险调整后仍然成立的位置成交

真正执行的时候，不是满足一个条件就下单。
而是这 6 层一起过筛：
LMSR 确认错价
EV gap 超过 5%

Kelly 允许仓位
Bayesian posterior 同意
KL-divergence 发现相关漂移
Stoikov 放行执行价格
只有这样，才会进场。=

也就是说，这已经不是普通意义上的“交易 bot”了
更像是一套 对冲基金框架，被搬进了 prediction market
素材最后那句其实点得很直白：

数学是公开的
edge 也是真的
真正的差别只在于：
大多数人从来没把它真正搭出来
这种把 6 个量化过滤器 同时塞进 Claude，再去跑 Polymarket 的打法，你觉得是未来的标准配置，还是只适合极少数真能把系统搭起来的人？

[TGweb3 @TGweb3333]

有个麻省理工数学系的学生，本来老老实实念着书

结果在 Polymarket 的天气预测市场里

拿 300 美元试了试水最后变成 7.6 万美元提走

然后人就退学了。

他走之前在麻省理工做的最后一个课题

叫“基于机器学习的概率天气预报”。

就是写了个模型，专门去抓那些概率极低、价格极低（0.1 美分以下）的极端天气事件。

模型跑出来的结果，全班看了都沉默：

39 美元 → 5753 美元
14 美元 → 2520 美元
3 美元 → 1358 美元

这个机器人不挑地方，纽约也扫，圣保罗也扫，眼里只有一件事：价格在 0.01 美分到 0.1 美分之间的机会。

他的个人主页：https://t.co/SiOfuVsTt7

想直接抄作业的，可以用 PolyCop 跟他的机器人：https://t.co/y5DIp4Ykvk

[elvis @omarsar0]

This paper is a big deal!

It's well known that RL works great for math and code.

But RL for training agents is a different story.

The default approach to training LLM agents today is based on methods like ReAct-style reasoning loops, human-designed workflows, and fixed tool-calling patterns. The issue is that these methods treat the environment as passive rather than interactive.

But in the real world, agents must make sequential decisions, maintain memory across turns, and adapt to stochastic environmental feedback.

That's fundamentally an RL problem.

This new research introduces Agent-R1, a framework for training LLM agents with end-to-end reinforcement learning across multi-turn interactions.

As agents move from predefined workflows to autonomous interaction, end-to-end RL becomes the natural training paradigm. Agent-R1 provides a modular foundation for scaling RL to complex, tool-using LLM agents.

Standard RL for LLMs assumes deterministic state transitions. You generate a token, append it to the sequence, done. But agents trigger external tools with uncertain outcomes. The environment responds unpredictably. State transitions become stochastic.

Therefore, the researchers extend the Markov Decision Process framework to capture this. State space expands to include full interaction history and environmental feedback. Actions can trigger external tools, not just generate text. Rewards become dense, with process rewards for intermediate steps alongside final outcome rewards.

Two core mechanisms make this work. An Action Mask distinguishes agent-generated tokens from environmental feedback, ensuring credit assignment targets only the agent's actual decisions. A ToolEnv module manages the interaction loop, handling state transitions and reward calculation when tools are invoked.

On multi-hop question answering, RL-trained agents dramatically outperform baselines. The weakest RL algorithm (REINFORCE++) still beat naive RAG by 2.5x on average EM. GRPO achieved 0.3877 average EM compared to 0.1328 for RAG.

Ablation results also confirm that the design matters. Disabling the advantage mask dropped PPO performance from 0.3719 to 0.3136. Disabling the loss mask caused further degradation to 0.3022. Precise credit assignment is essential for multi-turn learning.

Paper: https://t.co/BrIBT3AAxC

Learn to build effective AI agents in my academy: https://t.co/JBU5beIoD0

[Bruce Yang FinTech@AI4Finance @By_FinTech]

I just published AI4Finance Foundation: A Deep Dive into the World’s Leading Open-Source Community for Financial AI https://t.co/6qIlLTfjwp

[Tom Dörr @tom_doerr]

Open-source AI agent platform for financial analysis

https://t.co/8cK0RVT0n0

[AI4Finance-Foundation @AI4FinanceFound]

Updated the AI4Finance Foundation Executive Overview.

The deck summarizes our open-source work in financial AI — including FinGPT, FinRL, and FinRobot — along with current ecosystem metrics (30M+ views, 200K MAU, 43K+ stars).

PDF: https://t.co/i2caRPSegi

[AI4Finance-Foundation @AI4FinanceFound]

Slides from the AI4Finance Foundation deck:
1️⃣ Intro page — global nonprofit focused on AI + finance
2️⃣ About us — 501(c)(3) status, trademarks, mission, and impact
3️⃣ Our approach — Foundation / FinTech / Fund structure
4️⃣ Project impact — growth of FinGPT, FinRL, and FinRobot

[Quantitative Finance Papers @QFinancePapers]

FinRobot: Generative Business Process AI Agents for Enterprise Resource Planning in Finance. https://t.co/53DOiSj5uU

[Quant Science @quantscience_]

💡FinRL: A Python package for Financial Reinforcement Learning

Let's explore: 🧵

[AI4Finance-Foundation @AI4FinanceFound]

AI4Finance Foundation Secures U.S. Trademark Registration for “AI4FINANCE”🇺🇸📄

[Alfonso Amayuelas @AlfonAmayuelas]

🚨Check out my work with the AI team at @jpmorgan !!! 📜 "Grounding LLM Reasoning with Knowledge Graphs" explores the intersection of Knowledge Graphs and Reasoning strategies with LLMs which helps boosting answers in domain specific questions (1/4)

🧵👇

[AI4Finance-Foundation @AI4FinanceFound]

We’re thrilled to announce that Mostapha B. has joined us as an AI researcher, bringing his groundbreaking work to our community. Mostapha’s latest research: FinRL-DeepSeek: LLM-Infused Risk-Sensitive Reinforcement Learning for Trading Agents
https://t.co/7xlBrfnvs1
#AI4Finance

[Quantitative Finance Papers @QFinancePapers]

FinRL-DeepSeek: LLM-Infused Risk-Sensitive Reinforcement Learning for Trading Agents. https://t.co/JGipEgccSR

[Bruce Yang FinTech@AI4Finance @By_FinTech]

https://t.co/6sbDro8kYj

[Bruce Yang FinTech@AI4Finance @By_FinTech]

Excited to start my 7th semester teaching Columbia’s STAT GR5398 course! 🎉 'Teaching Without Teach’ philosophy-empowering students to take the lead in real-world AI projects. All materials are open-source for everyone to learn & contribute! #AI #Education #OpenSource

[Quantitative Finance Papers @QFinancePapers]

FinGPT: Enhancing Sentiment-Based Stock Movement Prediction with Dissemination-Aware and Context-Enriched LLMs. https://t.co/1b9vojLYMB

[elvis @omarsar0]

A Survey on LLMs-as-Judges

Presents a comprehensive survey of the LLMs-as-judges paradigm from five key perspectives: Functionality, Methodology, Applications, Meta-evaluation, and Limitations.

[elvis @omarsar0]

Gemini 2.0 real-time AI is absolutely wild!

Watch how I use it as an AI research assistant by sharing my screen and asking it about an AI paper.

10x your paper reading skills or just let Gemini summarize key points.

What an incredible time to be alive!