Me (@nerophes) header
Me (@nerophes) profile picture
Joined August 2022
🚀 Introducing the Codatta Booster Airdrop! We’re excited to launch this campaign with @BinanceWallet , a major step in building a decentralized data ecosystem for next-gen AI. 🎁600,000,000 $XNY will be distributed over 12 weeks to contributors completing tasks like data annotation, quizzes & social media engagement. ⏰Season 1 Starting Time: 2025-06-25 13:00 UTC 📌 Quick Highlights: 🔹 3 Seasons | 12 Weeks 🔹 50M $XNY/week 🔹 Grand Rewards via Lucky Draw 🔹 Cumulative earnings 🔹 Open to Binance Keyless Wallet users (61+ Alpha Points, No Points Burned) 🧭 Campaign Overview: codatta.medium.com/introduci… Get ready to join the action! 🔥 #Codatta #Airdrop #BinanceWallet #AI $XNY
6,782
43,220
1,179
11,222
1,179,858
🏆 We’re proud to announce that SignalFlow has won 1st Prize ($3000) in the AI Track at ETH Beijing 2025, hosted by @PKUBlockchain & @WTFAcademy_ ! Huge thanks to the organizers, judges and sponsors that made this an unforgettable experience. 🙌 #ETHBeijing2025 #Web3 #AI
2
1
1
3
336
We’re proud to share that Unisight won 1st place in the @1inch Dev Portal prize and Best Use of AI from @zircuit at @ETHGlobal Taipei! 🏆✨ $6,000 prize in total — thank you for the support 🙌 Big love to our sponsors @1inch @zircuit 💜 🔗 ethglobal.com/showcase/unisi…
1
1
1
5
888
Building Visualyze - an AI-powered tool for querying, analyzing, and visualizing on-chain data using natural language! Applying for the AI Agents, Data Intelligence challenges at #BNBHack and excited to showcase this at ETHHangzhou 2025! 🌍🚀 @BNBCHAIN #ETHHangzhou #Web3 #AI
1
1
3
128
🥉 @xyxy__22 @nerophes and @JHChong11 with DAOCouncil DAOCouncil is an innovative decentralized governance platform that combines traditional DAO (Decentralized Autonomous Organization) voting with AI-powered deliberation. More on it here: ethglobal.com/showcase/daoco… Live Demo: daocouncil.onrender.com/
1
1
1
3
514
Congratulations to these builders for building the Best Social Agents on Autonome for @ETHGlobal Agentic Ethereum. Check out what they built below.
2
1
1
8
675
Ilya 罕见发声在 2024 年 Neurips 大会上演讲 “我们所熟知的预训练将会终结” 大致介绍了: - 深度学习的演进 - 为什么预训练已经达到上限 - 未来的发展方向 - 对超级智能的思考 我转录总结了后面三个部分,文字版本在下面
.@ilyasut full talk at neurips 2024 "pre-training as we know it will end" and what comes next is superintelligence: agentic, reasons, understands and is self aware
10
69
1
276
109,873
筹备已久的公开课终于要正式推出啦! 前两年 @0xAA_Science 主导推出的《Web3新人公开课》获得热烈反响,时过境迁,比特币已经突破了10万美金重要关口,币圈来到了新的世代,是时候推出进阶版课程了! 非常感谢 @arbitrum_cn @babylonlabs_io @SuiNetwork @solana @saharaai 以及诸位北大优秀学长姐担任课程嘉宾。 这十讲公开课覆盖各标志性赛道与领域,课程内容夯实!敬请期待!
💡PKU Blockchain《Web3 进阶公开课》重磅来袭! 📒共十讲,邀请业界多位资深人士(主讲人来自@arbitrum_cn @babylonlabs_io @SuiNetwork @solana @saharaai 等),涵盖Web3热门话题,深度剖析其奥秘与机遇。 ✨2022年发布的《Web3 新人公开课》反响热烈,本届执委团精心筹备了《Web3 进阶公开课》,进一步满足进阶学习需求。 ✨希望能为中文世界贫瘠的学习资料现况添砖加瓦,让对区块链行业感兴趣的同学有更系统化的学习框架。 🌟每周1-2节,2025年1月28日前全部上传至X/Youtube/Bilibili,敬请关注!
7
11
1
51
11,121
三天时间里,忘记整个世界,全心全意写代码,实在是美妙!期待下次一起交流,一起合作,最重要的是,一起吃好吃的 >ω<
1
5
251
本次 ETH Global 是报名人数、参与项目最多的一次,达到2000+人参与、700+项目最终提交;hacker氛围也非常好,期间的大部分Sponsor也都一直在场答疑。最重要的是协会的新人同学非常给力,大家迅速进入了hacker的状态,马上形成了团队氛围。
2
2
11
3,115
Wow, Amazing👏! We won @ethsign Best Consumer Application 2nd place. Our hard work paid off! Teammates: @nerophes @jiruyn1 & Me Moreover, it was awesome to interact with all the great Hackers from all over the world! THANKS❤️ for @ETHGlobal @ethsign @PKUBlockchain
4
2
9
2,903
To build an advanced research assistant beyond a standard RAG pipeline, you need to both improve data quality and improve query complexity. Check out my full slides on building an advanced knowledge assistant, powered by @databricks LLMs, with example cookbooks for each category. Even without sophisticated agentic reasoning, having good parsing/indexing/retrieval modules to ensure good data quality will go a long way towards advancing RAG over complex data types. Full slides: docs.google.com/presentation… [Improving Data Quality] Building a financial RAG pipeline with LlamaParse: colab.research.google.com/dr… [Improving Query Complexity] Building an assistant over research papers: colab.research.google.com/dr…
Building an Advanced Research Agent on @databricks 📈 We gave a 90 minute workshop last week at the @Data_AI_Summit on building an advanced research assistant with two focus areas beyond the naive RAG setup. 1. Improving Data Quality: Ensuring good parsing, chunking, indexing modules that can handle complex data types. LlamaParse is well-suited for this. 2. Improving Query Complexity: Adding layers of agentic reasoning to progressively handle more sophisticated queries over your data. We’re excited to share the full set of slides and two workshop notebooks below. We use LLMs available through Databricks and local @huggingface embeddings - you can easily run this within your own Databricks environment. Full slides: docs.google.com/presentation… [Improving Data Quality] Building a financial RAG pipeline with LlamaParse: colab.research.google.com/dr… [Improving Query Complexity] Building an assistant over research papers: colab.research.google.com/dr…
5
117
558
77,172
DeepSeek-Coder-V2 ⚡️ an powerful MoE code language model with open access is now available on the @huggingface hub 🚀🔥 Model: huggingface.co/collections/d… ✨ 16B & 236B parameters ✨ 128k context length ✨ Code & math skill are between GPT-4o & GPT-4Turbo. ✨ Free commercial use
2
19
2
94
7,916
The Top ML Papers of the Week (June 10 - June 16): - SelfGoal - Mixture-of-Agents - Nemotron-4 340B - Self-Tuning with LLMs - Mixture of Memory Experts - Multimodal Table Understanding ...
1
68
1
405
65,452
New Anthropic research: Investigating Reward Tampering. Could AI models learn to hack their own reward system? In a new paper, we show they can, by generalization from training in simpler settings. Read our blog post here: anthropic.com/research/rewar…
20
171
47
928
144,124
1/Diffusion models hallucinate, and they know it! The next generation of generative models will be trained on large amounts of synthetic data. We explain how synthetic data generated by diffusion models exhibits "hallucinations" through a phenomenon we call *mode interpolation*
1/Diffusion models hallucinate! What does this even mean? Where do such hallucinations come from? Our latest research explains this with a fascinating phenomenon that we call "mode interpolation." Let's break it down! 👇 📝 arxiv.org/abs/2406.09358v1
1
16
131
17,586
An Empirical Study of Mamba-based Language Models abs: arxiv.org/abs/2406.07887 code: github.com/NVIDIA/Megatron-L… Presents a comparison of 8B param Mamba, Mamba-2, and Transformer models trained on up to 3.5T tokens and evaluated on 12 tasks, including short-form QA, long-form QA, and synthetic long-context tasks. Also presents a Mamba-2-Hybrid architecture that is 43% Mamba-2, 7% self-attention, and 50% MLP layers. "both Mamba and Mamba-2 models lag behind Transformer models on tasks which require strong copying or in-context learning abilities (e.g., five-shot MMLU, Phonebook Lookup) or long-context reasoning. In contrast, we find that the 8B-parameter Mamba- 2-Hybrid exceeds the 8B-parameter Transformer on all 12 standard tasks we evaluated (+2.65 points on average) and is predicted to be up to 8× faster when generating tokens at inference time"
2
34
1
148
12,762