programming ai agent:

https://useadrenaline.com/

blockchain applicatiion:

https://playdoge.io/en

coin mixer:

https://usdtmixer.com/index.html

https://github.com/sh0tn1k/simple-usdt-mixer

payment provider:

https://www.usdtcnpay.com/

https://github.com/CoinPayVIP/dujiaoka_usdt_payment

trc-20 related:

https://github.com/topics/trc20

https://tron.network/index

blockchain inspectors:

https://bscscan.com/

https://tronscan.org

trading platform:

https://www.okx.com/zh-hans

query: buy things with usdt

https://www.coinsbee.com

https://cryptonews.com/news/who-accepts-tether.htm

new proxy client:

https://github.com/clash-next/clash-meta

https://github.com/MetaCubeX/mihomo (supports proxy chaining)

https://wiki.metacubex.one/en

https://github.com/clash-verge-rev/clash-verge-rev

web3 security and poc:

https://github.com/SubWeb3Sec/DeFiHackLabs

https://github.com/0xCO2/Web3_Hack_PoC

https://github.com/Anugrahsr/Awesome-web3-Security

https://github.com/ManasHarsh/Awesome-Web3-security

https://github.com/descartes100/Web3Hack

Subscribe publish sources like blog, news, rss feed, recommendation engine to create video generation agents.

Github blog

Smart contract in EVM might be overkill, but it comes with the advantage of multiuser security.

A central solution might be beanie, an ODM (Object Document Mapper) in MongoDB, or pyneo4j-ogm, an OGM (Object Graph Mapper) in Neo4j

Playstation emulators can emulate up to PS5.

There are two Xbox emulators around. One is xemu, the original or the first Xbox emulator, which can run almost on any platform. The game for it can be downloaded at the vault. The other is xenia or the Xbox 360 emulator, which can only be running on top of Windows 10+.

Xbox 360 is hackable and the game for it can be downloaded as GOD format.

Greenlight is an electron app for local Xbox streaming and XGPU cloud gaming, supports multiple platforms.

To train a gaming AI and get potential financial rewards:

Buy a gaming console like Xbox Series S and use Arduino Leonardo or ESP32 to send gamepad signal. However since you do not have the authority to mimic official gamepad this might fail.

You may also use HID to gamepad hardware.

Make sure you do not delete the game accidentally. You may design such mechanism to prevent exiting the game and misbehave. You can create child accounts to prevent unwanted purchase.

Use Xbox 360 Emulator instead of running a real one.

Blockchain or Web3 is not silver bullet. Anything on Web3 can just be done in Web2 in centralized way.

Still smart contract compiler does not have Python API generator.

Local node: Geth, Hardhat, Ganache

Development environment: Ape, Brownie

Smart contract language: Solidity, Vyper

You have posted a lot of ideas into your QQ, WeChat, bookmarks and Bilibili favourites. Consider using tools to retrieve them?

卡点视频可以用Magenta Studio之类的MIDI生成模型或者时序输出模型来做，卡时间节点。

I hate Linkedin and Twitter. They both ban my accounts for no goddamn reason. Newly registered accounts shall do no operation for a long time before using features like browsing, following.

Consider finding some alternative platforms for job finding (remote work: V2EX, crypto careers, web3.career) and social network.

Use smart contract like sui move for auto re-execution of automated (agentic) video generation scripts and hacking scripts, reward these scripts for successful task fulfillment (view count bypassing 1000; ssh login success; crypto currency obtained), and punish or ban these scripts at failed condition.

Human execution will be recorded for automated execution, and it is free. Automated init execution will only be granted once.

Notice you do not want overflow errors in the smart contract.

Use retrieval system with chat history while building the chatbot. Mind different index modes in llamaindex, in order to better capture consecutive context range with dynamic sliding window size.

Tachiyomi is for free manga reading, and manga narriative video. Its plugin store has been removed and requires you to search for workarounds.

Deep recursive thoughts and subconscious operators like REM and DELEGATE might help for making the agent to think for itself, becoming conscious.

Tornado cash is for crypto mixing and now it is public archive. Alternatives are:

• Aztec zk.money

• Monero 0xTIP

• Messier 87

• Cyclone Protocol

• RenBridge

Use vimgolf for collecting vim usage data to train cybergod, however the cli sucks and requires learning the javascript request headers from browser.

A comprehensive guide from llamaindex hackrathon

vanna.ai can work as sql agent. what can generate cypher query for neo4j?

Coroutines and tasks (asyncio.create_task) can save time. You can also make blocking actions non-blocking by asyncio.to_thread.

InternLM-XComposer seems to be better than CogVLM, and can generate image with text interleaving articles.

The platform openxlab is a huggingface clone, and maybe it can prvide some model execution service.

A better alternative for SadTalker: Dream-Talk (still single image generation).

Audio2Gesture (pose generation by audio) can be exported to Metahuman.

Wav2Lip, Linly-Talker, VirtualWife, OpenVTuber

Prompt engineering: Pezzo for prompt version management, LangGPT for prompt template generation

Fine-tuning: Prompt2Model, GPT-LLM-Trainer

Video generation: MagicVideo-V2 by ByteDance, Lumiere by Google, VideoCrafter2 by Tencent, I2VGen by Alibaba

Anymate Anyone with OpenPose, MotionShop

DragonDiffusion for image detail editing

DeepSeekMoE 16B: the first MoE model in China

Auto Labeling: AutoLabel, AnyLabeling

RAG: QAnything by NetEase, StreamRAG for video search, DAnswer

Document understanding & OCR: Nougat, CnOCR

TTS: Bert-ViTS, GPT-SoViTS

Agent: ProAgent, crewAI, Dify, AutoGen, Agenta, AgentVerse, ChatDev, Agently

YOLOv8 AimBot

Make-A-Character

AnyDoor, AnyText

PhotoMaker

CoTracker

MagicAnimate

ImageDream

VideoPoet

Boximator

InpaintAnything

在BOSS直聘上面搜索腾讯、字节、阿里、微软等关键词，投递简历。官网上传简历的步骤比较繁琐。

外国的企业建议在Linkedin上面聊。发送邮件或者官网投送一般都没回复。（或许是我简历写的不好么）

英文简历不要用ChatGPT写，页数要够4页。

Arc近日将推出mac端的Arc自动浏览功能，iPhone端的Arc Search在Preview商店里面可以下载。

Arc浏览器貌似优化较差，速度较慢，界面用起来非常不习惯。

邮箱：careers@thebrowser.company

LeptonAI推出Lepton Search项目（对话式搜索引擎），需要用Lepton API和Bing API。

LeptonAI有状态机JSON格式输出/方法调用的实现。不过建议给大模型一定的思考空间，不要一上来就叫它给答案。

邮箱：info@lepton.ai

电报偶然一次打开了一个莫名其妙的图片 然后导致每次打开都崩溃

用root权限 在/data/data/org.telegram.messager/files下面删除cache4.db

最近登陆电报不能给国内手机发短信或者打电话了 新注册账号可能需要借助某宝

Images used by v86 can be run using QEMU locally. Or find under: https://k.copy.sh/<image_filename>

Visit v86 to run systems like: FreeDOS, KolibriOS, Windows 2000 etc in browser. Image file is written in WASM. Project is in Rust.

You can have memory dump, save & load state, network connection and disk access.

Linux is most 80386 (32bit) based and does not support 80286 (16bit). If you want to use Linux in 16bit mode special kernel is required.

The system sometimes feels lagging, less responsive. The only way to fix is to reboot.

Although monitor task completion time is helpful, but not general enough.

Simple hack:

1
2
3
4
5
6
7
8
9
10
11

import subprocess
def measure_system_responsiveness():
start_time = time.time()
subprocess.run(
["echo", "hello"], capture_output=True, encoding='utf-8', check=True
)
end_time = time.time()
exec_time = end_time - start_time
exec_per_second = 1 / exec_time
return exec_per_second

Lagging related package:

1
2

apt install nohang oomd psi-notify

Temperature monitoring packages:

1
2
3
4

pip3 install gpustat
pip3 install pyspectator
apt install lm-sensor # sensors -j

Create temperature statistics (high, low, mean) for all crucial components.

Set alert threshold, only trigger alert if temperature reoccurs for several times.

Send ntfy.sh notification as well as making audible alerts.

Notification shall specify device name, component name, current temperature, threshold, statistics.

Every device shall be equipped with this software as daemon, be it smartphone, MacBook, Linux PC, Windows PC.

lfs download is mandatory while doing git clone

skip with:

1
2

env GIT_LFS_SKIP_SMUDGE=1 git clone --filter=blob:none <repo_url>

DFPy, auto example based prompt engineering

Semantic search, question-answer generation (query generation), prompt engineering, template generation, usage generation, topic/category clustering, turn plain text summaries into markdown document.

Bad written code will suffer from longer process time (1000+ loc file), ambiguous project structure and so on. Currently the file cache system is just enough.

Related things:

autodoc with self-generated documentation

doc-comments-ai

AutoDoc-ChatGPT

write-the

Cody codebase understanding

symf (without readme, closed source?)

sourcegraph (remember how we browse source code of bilibili and potentially gta5 code)

livegrep

issue on rg to create persistant index

Documentation frameworks:

mkdocs

Useful libraries:

hnswlib

This device is wireless, but it does not mean you can use it wirelessly without connection. You must either connect it wirelessly to some other device or you have to plug in the audio wire (no device needed) in order to prevent it from automatic shutdown. It would be annoying if it shuts down when you had almost fallen asleep.

How to deploy Github Pages in /docs or from fork

Click to Settings -> Pages -> Source, select “Github Actions” and select back, then you can select your branch and /docs folder successfully.

Is that delay? I doubt. But I will record this anyway.

When reading bytes asynchronously from stdout using asyncio.create_subprocess_exec, the program has to be unbuffered.

1
2

python3 -u <script_path>

1
2
3
4
5
6
7
8
9
10
11

# this can only make `print` into unbuffered
import builtins
import copy
old_print = copy.copy(builtins.print)
def custom_print(*args, **kwargs):
if 'flush' not in kwargs:
kwargs['flush'] = True
old_print(*args, **kwargs)
# Override the built-in print function with the custom function
builtins.print = custom_print