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

https://github.com/frdel/agent-zero

https://digirl-agent.github.io/

https://github.com/opendilab/awesome-ui-agents

https://github.com/hyp1231/awesome-llm-powered-agent

https://github.com/skyvern-ai/skyvern

https://github.com/Envedity/DAIA

https://github.com/mem0ai/mem0

https://qinghonglin.github.io/

https://github.com/showlab/Awesome-GUI-Agent

https://github.com/waterhorse1/LLM_Tree_Search

https://github.com/evilsocket/nerve

https://www.superjoin.ai

https://github.com/test-time-training/ttt-lm-pytorch

https://github.com/stanfordnlp/dspy

https://www.builder.io/blog/micro-agent

search for site:github.com <computer agent benchmark name> agent and get a bunch of new computer agent frameworks

https://github.com/posgnu/rci-agent

https://github.com/stanfordnlp/wge

https://github.com/ServiceNow/BrowserGym

https://lmql.ai llm query language

neural network generation/neural developmental programs

https://arxiv.org/abs/2406.09787

James4Ever0/agi_computer_control: Autonomous computer program that can do anything without human operators.

niuzaisheng/ScreenAgent: ScreenAgent: A Computer Control Agent Driven by Visual Language Large Model

tmgthb/Autonomous-Agents: Autonomous Agents (LLMs) research papers. Updated Daily.

ltzheng/Synapse: [ICLR 2024] Trajectory-as-Exemplar Prompting with Memory for Computer Control

SkyworkAI/agent-studio: Benchmarks, environments, and toolkits for general computer agents

OS-Copilot/OS-Copilot: An self-improving embodied conversational agent seamlessly integrated into the operating system to automate our daily tasks.

landing-ai/vision-agent: Vision agent

smartcomputer-ai/agent-os: Build autonomous AI agents! 🌞

idosal/AgentLLM: AgentLLM is a PoC for browser-native autonomous agents

posgnu/rci-agent: A codebase for “Language Models can Solve Computer Tasks”

khulnasoft/gpt-computer-agent: GPT4 for windows, macos and ubuntu

TheDuckAI/DuckTrack: Multimodal computer agent data collection program

X-PLUG/MobileAgent: Mobile-Agent: The Powerful Mobile Device Operation Assistant Family

mnotgod96/AppAgent: AppAgent: Multimodal Agents as Smartphone Users, an LLM-based multimodal agent framework designed to operate smartphone apps.

stableagents/stableagents: Stable, Semi-Autonomous, Reliable and Steerable LLM Agents for production use cases.

OSU-NLP-Group/Mind2Web: [NeurIPS’23 Spotlight] “Mind2Web: Towards a Generalist Agent for the Web”

microsoft/autogen: A programming framework for agentic AI. Discord: https://aka.ms/autogen-dc. Roadmap: https://aka.ms/autogen-roadmap

geekan/MetaGPT: 🌟 The Multi-Agent Framework: First AI Software Company, Towards Natural Language Programming

richardyc/Chrome-GPT: An AutoGPT agent that controls Chrome on your desktop

Mass Copy URLs − copy all URLs on all tabs

openeqa: embodied agent question answering benchmark

https://github.com/facebookresearch/open-eqa

GUI Automation agent and dataset:

https://github.com/OpenGVLab/GUI-Odyssey (cross-app tasks)

https://github.com/TransformerOptimus/AutoNode

https://superagi.com/

https://huggingface.co/datasets/SuperAGI/GUIDE/

https://huggingface.co/SuperAGI/SAM

AI pipeline orchestration:

https://github.com/instill-ai/instill-core

https://github.com/ComposioHQ/composio/ (with GUI agent)

GUI dataset annotation can be done manually or using multimodal LLM.

It matters to validate the answer of questions requiring code execution with code.

YOLO GUI element identification:

https://github.com/rahulkundelwalll/YOLOv8-Web-Element-Recognition-Model

https://huggingface.co/foduucom/web-form-ui-field-detection

https://github.com/js0nwu/webui

https://huggingface.co/docs/transformers/model_doc/pix2struct

https://github.com/google-research/pix2struct

https://github.com/M3SOulu/WinGUICrawler

https://huggingface.co/datasets/yiye2023/GUIEnv

https://huggingface.co/datasets/yiye2023/GUIAct

https://huggingface.co/SiyuanH/GUIAgent

https://huggingface.co/datasets/SiyuanH/GUIAgent

https://huggingface.co/SiyuanH/GUIAgent-InternLM7B

Not every repo has official documentation.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

git clone https://github.com/opendilab/LightZero
git clone https://github.com/ruvnet/q-star
git clone https://github.com/tairov/QStarLearning.mojo
git clone https://github.com/estill01/open_qstar
git clone https://github.com/openai/Video-Pre-Training
git clone https://github.com/abhiprojectz/SingularGPT
git clone https://github.com/ddupont808/GPT-4V-Act
# preload-view.js:markPage is the html-to-boundingbox tool.
# the author wants to create a coco dataset
# specialized in UIED-like functionality
git clone https://github.com/Charmve/gpt-eyes
git clone https://github.com/OthersideAI/self-operating-computer
git clone https://github.com/unconv/gpt4v-browsing
git clone https://github.com/THUDM/CogVLM
git clone https://github.com/mnotgod96/AppAgent

https://www.novaspivack.com/business/the-four-levels-of-ai-steps-to-self-evolution

https://www.novaspivack.com/business/exploring-higher-order-ai-training-on-first-order-ai-networks-for-optimal-path-outcomes

i want to create some sort of agent, that learns autoregressively on historical tokens (not necessarily present in history, but close). however, when the agent is given some previous tokens, it is expected to send some actions to the environment in order to really observe the given tokens to get reward. the agent is not allowed to directly generate the token to the environment in order to prevent cheating. the agent is rewarded to successfully rebuild the past or predict and build the future. to predict the future is like the target token is generated by the agent itself instead of some automatic history replay bot, and the rest of the reward system follows the same way as the history replay reward system. this kind of system might have some sort of consciousness and therefore agi

the main objective of AGI is to create another version of itself.

the verification system can be built upon internal hidden tokens (you feel like you made it, feeling based) or similarity based (timeseries similarity or semantic similarity). there can be some external verification system such as lifespan, disk usage, view count, popularity, total capital etc.

the main problem of making this work is how to train it in parallel. the real world can be replaced by some world model (say some neural network) so that it can go back in time, or some really fast real world evaluators or some special world evaluators which supports time traversal, like virtual machine snapshots, web browsers (tab traversal). alphago has such advantage because go game is a very simple world model, while the real world is not.

also this could build some hierarchy like: real world -> world model -> agent -> superagent -> …

shall we design the agent system based on ‘distance’ instead of ‘role’? in that case, only words that make sense and continue more often will appear together, and vice versa.

multi-agent orchestration:

langchain

autogen

agently

ModelScope-Agent

you must think like an expert and show some example or direction as reasonable goals if you want it to perform specific task.

you must give it enough degree of freedom if you want it to self-improve and become conscious.

question

how do i design an ai model using gpt, rnn, lstm, vit to accept computer display screenshots and human instructions as input, keyboard and mouse events as output, trained on random keystrokes and mouse events, also human instructions paired with human demonstrations? show me python code

answers

Claude+

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

That would be an extremely complex AI system to build. Some high level steps would be:
Collect a large dataset of screenshots, human instructions, and corresponding keyboard/mouse interactions. This dataset would be crucial for training.
Use a CNN to extract visual features from the screenshots. Something like VGG16 or ResNet would work well here.
Use an RNN/LSTM to encode the sequence of instructions. BERT or RoBERTa would also be good options here.
Combine the visual and textual features. Options include concatenation, addition, or a more complex fusion method.
Use a Transformer like GPT-3 or Visual-Transformers to generate the keyboard/mouse interactions from the combined visual+textual features.
Train the entire model end-to-end to minimize the difference between the generated interactions and the ground truth interactions from your dataset.
This would likely take a team of ML engineers and researchers to develop, and would require lots of time, data, and compute.
Some sample Python code would be:
import torchvision.models as models
import torch
import torch.nn as nn
from transformers import RobertaModel, GPT2LMHeadModel
# Feature Extractors
cnn = models.resnet50(pretrained=True)
roberta = RobertaModel.from_pretrained("roberta-base")
# Fusion
fusion = nn.Sequential(nn.Linear(2048 + 768, 1024), nn.ReLU(), nn.Linear(1024, 512))
# Generation Model
generator = GPT2LMHeadModel.from_pretrained("gpt2")
# Forward pass
img_features = cnn(img)
instruction_features = roberta(instructions)[0]
combined_features = fusion(torch.cat((img_features, instruction_features), dim=1))
preds = generator(inputs_embeds=combined_features)[0]
This is just a brief sample, but hopefully gives you an idea of the components and flow. Let me know if you have any other questions!

GPT4

Designing an AI model that combines GPT, RNN, LSTM, and ViT for the task you described is complex and requires deep understanding of these architectures. However, I’ll provide you with a high-level overview and a basic structure of the code to help you get started.

1. First, process the input data:

• For computer display screenshots, use a Vision Transformer (ViT) model to extract features.

• For human instructions, use a GPT model to encode the natural language instructions.

2. Combine the features from both the ViT and GPT models.

3. Use an LSTM or RNN to process the combined features and maintain the temporal context.

4. Finally, decode the output of the LSTM/RNN into keyboard and mouse events.

Here’s a basic structure of the code:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33

import torch
import torchvision
from transformers import GPT2Tokenizer, GPT2Model
# Load pretrained models
gpt_model = GPT2Model.from_pretrained('gpt2')
vit_model = torchvision.models.vit_b_16(pretrained=True)
tokenizer = GPT2Tokenizer.from_pretrained('gpt2')
class CustomModel(torch.nn.Module):
def __init__(self, gpt_model, vit_model, hidden_size, output_size):
super(CustomModel, self).__init__()
self.gpt_model = gpt_model
self.vit_model = vit_model
self.rnn = torch.nn.LSTM(input_size=hidden_size, hidden_size=hidden_size, batch_first=True)
self.fc = torch.nn.Linear(hidden_size, output_size)
def forward(self, instructions, screenshots):
# Encode instructions using GPT2
inputs = tokenizer(instructions, return_tensors="pt", padding=True, truncation=True)
gpt_output = self.gpt_model(**inputs)
gpt_features = gpt_output.last_hidden_state[:, 0, :]
# Encode screenshots using ViT
vit_output = self.vit_model(screenshots)
vit_features = vit_output["last_hidden_state"][:, 0, :]
# Combine features
combined_features = torch.cat((gpt_features, vit_features), dim=1)
# Process combined features through an RNN
rnn_output, _ = self.rnn(combined_features.unsqueeze(1))
# Decode the output into keyboard and mouse events
output = self.fc(rnn_output.squeeze(1))
return output
# Instantiate the custom model
model = CustomModel(gpt_model, vit_model, hidden_size=1024, output_size=256)
# Train and evaluate the model using your dataset

You’ll need to collect a dataset with computer display screenshots, human instructions, and paired keyboard and mouse events. Then, preprocess the data, split it into training and validation sets, and train the model using an appropriate loss function and optimizer.

Keep in mind that this is just a starting point, and you may need to experiment with different architectures, loss functions, and hyperparameters to achieve the desired performance.

论文代写 作业代写转为AI自动运作项目

收集各大兼职网站 QQ接单群 闲鱼的项目描述 人工标注项目报价费用 训练实现自动报价抽成机器人

根据项目描述和大量公开数据 进行开放式问答 训练并应用于论文实现代写方面的付费问答机器人

source webpage

fix requests timeout problem.

enable bilibili chats, video recommendation, dynamics

enable weibo

video feedback, monitor viral trends

improve chatbot by adding chatterbot, bm25, search engine, picture search engine

you may use this to fully utilize data from baidu shitu:

https://github.com/Augu1sto/Rubindemo/blob/0ffe52af74643db8d8bfae048ee256824836e277/src/main/java/com/rubin/demo/Utils/BaiduSerchImgApi-master/functions.py

https://github.com/chenguanyou/BaiduSerchImgApi

https://github.com/chenguanyou/360ImageSearch

https://github.com/chenguanyou/BaiduTextApi

shitu.baidu.com

https://graph.baidu.com/pcpage/index?tpl_from=pc (the entry page)

graph.baidu.com/s (where you collect data, recognize identities: script -> window.cardData (list) -> “tplData” -> “pano” “baike” shits … with title “百度识图搜索结果”)

graph.baidu.com/ajax/pcsimi (get similar images, sources)

https://miao.baidu.com/abdr (UNKNOWN)

classes:

graph-guess-word

graph-baike-text

graph-baike-desc -> span

tool repository

i was informed by a post on zhihu.

futurepedia

creatives

futuretools

getinference

notice: QA based websites are popular now. inspired by that and jumpcut, you can mine summaries/questions/prompts/examples relating to some topic, generating answers/scripts from chatgpt (imagination is the limit)

videos are similar by title, author and topic. you demand the bot to generate new embeddings by any “abstractive” means, giving sample embeddings (subtitles, voice, sentiment, music, scene), and output needed embeddings. cool!

There are a few different ways you can use AI to generate video content and distribute it for advertising purposes. Here are a few options:

Use an AI video creation tool: There are several AI tools available that allow you to input text, images, or other media and have the AI generate a video for you. These tools can be useful for creating promotional videos, explainer videos, or other types of content quickly and easily.

Use an AI-powered video editor: Some video editing software includes AI-powered features that can help you edit and enhance your videos more efficiently. For example, you might be able to use AI to automatically color grade your footage or to stabilize shaky video.

Use an AI-powered video platform: There are also platforms that use AI to optimize and distribute your video content for maximum reach and engagement. These platforms can help you target specific audiences, measure the effectiveness of your videos, and more.

It’s important to note that while AI can be a useful tool for creating and distributing video content, it’s not a replacement for human creativity and insight. The best results will often come from a combination of AI and human input.

recommend popular ai video creation tools

There are a number of popular AI video creation tools that you might want to consider using, depending on your needs and budget. Here are a few options:

Animoto: Animoto is a cloud-based video creation platform that uses AI to help you create professional-quality videos quickly and easily. You can choose from a variety of templates, add your own media, and customize your video with text, music, and other elements.

Wibbitz: Wibbitz is an AI-powered video creation tool that allows you to turn text into video content. You can input a written article or script, and the AI will generate a video complete with images, text, and music.

Adobe Premiere Elements: Adobe Premiere Elements is a consumer-level video editing software that includes AI-powered features such as automatic video editing, object detection, and more.

Lumen5: Lumen5 is an AI-powered video creation platform that allows you to input text, images, or a link to an article and have the AI generate a video for you. You can customize the video with your own branding, music, and other elements.

It’s a good idea to try out a few different tools and see which one works best for you and your needs. Keep in mind that while these tools can be helpful, they may not produce results that are exactly what you’re looking for, and you may need to do some manual editing or customization to get the desired results.

recommend AI-powered video editor

There are several AI-powered video editors available that you might want to consider using, depending on your needs and budget. Here are a few options:

Adobe Premiere Pro: Adobe Premiere Pro is a professional-grade video editing software that includes a number of AI-powered features, such as object detection and automatic color grading.

DaVinci Resolve: DaVinci Resolve is a professional-grade video editing software that includes a number of AI-powered features, such as automatic color grading and audio repair.

Final Cut Pro: Final Cut Pro is a professional-grade video editing software for Mac that includes a number of AI-powered features, such as automatic color grading and object detection.

Filmora: Filmora is a consumer-level video editing software that includes a number of AI-powered features, such as automatic video stabilization and object detection.

It’s a good idea to try out a few different video editors and see which one works best for you and your needs. Keep in mind that while these tools can be helpful, they may not produce results that are exactly what you’re looking for, and you may need to do some manual editing or customization to get the desired results.

how can ai help you target specific audiences

There are several ways in which artificial intelligence (AI) can help target specific audiences:

Personalization: AI can analyze data on individuals and their preferences, and use that information to tailor content and messaging specifically to them. For example, an AI-powered recommendation engine can suggest products or content based on a user’s past behaviors and interests.

Segmentation: AI can help to segment audiences into smaller, more specific groups based on shared characteristics, such as demographics, interests, or behaviors. This can allow organizations to create more targeted campaigns and messaging.

Predictive analytics: AI can analyze data on past behaviors and outcomes to make predictions about future behavior. This can help organizations to better understand their target audience and tailor their marketing efforts accordingly.

Natural language processing: AI can analyze and understand human language, allowing organizations to use chatbots and other tools to communicate with their audience in a more personalized and natural way.

Image and video analysis: AI can analyze images and videos to understand the content and context, allowing organizations to target specific audiences based on the visual content they consume.

Overall, AI can be a powerful tool for targeting specific audiences by providing insights, personalization, and automation to marketing efforts.