Odkazy z přednášky na FF 2024 - Kam kráčíš, umělá inteligence?

odkazy.md

Prezentace na google drive

https://docs.google.com/presentation/d/1AHTHncolk60ccyERASaZhy_6UlwBnuEu/edit?usp=sharing&ouid=102632222370728552977&rtpof=true&sd=true návod na zprovoznění ai artu, ale i jazykových modelů doma a technické materiály: https://gist.github.com/racinmat/74d0f08618c1c39edf1e06a0bddd04ce

Mé předchozí přednášky

• přednáška o AI na AnimeFestu 2019, hlavně StyleGAN
  • záznam https://www.youtube.com/watch?v=WWrhXgaoSWM&ab_channel=AnimefestCZ
  • prezentace: https://drive.google.com/file/d/12jW7YbojmWCK6XPFqHpFyANgM6pHlY0X/view
  • zdroje: https://gist.github.com/racinmat/b01d8cdb4f515a136644542dac19a5ea
  • aktualizovaná přednáška pro ČVUT, 2021
    • záznam https://www.youtube.com/watch?v=puXb4ikcaAE
    • zdroje https://gist.github.com/racinmat/d55c7ff4bce4592d48b05058f4b00900
  • aktualizované zdroje a přednáška, advík 2021: https://gist.github.com/racinmat/f654b6a597263f625d8c67f878a348cf
  • aktualizované zdroje a přednáška, anime jikan 2022: https://gist.github.com/racinmat/2f988eef70372ff90679ef88f450b8b8
• přednáška o AI na Animefestu 2023, difuzní modely a LLMs
  • záznam https://www.youtube.com/watch?v=-TEjquDga04&t=4243s
  • Zdroje a odkaz na přednášku: https://gist.github.com/racinmat/20a380a671fe3c84d1cd29741a800bfc
  • aktualizované zdroje a přednáška, advík 2023: https://gist.github.com/racinmat/b50f740fe2ce91325b3bf85497b7c2ea
• přednáška o LLMs na FF 2023 https://gist.github.com/racinmat/bb59ed677015bfb816628f2f6faecc0b
  • aktualizovaná verze, ostrakon 2023 https://gist.github.com/racinmat/6f663fa82c44c8da0d2fcdb241054954
• přednáška o AI artu na FF 2023 https://gist.github.com/racinmat/e9db2957115455d8ecba566e4f430e5c
  • aktualizovaná verze, ostrakon 2023 https://gist.github.com/racinmat/dd04c4df4289bdc546b6d690c93ee6ca
• workshop na stable diffusion, Advík 2023: https://gist.github.com/racinmat/efd25a488cac89f7c8fb65b5e9327452

ne-akademické články

• turing test

  • https://bigthink.com/the-future/turing-test-imitation-game/
  • https://en.wikipedia.org/wiki/Turing_test

• google lamda

  • https://www.wired.com/story/lamda-sentient-ai-bias-google-blake-lemoine/
  • https://www.washingtonpost.com/technology/2022/06/17/google-ai-lamda-turing-test/
  • https://futurism.com/fired-google-engineer-ai-sentience
  • https://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/
  • https://www.scientificamerican.com/article/google-engineer-claims-ai-chatbot-is-sentient-why-that-matters/
  • https://www.theguardian.com/technology/2022/aug/14/can-artificial-intelligence-ever-be-sentient-googles-new-ai-program-is-raising-questions

• počet vrstev deep NN

  • víc než 3 http://wiki.pathmind.com/neural-network#:~:text=More%20than%20three%20layers%20(including,qualifies%20as%20%E2%80%9Cdeep%E2%80%9D%20learning.
  • 3 a víc https://stats.stackexchange.com/questions/229619/minimum-number-of-layers-in-a-deep-neural-network ale včetně vstupní vrstvy je to taky víc než 3

• awesome llms https://github.com/Hannibal046/Awesome-LLM

• word2vec

  • https://en.wikipedia.org/wiki/Word2vec
  • http://www.mlguru.com/cs/2015/03/
  • http://www.fit.vutbr.cz/~imikolov/rnnlm/thesis.pdf

• rlhf

  • https://huyenchip.com/2023/05/02/rlhf.html
  • https://www.reddit.com/r/reinforcementlearning/comments/zqfw7r/why_cant_we_do_supervised_learning_in_step_3_of/
  • http://new-savanna.blogspot.com/2023/02/chatgpt-goes-wild-is-solidgoldmagikarp.html

• transformery https://github.com/gaussalgo/L2L_MLPrague23/blob/main/presentation.pdf

• tokenizace

  • https://blog.quickchat.ai/post/tokens-entropy-question/
  • https://platform.openai.com/tokenizer
  • https://github.com/openai/tiktoken
  • https://openaipublic.blob.core.windows.net/encodings/cl100k_base.tiktoken

• bugy u tokenizeru

  • solidgoldmagikarp https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldmagikarp-plus-prompt-generation
  • pokračování https://www.lesswrong.com/posts/Ya9LzwEbfaAMY8ABo/solidgoldmagikarp-ii-technical-details-and-more-recent
  • https://www.reddit.com/r/countingcirclejerk/comments/11ng6oa/usolidgoldmagikarp_will_be_forever_immortalized/
  • https://www.reddit.com/r/artbn_bots/wiki/rc_hoc_500/
  • https://github.com/artbn/RC/blob/master/hoc.txt

• vysvětlení transformerů

  • http://jalammar.github.io/illustrated-transformer/
  • https://towardsdatascience.com/transformers-explained-visually-not-just-how-but-why-they-work-so-well-d840bd61a9d3

• red pajamas 7B

  • online https://api.together.xyz/open-chat
  • blogpost https://www.together.xyz/blog/redpajama-7b

• superHOT

  • https://kaiokendev.github.io/context
  • https://kaiokendev.github.io/til

• openai modely https://platform.openai.com/docs/models/gpt-4

• prompt hacking https://learnprompting.org/docs/category/-prompt-hacking

  • prompt injection bingu https://greshake.github.io/

• prompt leaking

  • https://matt-rickard.com/a-list-of-leaked-system-prompts
  • https://arstechnica.com/information-technology/2023/02/ai-powered-bing-chat-spills-its-secrets-via-prompt-injection-attack/

• capture the flag hry

  • https://gandalf.lakera.ai/
  • https://pihack.stratosphereips.org/

• hádání filmů z emoji

  • https://www.reddit.com/r/WebGames/comments/136iqpn/puzzlemoji_i_made_a_single_player_emoji/
  • https://flowgpt.com/prompt/qj3XLSqxRq7dLWGpSYJCY

• falcon

  • https://falconllm.tii.ae/index.php
  • https://huggingface.co/tiiuae/falcon-40b
  • online demo: https://huggingface.co/blog/falcon

• google no moat https://www.economist.com/leaders/2023/05/11/what-does-a-leaked-google-memo-reveal-about-the-future-of-ai

• gpt3 http://jalammar.github.io/how-gpt3-works-visualizations-animations/

• gpt4 parameters https://the-decoder.com/gpt-4-has-a-trillion-parameters/

• gpt4o https://openai.com/index/hello-gpt-4o/

• claude 3 https://www.anthropic.com/news/claude-3-family

• chatgpt growth https://explodingtopics.com/blog/chatgpt-users

• deepmind chinchilla https://www.deepmind.com/publications/an-empirical-analysis-of-compute-optimal-large-language-model-training

• llama

  • llama 1
    • https://the-decoder.com/metas-llama-language-model-shows-that-parameters-are-not-everything/
    • llama stable diffusion moment https://simonwillison.net/2023/Mar/11/llama/
    • gpu requirements https://www.reddit.com/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/
    • openllama https://github.com/openlm-research/open_llama
  • llama 2
    • https://llama.meta.com/llama2/
    • https://ai.meta.com/research/publications/llama-2-open-foundation-and-fine-tuned-chat-models/
  • llama 3
    • https://llama.meta.com/llama3/
    • https://ai.meta.com/blog/meta-llama-3/
  • performance https://www.interconnects.ai/p/llama-3-and-scaling-open-llms
  • llama family https://github.com/RUCAIBox/LLMSurvey

• huggingface open llm leaderboard https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard

• hellaswag

  • paper https://arxiv.org/pdf/1905.07830.pdf
  • validační data https://raw.githubusercontent.com/rowanz/hellaswag/master/data/hellaswag_val.jsonl

• GPT4Chan

  • video s vysvěltením https://www.youtube.com/watch?v=efPrtcLdcdM&ab_channel=YannicKilcher
  • huggingface https://huggingface.co/ykilcher/gpt-4chan
  • web modelu https://gpt-4chan.com/
  • zdroják https://github.com/yk/gpt-4chan-public
  • 16bit i 32bit modely dostupné přes torrent
    • https://archive.org/details/gpt4chan_model_float16
    • https://archive.org/download/gpt4chan_model_float16
    • https://archive.org/details/gpt4chan_model
    • https://archive.org/download/gpt4chan_model
  • dataset raiders of the lost kek https://zenodo.org/record/3606810#.YpjGgexByDU

• 4bit operace https://astojanov.github.io/projects/clover/

• autogpt https://docs.agpt.co/

  • zdroják https://github.com/Significant-Gravitas/Auto-GPT

• agentgpt https://agentgpt.reworkd.ai/

• langchain https://python.langchain.com/docs/get_started/introduction.html

  • meetup s ukázkami langchainu https://www.youtube.com/watch?v=fl1jluGawzQ&ab_channel=Applifting

• promptsource https://github.com/bigscience-workshop/promptsource

• deepmind chinchilla https://www.deepmind.com/publications/an-empirical-analysis-of-compute-optimal-large-language-model-training

• právník chatgpt https://www.cbsnews.com/news/chatgpt-judge-fines-lawyers-who-used-ai/

• žaloba proti microsoftu https://www.theverge.com/2023/1/28/23575919/microsoft-openai-github-dismiss-copilot-ai-copyright-lawsuit

• character.ai https://beta.character.ai/

• alpaca https://crfm.stanford.edu/2023/03/13/alpaca.html

  • zdroják https://github.com/tatsu-lab/stanford_alpaca

• vicuna https://lmsys.org/blog/2023-03-30-vicuna/

• guanaco https://guanaco-model.github.io/

• export z character.ai na on-prem

  • https://np.reddit.com/r/PygmalionAI/comments/10h37u4/my_basic_guide_to_pygmalion_for_begginers/
  • https://www.reddit.com/r/CharacterAi_NSFW/comments/112d104/converting_characterai_to_tavern_ai/
  • https://github.com/0x000011b/characterai-dumper

• překlady

  • https://metaroids.com/learn/google-translate-vs-chatgpt-vs-deepl-translator-ultimate-showdown/
  • https://slator.com/tencent-pits-chatgpt-translation-quality-against-deepl-google-translate/
  • WMT konference
    • 2023 https://www2.statmt.org/wmt23/pdf/2023.wmt-1.1.pdf
    • 2023 cuni paper https://www2.statmt.org/wmt23/pdf/2023.wmt-1.8.pdf
    • 2024 https://www2.statmt.org/wmt24/

• DPO https://patmcguinness.substack.com/p/fine-tuning-llms-with-direct-preference

• tokeny a tokenizery

  • https://github.com/pavelkraleu/llm-language-tokens/blob/main/strawberry/language_token_count.png

• openai "vzpoura"

  • https://en.wikipedia.org/wiki/Removal_of_Sam_Altman_from_OpenAI
  • https://www.nytimes.com/2023/11/18/technology/sam-altman-open-ai.html
  • https://en.wikipedia.org/wiki/Ilya_Sutskever
  • https://www.reuters.com/technology/artificial-intelligence/openai-ceo-says-company-could-become-benefit-corporation-information-2024-06-15/

• Elon Musk

  • based-ai https://www.vice.com/en/article/qjkgym/elon-musk-based-ai
  • truth-gpt https://www.theverge.com/2023/4/17/23687440/elon-musk-truthgpt-ai-chatgpt
  • tesla je AI company https://www.fool.com/investing/2024/04/27/elon-musk-called-tesla-an-artificial-intelligence/
  • zákaz AI výzkumu na 6 měsíců, zatímco vyvíjí xAI https://12ft.io/https://www.businessinsider.com/elon-musk-building-ai-venture-while-calling-for-industry-pause-2023-8
  • žádost o pozastavení AI výzkumu https://futureoflife.org/open-letter/pause-giant-ai-experiments/
  • xAI https://www.theguardian.com/technology/article/2024/may/27/elon-musk-xai-raises-openai-funding-round
  • žaloba
    • https://www.reuters.com/legal/elon-musk-sues-openai-ceo-sam-altman-breach-contract-2024-03-01/
    • https://www.theverge.com/2024/6/11/24176462/elon-musk-openai-lawsuit-sam-altman-dropped
    • https://www.ndtv.com/world-news/change-name-to-closedai-elon-musk-tells-sam-altmans-openai-heres-why-5191645
    • https://openai.com/index/openai-elon-musk/
    • https://www.euronews.com/next/2024/03/06/openai-hits-back-at-elon-musks-lawsuit-and-says-he-wanted-full-control-of-chatgpt-maker

• Anthropic - Claude online demo https://claude.ai/new

• jailbreaky

  • https://github.com/verazuo/jailbreak_llms
  • https://jailbreak-llms.xinyueshen.me/
  • https://jailbreakbench.github.io/

• kvantizace

  • https://oobabooga.github.io/blog/posts/gptq-awq-exl2-llamacpp/

papery

• chain of thought https://arxiv.org/pdf/2201.11903.pdf
• position interpolation https://arxiv.org/pdf/2306.15595.pdf
• lora https://arxiv.org/pdf/2106.09685.pdf
  • zroják https://github.com/microsoft/LoRA
• tree of thoughts https://arxiv.org/pdf/2305.10601.pdf
• GPT na překld https://arxiv.org/pdf/2301.08745.pdf

PygmalionAI

• hlavní stránka https://pygmalion.chat/
• discord https://discord.gg/pygmalionai
• dokumentace https://docs.pygmalion.chat/
• neplést s http://pygmalion.ai/, to je něco úplně jiného
• přehled potřebné RAM podle velikosti modelu https://rentry.co/pygmalion-ai
• jeden z návodů na zprovoznění: https://rentry.org/pygmalion-local
  • není úplně aktuální, je třeba nainstalovat u KoboldAI všechny závislosti pomocí install_requirements.bat puštěného jako admin.
• rozcestník na některé guides https://rentry.org/wAIfu_Pygmalion_Guides
• seznam postav a ukázky s nimi https://discord.com/channels/1066323170866495609/1073045743885090846 a mnoho dalších kanálů na tom discordu
• rock lee vs Andrew Tate

web UI

• Tavern AI https://github.com/TavernAI/TavernAI
• oobabooga web ui
  • github https://github.com/oobabooga/text-generation-webui/
  • reddit https://www.reddit.com/r/oobaboogazz/
  • discord https://discord.gg/jwZCF2dPQN
• KoboldAI pro 4bit modely https://github.com/0cc4m/KoboldAI
• doporučený fork KoboldAI https://github.com/henk717/KoboldAI/releases
• silly tavern https://docs.sillytavern.app/
  • zdroják https://github.com/SillyTavern/SillyTavern
• llama cpp https://github.com/ggerganov/llama.cpp
• kobold cpp https://github.com/LostRuins/koboldcpp

ostatní

• galerie promptů:
  • https://prompts.aidg.club/
  • https://aetherroom.club/
  • https://github.com/cr4bl1fe/AIDSPrompts
• seznam rolí/person https://rentry.org/wAIfu_Bot_List_Final#generation-2
• 4chan rozcestník na vtuber chatboty https://boards.4channel.org/vt/thread/52434778
• rozcestník na hodně věcí https://rentry.org/diy-virtualwaifu
• aids wiki https://aids.miraheze.org/wiki/Main_Page
• tips https://rentry.org/pygtips
• koboldai discord https://discord.gg/UCyXV7NssH
• diskuze o instalaci na pygmalionAI discordu https://discord.com/channels/1066323170866495609/1066342105837998120/1125408856336506900
• další tutoriál na zprovoznění https://rentry.org/ipfub
• jailbreaky pro tavern ai https://rentry.org/GPTJailbreakPrompting
• počty parametrů https://www.reddit.com/r/Infographics/comments/1c81n2o/the_size_of_llms_apr_2024/
• benchmarky
  • původní open llm leaderboard https://huggingface.co/spaces/open-llm-leaderboard-old/open_llm_leaderboard
  • neko institute https://huggingface.co/Neko-Institute-of-Science
  • blogpost https://huggingface.co/docs/leaderboards/open_llm_leaderboard/archive
• nový open llm leaderboard https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard
  • blogpost https://huggingface.co/spaces/open-llm-leaderboard/blog
  • https://www.tomshardware.com/tech-industry/artificial-intelligence/chinese-llms-storm-hugging-faces-chatbot-benchmark-leaderboard-alibaba-runs-the-board-as-major-us-competitors-have-worsened
• lmsys chatbot arena https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard
• uvažování a inteligence
  • shift cipher https://arxiv.org/pdf/2309.13638
  • llm reasoning https://slideslive.com/39019842/llms-reasoning-and-the-path-to-intelligence?ref=folder-133929
• RAG
  • https://www.promptingguide.ai/techniques/rag
  • https://talks.karmi.cz/czpycon2023/
• roleplaying
  • persony https://www.characterhub.org/?search=&first=50&topics=&excludetopics=&page=1&sort=default&venus=false&min_tokens=50
  • https://www.reddit.com/r/KoboldAI/comments/117vomu/where_to_download_custom_characters/
  • https://www.dreamtavern.ai/
    • No machine export, need to copypaste to sillytavern manually
    • Guide https://RENTRY.ORG/PYGMABOORU
    • Contains pngs where all data is stored as metadata of the png https://botprompts.net/
    • Contains just jsons https://beta.character.ai/ https://chub.ai/ https://janitorai.com/s
    • Has character definitions enabled
    • Can be downloaded using https://www.reddit.com/r/SillyTavernAI/comments/168ya0v/i_made_a_page_where_you_can_easily_download_card/ just replace janitorai.com with janitorai.me in the url on page of certain character
    • https://janitorai.me/ works too

online překládací a titulkovací program založený na faster-whisper

• https://github.com/ufal/correctable-lecture-translator