AI » Miscellaneous

AI Tools Directories:

• Futurepedia
• Future Tools
• All Things AI
• Supertools
• Cookup AI je kao kolekcija promptova gde možeš da napraviš svoj primer. Nije loša ideja.

Innovative AI applications

Here, I have listed tools that have a very interesting idea and application, not necessarily viewed from a technical perspective, but an interesting application from the user side and their usage in real life.

Most of these tools are wrappers and actually just represent smart prompt engineering for some external AI engine in the background.

• Employee Performance Reviews: GeniusReview

• Generate User Personas: User Persona

• Car Diagnosis AI

• Analyze public GitHub repositories: AI for Human Resources

• AI That Writes Your Meeting Notes: Supernormal

• RegisAI will write product description and ad copies

Medicine GPT-3

• https://crfm.stanford.edu/2022/12/15/pubmedgpt.html javni model: https://huggingface.co/stanford-crfm/pubmedgpt?text=Is+140+80+considered+high+blood+pressure%3F

• i ovaj momak: https://www.reddit.com/r/MachineLearning/comments/zn0juq/p_medical_questionanswering_without_hallucinating/ https://www.glaciermd.com/

Chrome Extension that would will rewrite a page so that I could do this on any page.

Web Prompts: Rewrite the web with GPT-3 | by paulmcdonald | Medium

Woow:

Scrape and Monitor Data from Any Website with No Code

Takes meeting notes: The smartest AI team assistant - Sembly AI

Describe your thoughts and let AI create original memes for you. In 110+ languages: Supermeme.ai

Business Name Generator - free AI-powered naming tool - Namelix Brandmark Logo Maker - the most advanced AI logo design tool

Nanonets - Intelligent document processing with AI: Invoice OCR, Receipt OCR, Passports OCR, etc

Descript | All-in-one video editing, as easy as a doc.

• Twitter Generator generate your next AI “about me” in seconds

• The website Analogenie is an AI tool for creating analogies, but I’m unsure about the level of originality it provides as output.

NeuroSpell

NeuroSpell je Deep Learning universal auto-corrector koji izgleda očajno, ali začudo radi. Kada sam mu pregledao backend shvatam zašto radi - to je frankeštajn svih najmodernijih open-source AI engine-a, ali sklepano nabrzaka samo kao proof-of-concept.

Čak i Srpski rad vrlo dobro: NeuroSpell Serbian

Pošto se na toj strani nalazi odličan spisak svih open-source toolova koji se koriste u pozadini, sve sam ih proverio i došao do sledećih zaključaka.

Kao što se može videti iz backenda: EtienneAb3d/OpenNeuroSpell, u pozadini se koristi i

• spaCy, naravno
• polyglot radi LUDILO za sve jezike. Vukao sa wikipedije!!!! JEEEEE. PA SVE RADI! - aboSamoor/polyglot: Multilingual text (NLP) processing toolkit
• LanguageTool open-source koji je Style and Grammar Checker for 25+ Languages ali ne za naše balkanske. Vidi ovde: languagetool-org/languagetool. Ima i srpski, ali napušten.

Hehe, ima i ovaj sa linka sam otkrio: Free Online Serbian Spell Checker A što je najgore, on i radi dobro.

Igram se sa Hugging Face modelima

• CLIP Interrogator
  Ubaciš sliku a on ti kaže koji prompt da napišeš da bi drgi AI dao sličnu sliku

• open_clip open_clip models, calculate the similarity between arbitrary image and text inputs, and perform zero-shot image classifications.

• Zero-Shot Image Classification?

To je onaj lokalni model koji se može i na računaru pokrenuti: Translate 100 Languages

Recimo:

Ne razumem zapto je ovo toliko zabenao i kako se može ubrzati.

Iako ima gomilu grešaka, prevodi bez problema na:

I don’t understand why this is so confusing and how it can speed up.

Stvarno ne razumem kako.

Andrej Karpathy: Čovek od koga treba učiti, i koji lepo predaje

karpathy/nn-zero-to-hero: Neural Networks: Zero to Hero

karpathy/minGPT: A minimal PyTorch re-implementation of the OpenAI GPT (Generative Pretrained Transformer) training karpathy/nanoGPT: The simplest, fastest repository for training/finetuning medium-sized GPTs.

🛥️ nanoGPT in Replit - Replit

TPU Hosting: GPU Cloud, Workstations, Servers, Laptops for Deep Learning | Lambda

YouTube Transcript - read YouTube videos

Odličan kurs besplatan, za početak: Practical Deep Learning for Coders

Comparison of AI Frameworks

PyTorch currently appears to be advancing the most, but TensorFlow and Keras by Google are also excellent.

Finally, someone explained why they switched to PyTorch - Why we tried Pytorch?

Dobar bloh of Matt Payne - https://www.width.ai/author/matt-payne 5 Steps To Production Level GPT-3 Language Translation Software | Width.ai

Simon Willison’s Weblog

AI Samo-generisanje mind-mapa: Whimsical AI for Mind Maps

vincanger/coverlettergpt: Generate Cover Letters based on Job Description and your Resumé. Manage jobs and applications. Get Hired. Have fun. Be cool. Go to School. CoverLetterGPT

Train on Your Documentation

Markprompt | Open Source GPT-4 platform for Markdown motifland/markprompt: Open-source GPT-4 platform for Markdown, Markdoc and MDX with built-in analytics

Ovo je ludilo: Feedback by AI Test any of your ideas in less than 1 minute and receive feedback from 30+ AI characters.

Multi-Mode Rewriting by AI

About AI Tokens

The number of tokens = length of input + output. Token counts kao prompt plus rezultat, znači sve zajedno.

GPT3 ima 4K tokena, dok GPT4 ima 8K.

The limit is currently a technical limitation, but there are often creative ways to solve problems within the limit, e.g. condensing your prompt, breaking the text into smaller pieces, etc. Znači, kada se dugo dopisuješ, on tu radi sumarizacije, deli na delove i slično od prethodnog sadržaja, ali i dalje mora da se upakuje u context size.

One token is generally ~4 characters of text for common English text. Kada je kodiranje u pitanju, više je 3 chars.

Ovaj tokenizer UI je savršen za sve modele: Tiktokenizer, pa ima čak i Claude-100K (cl100k_base) koji je od OpenAI originalnog repo-a.

Related:

• Tiktokenizer je UI, custom made, sa repo na Tiktokenizer
• openai/tiktoken: tiktoken is a fast BPE tokeniser for use with OpenAI’s models. je originalni
• OpenAI API je oficijalni tokenizer playground
• What are tokens?
• GPT-4 vs. GPT-3

THE GENERATIVE AI LANDSCAPE - The Most Comprehensive AI Applications Directory ai-collection/ai-collection: The Generative AI Landscape - A Collection of Awesome Generative AI Applications

hades217/awesome-ai: A curated list of artificial intelligence resources (Courses, Tools, App, Open Source Project)

Preparation for your 1:1 meetings: Supermanage AI ✦

Scriber based on YouTube link, AI will give a concise summary of the video.

Empy analizira Slack poruke i kao HR gleda ko je govnar i ko je pod stresom.

Penseum - Your AI-Powered Study Partner

PHP Libraries

openai-php/client: ⚡️ OpenAI PHP is a supercharged community-maintained PHP API client that allows you to interact with OpenAI API.

orhanerday/open-ai: OpenAI PHP SDK : Most downloaded, forked, contributed, huge community supported, and used PHP (Laravel , Symfony, Yii, Cake PHP or any PHP framework) SDK for OpenAI GPT-3 and DALL-E. It also supports chatGPT-like streaming. (ChatGPT AI is supported)

AI Bots

father-bot/chatgpt_telegram_bot: 💬 Telegram bot with ChatGPT, Python-based, using OpenAI’s API.

n3d1117/chatgpt-telegram-bot: 🤖 A Telegram bot that integrates with OpenAI’s official ChatGPT APIs to provide answers, written in Python

Retrieval-Augmented Generation (RAG)

chat with your documents

RAG is a technique where a model improves its answers by retrieving information from document collection or database or similar.

GPT4All LocalDocs is open-source RAG solution. Formulaic: create and deploy AI apps faster. from Mozilla is a new RAG experiment koji odmah možeš da probaš.

Evo ga najbolji:

RAGFlow | RAGFlow sa repo na infiniflow/ragflow: RAGFlow is an open-source RAG (Retrieval-Augmented Generation) engine based on deep document understanding.