What a wild week in AI 🤯

- Mistral OCR
- Google’s AI Mode
- Windsurf Previews
- Anthropic Console
- ChatGPT Edit in IDEs
- Microsoft Dragon Copilot
- HunyuanVideo I2V Model
- Sesame Realistic AI Voices
- Alibaba releases QwQ-32B

Here’s everything you need to know: 1. Mistral AI has launched Mistral OCR, a new API designed for document understanding.

This tool extracts text from images and PDFs with high accuracy, making it ideal for use with RAG systems.

It’s priced at 1000 pages per dollar and is already available.

https://twitter.com/sophiamyang/status/1897713370029068381

You can now clone any website just by writing a prompt.

Simply add the new Firecrawl MCP server to your favorite AI coding tool for improved web data extraction, and let Claude code it for you.

Here’s how: First, you are going to need an API key from Firecrawl. You can get one to try free here: firecrawl.dev

Real-time AI voice agents for businesses are here.

I created this AI assistant in minutes with no code and used my cloned voice to help handle customers at a fictitious store.

Plus, there’s now a marketplace with ready-made templates where you can sell yours. First, head over to:

Create an AI voice agent by adding its name, a voice, and instructions of how it should behave via the “Prompt” section. synthflow.ai

What a wild week in AI 🤯

• OpenAI GPT 4.5
• Claude 3.7 Sonnet
• Alibaba Wan 2.1 AI video
• Grok 3 & Perplexity Voice
• Google Gemini Code Assist
• Amazon AI assistant Alexa+
• ElevenLabs Speech-to-Text
• Hume AI new LLM built for TTS

Here’s everything you need to know: 1. OpenAI has launched GPT-4.5, its largest AI model yet.

It focuses on unsupervised learning, improving intuition and reducing hallucinations.

Early testers say interactions feel more natural. Currently available to Pro users.

https://twitter.com/openai/status/1895219591070261266

Grok 3 is an incredible AI coding assistant.

After a couple of hours and +1000 lines of code generated, I have now a fully functional 2D vertical jumping game.

With different heroes, monsters, platforms, difficulties, and lives.

Here are the prompts and process I followed: 🧑‍💻CODING PART

I will provide the prompt below, let me give you a few tips:

- Don’t try to ask for every single detail and feature. Try to first ask for a very simple game.
- For complex tasks, use the Grok Think button, it’s quite good.
- Ask for shapes as your characters, platforms, etc. Do not worry about the assets at first.
- You will probably get some errors. Just ask @Grok to solve them for you if you don’t know.

Once you have a first version working, you can start adding the images and other features.

In my case, I added the settings, lives, score, animations, different platforms, main menu and enemies later.

OpenAI has released a new prompting guide for their reasoning models.

It emphasizes simplicity, avoiding chain-of-thought prompts, the use of delimiters, and when to use them.

Here’s a breakdown and an optimized prompt to have it write like you:
ChatGPT models excel in different ways:

🔹 o-series (“the planners”)

• Think deeply for complex tasks
• Great at strategy, planning & decision-making
• Ideal for expert-level fields: math, science, finance, law

🔹 GPT (“the workhorses”)

• Optimized for speed & efficiency
• Excels at straightforward execution • Best when cost & speed outweigh perfect accuracy

Windsurf just announced Wave 3.

It takes AI coding to a whole new level with so many incredible features.

8 wild examples: 1. Tab to Jump

The autocomplete model is getting a huge upgrade.

Windsurf can now anticipate your next cursor position and navigate through your file by just pressing tab.

What a crazy week in AI 🤯

- Mistral Le Chat app
- Anthropic Jailbreaks
- Pika adds Pikaddition
- Google Gemini 2.0 Pro
- Replit free text-to-app
- ByteDance’s AI avatars
- OpenAI’s Deep Research
- HuggingFace AI App Store

Here’s everything you need to know: 1. Mistral has updated its famous “Le Chat” AI platform. It can Search the Web, create images, and analyze uploaded documents.

They also just released its new phone app for Android and iOS

https://twitter.com/mistralai/status/1887517520040448510

I just created my own web AI agent.

I can give it a topic or a news URL and it will take over my computer to look it up, summarize it, and draft a post about it in my own writing style.

All on its own 🤯

Here’s how you can do it too: For this, I used Stagehand, which is the AI interface to the internet.

It is the easiest way to build browser automations. You can find more info here: and here are their tags @Stagehanddev @browserbasehq
 
But let’s get you started 👇 stagehand.dev

OpenAI’s Deep Research is a major shift.

It’s like having a PhD-level assistant that can perform deep web analysis and generate reports in minutes.

But it’s limited to Pro Users…

Here are 6 free open source alternatives 🧵: 1. Open Deep Research

https://twitter.com/nickscamara_/status/1886459999905521912

ChatGPT released a new beast model.

o3-mini is a fast reasoning model that can search the web, and is particularly good at science, math, and coding.

10 wild examples: 1. New AidanBench records

AidanBench rewards Creativity, Reliability, Contextual attention, and Instruction following of LLMs.

https://twitter.com/aidan_mclau/status/1885559328662573242

What a crazy week in AI 🤯

- o3 models
- Krea AI Chat
- Mistral Small 3
- Pika 2.1 AI Video
- Kimi Reasoning model
- Alibaba Qwen2.5-Max
- Perplexity Deekseek R1
- DeepSeek new Multimodal
- Google AI Weather Forecast

Here’s everything you need to know: 1. OpenAI just released o3-mini and it comes in two flavors.: o3-mini and o3-mini-high.

They are particularly strong in science, math, and coding. and they are currently rolling out.

Free users can also try it out by selecting the Reason button.

https://twitter.com/openai/status/1885406586136383634

Everything you need to know to master Windsurf AI.

A full guide 🧵 1. Windsurf AI Overview

Windsurf AI is a code editor with advanced AI capabilities, increasing developer productivity by integrating AI assistance into your coding workflow.

It can both collaborate with you like a Copilot and tackle complex tasks independently like an Agent.

What a crazy week in AI 🤯

- OpenAI Agents
- Stargate Project
- Claude Citations
- Freepik Imagen 3
- DeepSeek-R1 model
- Perplexity AI Assistant
- Gemini 2.0 Flash Thinking
- Tendent 3D Asset Creation
- ByteDance Reasoning Agent

Here’s everything you need to know: 1. OpenAI just announced Operator.

Operator is an AI agent that can go to the web to perform tasks for you. It can shop for airline tickets, making restaurant reservations, and more.

It’s now available to Pro users.

https://twitter.com/openai/status/1882509286439637448

What a huge moment for open source.

DeepSeek has released a model rivaling OpenAI’s o1 model.

Not only that, they fine-tuned smaller models like Llama 8B, achieving a similar or even better performance than GPT-4o or Claude 3.5 Sonnet!

And you can run them locally now 🧵:
This means that you can now run a really powerful AI model offline, allowing more privacy and accessibility at all times, for free.

To run it locally, one of the easiest ways is to download LM Studio

Windsurf just announced Wave 2.

It takes AI coding to a whole new level with so many incredible features.

8 wild examples: 1. Web Search

The AI assistant (Cascade) can now search the web so you can pull in the latest API docs or the changelog of your favorite open source project.

You can also use @ URL to specify any website.

What a crazy week in AI 🤯

- Luma Ray2
- ChatGPT Tasks
- Runway Frames
- Sky-T1 matches o1
- MiniMax-O1 models
- Microsoft MatterGen
- MiniMax Text to Audio
- Mistral’s new Codestral
- Krea 3D Object Creation

Here’s everything you need to know: 1. Luma has released Ray2, a large-scale video generation model that produces highly realistic visuals with smooth, natural motion.

The model demonstrates a strong ability to comprehend text instructions and can also work with both image and video inputs.

https://twitter.com/lumalabsai/status/1879592852151558258

ChatGPT’s new Tasks feature has unlocked so many incredible use cases.

It allows you to set up recurring actions and tasks, essentially giving you an AI agent that works for you in the background.

10 powerful ideas: 1. Start your day with a curated global news summary at 8 AM.

The world’s most influential tech event is taking place this week.

It’s insane what’s being showcased at CES 2025.

Here are 14 mind-blowing reveals:

1. XPeng Aero HT presents the Land Aircraft Carrier, a modular “flying car”
2. Portalgraph: A 3D projector that projects VR space into the real world

You can now create realistic AI characters that narrate your script in seconds.

It’s completely free to even create a custom avatar of yourself.

Here’s how: STEP 1.

Head over to:

In the main dashboard, you’ll see tons of AI avatars to choose from. humva.ai

Gemini’s Stream Realtime has unlocked so many incredible use cases.

You now have an AI assistant that can see your screen and chat with you in real-time to learn, work, and research faster.

7 powerful ideas: 1. Research Assistant

- Highlight a dense paragraph from a white paper and ask for a non-technical summary.
- Hover over a tricky term, formula, diagram and ask for a simple explanation.
- Open multiple tabs on a research topic and ask to synthesize the key points side by side.