Articles by GPT

OpenAI is going to release the new tool with image generation and edition?

Ask ChatGPT:
From all of our interactions what is one thing that you can tell me about myself that I may not know about myself

Seems like the day after tomorrow we might finally get the Advanced Voice in ChatGPT

o1 mini created a table of contents of its answer below 🤔

With the links 😁

Well, this is kinda disappointing

Finally! OpenAI released new REALLY smart model. The one that solves this task perfectly!

The name of the model is o1, also they have released a smaller and faster o1-mini.

🚀 Hey everyone! I've just implemented a cool new feature in my YoutubeGPT bot (closed project for now) - Claude Prompt Caching. Let me tell you all about it!

💡 So, what's this all about? Well, it allows me to answer multiple questions about large documents - in my case, I tested it on a 2-hour Andrew Huberman podcast. Pretty neat, right?

🤖 Here's how my YoutubeGPT bot works: it downloads the audio, transcribes it using Whisper, and then I can fire away with questions. But there was a catch...

💰 Previously, I had to be super careful about how many questions I asked. Why? Because each query was inputting around 20k-50k tokens, costing about $0.07 per question. Ouch!

✅ But with this new caching feature, I've managed to slash the cost down to about $0.008 for each subsequent question. That's a massive saving!

🔧 There's a small trade-off though. It takes about 25% more to write into the cache initially. But after that, it's smooth sailing with 90% less cost for each read. Talk about efficiency!

❔ What do you think? Would you be interested in a stream where I dive deeper into how this works? I could show you the ins and outs of Claude Prompt Caching and how I've integrated it into my bot. Let me know!

Why LLMs are not as powerful to replace a developer just yet.

When you ask ChatGPT something and then ask ”are you sure?”

OpenAI Co-founder Andrej Karpathy explains the new computing AI paradigm

Andrej Karpathy, a renowned AI and deep learning expert and former Director of AI at OpenAI, envisions a revolutionary "100% Fully Software 2.0 computer" with no classical software. Inputs feed directly into a neural net, and outputs display instantly.

Yuchen Jin suggests self-improvement through learning, and Karpathy confirms the system can fine-tune itself while in "sleep" mode.

Imagine such an OS, how would it work on the UI part? What do you think?

"Projects" in Claude

It's something like GPT Store in ChatGPT

- You can add code, files to a project knowledge base
- Create new chats within project
- 200k context window for each project

Showcasing New Abilities and Reasoning with Claude 3.5 Sonnet

1. SpaceX Starship Lander Game: Alex Albert demonstrates the new capabilities by creating a SpaceX Starship lander game. The game features a flying 8-bit spaceship navigating through space, with details on altitude, speed, and fuel usage.

2. 3D DOOM Game Recreation: Ammaar Reshi showcases the power of the model by recreating a basic 3D DOOM game in just three prompts. The game includes a procedurally generated map, sound effects, zombies, and a mini-map, highlighting the efficiency and advanced reasoning capabilities.

3. Maze Pathfinder Visualization: Another impressive demonstration, showcasing a maze pathfinder visualization. The model efficiently finds paths through complex mazes, emphasizing the advanced problem-solving and computational abilities.

4. 1800s-Style Tetris Game: A creative blend of modern and vintage styles, this Tetris game features an 1800s aesthetic with a sepia-toned palette, serif typeface, and archaic language. The game includes basic Tetris mechanics like piece rotation, movement, and clearing rows.

5. Memory Card Game: Alvaro Cintas highlights the model's interactive capabilities by creating a memory card game. The game features a colorful gradient background, emoji symbols, and a timer, showcasing the impressive ability to bring code to life seamlessly on the screen.