Hello, Starters!

We're already getting too used to our classic models, chatbots, and image generators. It's always good to look around and see what new offers are coming up in the AI scene, and we assure you, they're worth your time!

Here’s what you’ll find today:

• GitHub Models are here for devs

• MIT’s novel AI calibration method

• Black Forest Labs unveils FLUX

• OpenAI is working on a ChatGPT detection tool

• Nvidia’s latest chips are getting delayed

• And more.

🧑‍💻 GitHub Models are here for devs (4 min)

Developers are now getting help from GitHub to become the AI engineers of the future. The platform has recently launched a built-in playground where users can access leading models in the industry and harness them to develop their AI applications. The offerings go from GPT-4o to Mistral Large 2, providing an innovative perspective for coding.

This approach is just another example of the democratisation of AI. GitHub users will be able to employ models from companies like Meta, Mistral, and Microsoft, to name a few, to uncover novel ways in which AI can aid in the development process, leading to groundbreaking advancements as we inch closer to AGI.

🌡️ MIT’s novel AI calibration method (3 min)

Although most models have become improved enough for us to rely on them, there are still certain tasks or queries where we might want to double-check the information they're providing. That's why a research group at MIT has come across a calibration method, dubbed Thermometer, that can prevent models from giving incorrect outputs.

"Thermometer" consists of building a smaller model that assists the main one and provides the right "temperature" to improve its accuracy when facing a new task. This technique is based on the 'temperature scaling' calibration method and shows great results in reducing computational costs.

🖼️ Black Forest Labs unveils FLUX (3 min)

Black Forest Labs is a German startup founded by some of the minds behind the technology of Stable Diffusion, and they've brought a new contender to the image generation arena: FLUX.1. This text-to-image model is built in a hybrid architecture that mixes both transformer and diffusion techniques to reach incredible results.

According to the company, FLUX.1 surpasses offerings like Midjourney or DALL-E, as it showcases improved image quality and correlation to text prompts. Also, results note that the model is slightly better at producing hands, an issue that constantly haunts most AI image generators.

💬According to reports from the Wall Street Journal, OpenAI has an AI-text detector in the works, distinct from the one they retired last year. This version provides better results at discovering changes like paraphrasing; however, the company isn't completely convinced about its launch, as it may hurt its business model and unfairly affect certain users.

⏳Being the main provider in the AI era comes at a cost. Many companies rely on Nvidia for their innovations, and they'll now have to wait until early 2025 to get ahold of the company's Blackwell chips, as sources claim they've encountered design issues late in the production process, pushing back the initial shipping date.

⚡️Quick links

Neuralink implanted second trial patient with brain chip, Musk says

Stability AI speeds up 3D image generation with Stable Fast 3D

Microsoft says OpenAI is now a competitor in AI and search

The EU’s AI Act is now in force

Meta is reportedly offering millions to use Hollywood voices in AI projects

Thank you for reading!