Weekly Tech News of July Third Week 2024

Project Strawberry: OpenAI’s Next Big Leap in AI Reasoning

OpenAI is enhancing its AI’s reasoning abilities by itself using the project called “Strawberry. Bloomberg and Reuters have recently revealed about this initiative, based on the Q* algorithm with the aim of enhancing AI’s planning and online research skills.

Using the Strawberry mechanics, you can navigate the internet autonomously for “deep research. Plus, can improve your performance using the novel post-training method even after initial training.

Within this OpenAI demonstrated new human-like reasoning skills you can upsurge your AI’s reasoning productivity. This effort is part of a broader trend in the AI industry to enhance reasoning capabilities which is essential for complex problem-solving and scientific discoveries.

Official Source:

https://www.reuters.com/technology/artificial-intelligence/openai-working-new-reasoning-technology-under-code-name-strawberry-2024-07-12/

Prover-Verifier Games— OpenAI’s Latest Update: Enhance Your Large Language Model (LLM) Outputs

Enhance the legibility of your large language model (LLM) outputs, especially the math problems, using the Prover-Verifier Games that enhance not only outputs but also clarity and verifiability.

The name of this latest update is inspired by a game-theoretic framework that will allow you to optimize your LLM’s accomplishment solely for correctness and can reduce legibility as well. This training algorithm carries the provers that generate correct, legible solutions and produce sneaky & misleading solutions.

This iterative process strengthens the verifier’s ability to distinguish between correct and incorrect answers. In addition to this, it improves human evaluators’ accuracy and highlights that make a balance between performance and legibility and finally drives a clearer and more reliable AI-generated text that benefits various complex tasks.

Official Source:

https://openai.com/index/prover-verifier-games-improve-legibility/

GPT-4o Mini: OpenAI’s Latest and Fastest Model Debuts on Azure AI

OpenAI’s GPT-4o mini is now available on Azure AI to enhance the performance and safety features. It is significantly smarter than GPT-3.5 Turbo and is more cost-effective with having a context window of 128K and improved multilingual capabilities as well.

GPT-4o mini supports text processing with future updates for image, audio, and video. Azure AI includes default safety features, data residency across 27 regions, and global pay-as-you-go availability.

This model delivers high throughput and flexibility that support various applications, like GitHub Copilot. Its fine-tuning and batch services ensure its capabilities, like high-quality, cost-effectiveness, AI deployments for over 53,000 customers, Fortune 500 companies, and more.

Official Source:

https://azure.microsoft.com

Meet AlphaFold 3: An AI Predictor that will foretell about life’s structure and interaction

Google DeepMind’s AlphaFold 3, launched in May is consistently extending its capabilities and reach to a level where it can predict the structure and interactions of all molecules of human life such as DNA, RNA, and ligands. This approach or update is recognized with the name “AlphaFold 2”

It aids in drug design and genomics research and tackles complex biological interactions. Researchers access AlphaFold 3 via AlphaFold Server, a tool that generates molecular structures from user-provided sequences.

The model’s training data now includes diverse biomolecules that enhance their accuracy. AlphaFold 3 simplifies the process of handling varied molecule types just by using a diffusion-based generative model. This advancement promises significant progress in fields like drug design and genomics.

Official Source:

https://blog.google/technology/ai/how-we-built-alphafold-3/?utm_source=tw&utm_medium=social&utm_campaign=og&utm_content=&utm_term=