Open Source LLM war begins.

PLUS : OpenAI version of Search Engine is getting ready.

Greetings, AI Enthusiasts

AI giants are getting into the OpenSource LLM wars with their new LLM models.

OpenAI is trying to dethrone “Google - The Search engine King”

In addition, there are more interesting news, tools, chart and prompt this week.

Lets explore!

AI news and Insights:

  • OpenAI’s search engine

  • Open source LLM models

  • AI in Paris 2024 Olympics

  • New no-code app

  • StackOverflow survey about AI use.

  • More interesting news.

  • Free AI tools

OpenAI’s new search Engine : SearchGPT

Video Credit : OpenAI

✅️ OpenAI was in the news before for trying to dethrone "Google - The Search Engine King". With SearchGPT, OpenAI promises to redefine online search and has put a stop to all the search related speculations

The Essentials:

  • SearchGPT will provide immediate responses to user queries.

  • Let users ask more questions to get clear information.

  • Shows the right source of its information responses.

  • Can create pictures, charts or graphs to explain things.

  • Works with content creators to present the best and the latest information.

The Takeaway:

Currently, SearchGPT is in the Prototype phase. If you are curious to test, sign up on the OpenAI website, to get early access.

Meta’s Open-source LLM Model : Llama 3.1 405B

Image Source : Rowan Cheung

Meta has released the most powerful Open-source LLM model Llama 3.1 405B. It is the rival to the closed AI model's capabilities and represents a significant advancement in AI.

The Essentials:

  • It is the world's largest Open-source AI model with 405B parameters.

  • The 8B and the 70B models are upgraded with 128K context length.

  • It provides Multilingual support and improves its reasoning capabilities.

  • It has been trained with over 15 trillion tokens.

The Takeaway:

Meta is democratizing AI with the introduction of Llama 3.1. Although, Meta offers Llama3.1 405B in Open format, but it requires significant computational resources for deployment and fine-tuning the model.

Mistral’s Open-source LLM model : Mistral Large 2

Image Source : Mistral

Mistral Large 2, the flagship language model of Mistral is released. It sets a new benchmark in performance-to-cost ratio.

The Essentials:

  • 123 billion parameters

  • 128,000 token context window.

  • Enhanced code generation and improved reasoning.

  • Supports dozens of languages, along with 80+ coding languages including Python, Java, C, C++, JavaScript, and Bash.

  • Mistral Large2 is not fully open-sourced like Llama3.1 405B.

The Takeaway :

Mistral Large 2 is available for research and non-commercial use. A commercial license is required for business applications.

NVIDIA’s two new LLM models: Minitron 4B and 8B

Image Source : Pavlo

NVIDIA released two new models Minitron 4B and 8B derived from their bigger model Nemotron-4 15B. These two AI language models are smaller yet powerful.

The Essentials:

  • It only requires 40 times fewer training tokens when compared to training a model from scratch.

  • Less training results in 1.8x compute cost savings.

  • Minitron 4B/8B models show a 16% improvement in MMLU benchmark scores.

  • They perform comparably with other models like Mistral 7B, Gemma 7B and Llama-3 8B.

The Takeaway:

NVIDIA pruned the bigger model Nemotron-4 15B by removing less important parts and then distilled it to a more concentrated form, increasing efficiency.

Paris 2024 Olympics are not just sports; they are a showcase of cutting-edge AI technology.

The Essentials:

  • Countless CCTV cameras are set up to monitor the Olympic activities

  • Intel is providing 5G, AI, and 3D imaging tech for enhanced logistics, broadcasting, and fan experience.

  • Omega, the official timekeeper, uses AI for precise timekeeping and capturing precise moments of Athlete's performances.

  • Google is partnering with NBC Universal and Team USA to integrate AI in live broadcasts and enhance user searches related to the Olympics.

The Takeaway :

AI will play a significant role along with Athletes at the Paris Olympics. Major Tech companies have joined together to enhance user experience though user privacy is still an important issue to address.

No code AI app from a unicorn

Image Source : Airtable

You can build an app with no coding experience. AI will build apps in seconds. $11 billion dollar unicorn Airtable's new AI tool named Cobuilder is a game-changer to build apps.

The Essentials:

  • Describe your app idea in plain English, Cobuilder turns into a real working app in seconds. You can also edit and tweak the app based on your needs.

  • With Natural Language Processing (NLP), user input is interpreted and fed to LLM to turn inputs into actual code.

  • Integrates with Airtables database, workflow and templates to build an app.

  • Airtable ensures that customer data is not retained anywhere in the process.

The Takeaway:

With Cobuilder you might not need any coding experience to build simple apps, but for complex, custom-coded apps, coding cannot be replaced just yet.

StackOverflow survey about AI use

Image Source : StackOverflow

70% of professional developers believe AI is NOT a threat to their jobs.

Key numbers :

  • AI Tool Usage:

    • ↑ Increased from 24.88% (2023) to 34.14% (2024)

  • Most Popular AI tool:

    • ChatGPT - 76% of developers.

    • Github Copilot - 40% of developers

  • Generative AI Adoption:

    • 70% of professional developers use it.

  • Productivity:

    • 1.5 - 2 hours saved per day.

  • Workplace Policies:

    • 62% have AI usage policies

    • 33% encourage AI tool use

    • 29% restrict or prohibit AI tool use

  • Trust in AI :

    • ↓ 34% trust AI for secure code

    • ↓ 31% trust AI for accurate code

Interesting News

Apple Pledge to keep AI safe:

Apple has joined other tech companies in signing a voluntary pledge towards responsible development of AI. This includes watermarking AI-generated videos and conducting security testing.

Elon Musk is using users tweets to train xAI

Elon Musk, without informing the users, or regulators, have started using user's data for training xAI possibly Grok2 and 3. Now there is an option for users to opt out of this data sharing through the X settings. This has raised concerns about data privacy and compliance with EU regulations.

JP Morgan is launching a new Generative AI tool

JP Morgan is launching a Generative AI tool for employees. This tool will help employees to summarize documents, help in writing and generate new ideas.

3 Free AI Tools for you to use

  • Udio - To create custom music tracks for any topic or genre

  • Invideo - Produce videos using text prompts.

  • OpenArt - Create digital art from text prompts.

  • OpenArt Sketch to Image - Turn original drawings into full digital art.

Magic Prompt

Image Source : @fotomario

An awe-inspiring 3D render of the Paris 2024 Olympics logo, set against the backdrop of a fiery, burning Olympic flame. The flamesdance and intertwine to form the iconic Olympic rings and typography, while the Eiffel Tower stands tall in the distance. The cinematic-style poster is rich in color and detail, evoking a sense of excitement and anticipation for the upcoming games., poster, typography, 3d render, cinematic, vibrant, ukiyo-e.

FEEDBACK

Your feedback is precious and helps us tailor future newsletters to better suit your needs!

We welcome your feedback and insights. Please reply to this email with any comments or suggestions you'd like to share.

Appreciate your time.

AI-ways forward,Naveen Krishnan,Founder and Writer,The AI Catalyst Solutions Inc.