• AI Emergence
  • Posts
  • OpenAI’s big bets - Project Strawberry and Orion (next frontier model)

OpenAI’s big bets - Project Strawberry and Orion (next frontier model)

Along with: What’s your view on the California AI regulation bill?

Hey there, 

This week looks like one of those weeks that could have a huge impact on the direction of AI in the coming years. Why do I say so?

First, OpenAI’s big bets were in the limelight. And you would remember I think the next frontier model from OpenAI could differentiate hype vs. reality.

Second, the California AI regulation bill is close to becoming a law and if it gets implemented as is, many feel that it would stifle innovation. And it is happening in a place that hosts the biggest players in AI. 

I am watching out this space - how about you?

What would be the format? Every week, we will break the newsletter into the following sections:

  • The Input - All about recent developments in AI

  • The Tools - Interesting finds and launches

  • The Algorithm - Resources for learning

  • The Output - Our reflection 

  • Question to ponder before we meet next!

Table of Contents

According to The Information, there have been more updates on Project Strawberry, a potential technical breakthrough aimed at improving its model's ability to solve math problems- a task that has traditionally posed significant challenges for large language models (LLMs).

Employees at OpenAI have also used it to solve the NYT Connections - a word puzzle game. 

The model is reportedly designed to take time to "think" through problems. So it would take more time to think but the results would be much better than traditional models.

But what’s more? 

While Strawberry excels at language model improvement and mathematical reasoning, the report mentions a new model named Orion, the next frontier model for OpenAI. One of the main purposes of Project Strawberry is to help generate high-quality training data and reduce issues related to hallucinations.

Both these updates are expected to come in the Fall of this year. (source)

California is actively pursuing SB 1047, a trailblazing AI safety bill, amid a backdrop of heated debate among tech giants, policymakers, and AI researchers. 

The bill mandates stringent measures for AI developers spending over $100 million, including safety testing, third-party audits, and the implementation of a kill switch. 

Despite facing opposition from industry leaders like Google and Meta, who argue it stifles innovation, the bill has garnered support from figures such as Elon Musk and organizations expressing cautious optimism about its amendments. 

As it moves through legislative hurdles, Senator Scott Wiener champions the bill, highlighting its necessity in regulating AI akin to other significant technologies, amidst a broader call for a unified federal regulatory approach. (source)

Google rolls out custom AI experts and advanced image generation in Gemini update

Google Gemini has introduced some new features that let users customize Gemini to create their own personal AI experts on any topic they want, starting with the introduction of Gems and the Imagen 3 image generation model. 

Gems, a customizable tool within Gemini Advanced, Business, and Enterprise, allows users to create personal AI experts tailored to specific topics, from coding to career advice. These virtual experts can assist with projects, brainstorming, and routine tasks, enhancing productivity across various applications. 

Concurrently, Imagen 3 is set to enhance creative capabilities by generating high-quality images from textual prompts. This latest model promises improved realism and artistic versatility, expanding its deployment to more users and languages shortly. (source)

Anthropic launched Artifacts in June as a feature preview and now it is finally available to all the users across their free, Pro, and Teams plans.

You can now create and view Artifacts on both iOS and Android apps.

Artifacts provide a dedicated space to instantly view, refine, and expand upon the work you create with Claude which is pretty amazing for prototyping. (source)

Meta’s Sapiens models revolutionize human-centric vision tasks with transformers

After announcing SAM 2, Meta introduced us to Sapiens, a family of vision models optimized for human-centric tasks, including 2D pose estimation, body part segmentation, depth estimation, and surface normal prediction. 

These models leverage transformer attention mechanisms and multi-headed self-attention to precisely process high-resolution images. 

Meta reports that Sapiens consistently outperforms previous models in crucial tasks such as pose estimation, motion capturing, and depth prediction, which is setting new benchmarks in the field. (source)

Microsoft is planning to allow developers to use Recall AI by October which has been delayed because of some controversies in the security and user experience. 

The feature, intended for Windows 11, captures nearly everything on the user's PC, allowing for searches and reviews of past activities via an explorable timeline. 

As per the statement by Pavan Davuluri security enhancements like optional activation, database encryption, and Windows Hello authentication, are in progress. (source)

Salesforce has launched a new suite of open-source large multimodal AI models named xGen-MM (or BLIP-3), marking a significant step forward in the AI sector's ability to process and generate content from a combination of text, images, and other data types. 

This suite includes not only the models but also curated large-scale datasets and fine-tuning code, which are intended to spur further research and development in the field. 

These models are designed to handle complex tasks like simultaneous image and text analysis, which could have applications in diverse fields such as medical diagnostics and autonomous driving. (source)

Boston Dynamics has recently showcased its latest Atlas humanoid robot in a new video featuring the robot performing push-ups. 

While the commercial utility of a robot doing push-ups may be limited, the video impressively demonstrates Atlas' impressive actuators, which enable a remarkable range of motion. 

This display contrasts with previous footage of Atlas, where the robot bent and twisted in ways that are beyond human capability, highlighting its advanced engineering and versatility. (source)

Perplexity AI, embroiled in plagiarism controversies, plans to launch search ads in Q4.

Despite past accusations from Forbes and Wired about content misuse, Perplexity has updated its citation methods and introduced a revenue-sharing model for publishers influenced by ad-generated revenue.

The company's app, having over two million downloads, handles over 230 million queries monthly, showing substantial U.S. growth.

Perplexity's forthcoming ad model targets various sectors, with ad costs set significantly higher than average industry rates. (source)

In a recent post on X, Andy Jassy introduced us to Amazon Q, Amazon's GenAI assistant for software development which has revolutionized the process of updating foundational software by integrating a new code transformation capability. 

This innovation dramatically reduced the time required for Java upgrades, slashing what normally takes 50 developer days to just a few hours and saving an estimated 4,500 developer-years of work. 

More than half of Amazon's production Java systems were upgraded to modern versions in less than six months, with 79% of auto-generated code reviews passing without modifications. (source)

Tool: Mindsera

In the competitive business world, maintaining mental clarity and strategic focus is paramount, taking a moment to reflect and process our thoughts can be incredibly valuable. MindSera, a journaling tool with AI analyzing feature. It offers AI-powered prompts that guide users through reflective practices to improve problem-solving capabilities and emotional resilience.

Problem Statement: It’s actually not a problem statement, it’s a regular exercise. Take a moment each day to jot down new insights, fleeting thoughts, or memorable experiences. Transform your daily reflections into a vibrant tapestry of learning and personal growth. The AI analyzer will serve as a guide, helping you analyze your thoughts and providing answers based on your prompts.

How to Access:

  • Log in to MindSera.

  • Write your thoughts.

  • Select the text.

  • Choose "Analyze" to enable the AI to give you an answer based on your selected prompt.

  • Receive your analyzed journal.

Note: This is a paid version, so while it's a great app for journaling, accessing the AI facility requires payment.

  • In the recent episode of Leading with Data, I had an impressive conversation with Gaurav Agarwal, CEO of RagaAI. He shared his views on the development of AI technologies, particularly in autonomous driving, and how he believes building strong teams is essential for success.

What is your view on the California AI Regulation bill? I would love to know how you are feeling about it.

It feels like a difficult, but very significant choice in the way AI develops from here.

Question to Ponder

What is your view on the California AI Regulation bill?

Login or Subscribe to participate in polls.

Keep learning!

Kunal

How do you rate this issue of AI Emergence?

Would love to hear your thoughts

Login or Subscribe to participate in polls.

Reply

or to participate.