OpenAI Insider: AGI by 2027

Today’s top AI Highlights:

1. Stability AI releases opensource model for audio samples and sound design

2. Microsoft releases first large-scale foundation model of the atmosphere

3. Former OpenAI Superalignment team member says AGI will be achieved by 2027

4. Asana unveils AI Teammates for complex workflows and elevated teamwork

5. Upload an audio and extend it with Udio’s new AI feature

& so much more!

Read time: 3 mins

Latest Developments 🌍

Stability AI’s Opensource Text-to-Audio Model 🎶

Stability AI has released an open-source text-to-audio model Stable Audio Open. The model can generate up to 47 seconds of audio samples at 44.1kHz, including drum beats, instrument riffs, and sound effects. You can also fine-tune the model with your own custom audio data. This release provides a valuable tool for sound designers and musicians to experiment with generative AI for sound design.

Stable Audio Open is different from Stability AI’s commercial text-to-audio model, Stable Audio that generates full tracks up to 3 minutes long with a proper structure. The new open-source model prioritizes audio samples, sound effects and production elements. While it can generate short musical clips, it is not optimised for full songs, melodies or vocals.

$1 Trillion AI Investment Market by 2027 💸

In a podcast with Dwarkesh Patel, a former OpenAI employee turned investor Leopold Aschenbrenner spoke at length about the burgeoning AI landscape. They discussed the rapid scaling of AI, the “trillion-dollar cluster” for the development of superintelligence, and dangers if AGI is not controlled. He hadn’t signed the non-disparagement agreement and had let go of his vested equity to speak freely about OpenAI. Here are some key takeaways:

1. AI is becoming an “industrial process”, requiring massive infrastructure investments beyond just writing code.

2. By 2026, training clusters will need a gigawatt of power, equivalent to the Hoover Dam, costing tens of billions of dollars. By 2030, the cluster will reach 100 gigawatts and require 100 million H100 equivalents, costing a trillion dollars and consuming over 20% of US electricity production.

3. This scale of compute will necessitate new power plants and fabrication facilities, driving a $1 trillion AI investment market by 2027.

4. Leopold has started a new series of blogs “Situational Awareness” which is about understanding the scale of AI progress and its implications. Most people underrate the speed and impact of AI development. It’s like the pandemic, where many people initially dismissed the threat, only to be overwhelmed by the reality of its impact.

5. In one of his blogs, Leopold states that AI is rapidly advancing towards the goal of achieving AGI by 2027. The progress from GPT-2 to GPT-4 has demonstrated a remarkable increase in AI capabilities, moving from simple sentence construction to acing high school exams and performing complex coding tasks.

   1. AI models are expected to see a 3–6 orders of magnitude increase in effective compute by 2027, driven by both hardware improvements and algorithmic efficiencies.

   2. Techniques like RLHF and chain-of-thought prompting have significantly improved AI capabilities, making them more practical and useful. AI will evolve from chatbots to agents that can work alongside humans as effective coworkers.

   3. By 2027, AI models might not only perform tasks currently done by researchers and engineers but also automate AI research itself

1. Leopold was part of OpenAI’s Superalignment team. While firing him, OpenAI cited a “leak” of a document he had shared with three external researchers for feedback on safety in AI development, claiming it contained sensitive information about AGI timelines.

2. Further, he mentioned about a security memo he wrote to the board, outlining concerns on OpenAI’s lack of security, and the board hassled OpenAI’s leaders on this. He received a warning for this and says that it was a major reason for his termination.

AI Takes the Challenge of Accurate Weather Prediction 🌪️

The world’s first large-scale foundation model for the atmosphere is here! Microsoft has released a powerful new AI model “Aurora” with 1.3 billion parameters designed for high-resolution forecasting of weather and atmospheric processes. This model, trained on over a million hours of diverse weather and climate data, tackles the limitations of current weather prediction models, which often struggle to accurately forecast extreme events. Aurora promises major improvements in our ability to anticipate and prepare for changing weather phenomena.

Key Highlights:

1. Diverse Datasets: Aurora is trained on data from a variety of sources, including climate simulations, reanalysis products, and operational forecasts for the model to learn a more comprehensive understanding of atmospheric dynamics.

2. Computational Efficiency: Aurora operates at a computational speed-up of approximately 5,000 times compared to the state-of-the-art numerical forecasting system, the Integrated Forecasting System (IFS). This remarkable efficiency translates to faster and more cost-effective weather predictions.

3. Superior Performance: Aurora outperforms the current best specialized deep learning models, including GraphCast, and the gold standard numerical weather prediction model, IFS HRES, across a range of tasks and resolutions.

Your New Co-worker: Asana’s AI Takes on Tasks 👭

The famous AI work management platform Asana has launched their new “AI teammate” to help teams maximize their impact and achieve goals faster. These AI teammates are not just simple chatbots, they can:

• Advise teams where to focus by surfacing insights around potential risks

• Automate workflows at scale by performing tasks, triaging requests, and assigning work

• Adapt to individual workstyles

• Be a “multi-player” where multiple people can collaborate together with AI in real-time

Built upon their proprietary Work Graph data model, these AI agents offer a unique advantage: they operate within the context of organizational goals and workflows, rather than relying on vast, often unreliable data sets. This means AI teammates can offer more nuanced and relevant insights and actions, taking on tasks like risk assessment, workflow optimization, and even proactive information gathering.

Tools of the Trade ⚒️

1. User Uploads by Udio: Released as a new feature in the text-to-song platform Udio, you can now upload an audio clip and extend this clip either forward or backward by 32 seconds using up to 2 minutes of context and a simple text prompt.

1. Univerbal: An AI language tutor that helps you practice speaking in over 20 languages through real-world scenarios like ordering food in a bakery. It gives instant feedback on grammar and vocabulary, and lets you customize topics or create your own scenes for personalized learning.

2. Wix App Builder: Create a branded mobile app for your business using AI without any coding. You can design and customize the app fully, sync it with your Wix site, and publish it on the App Store and Google Play.

3. Awesome LLM Apps: Build awesome LLM apps using RAG for interacting with data sources like GitHub, Gmail, PDFs, and YouTube videos through simple texts. These apps will let you retrieve information, engage in chat, and extract insights directly from content on these platforms.

Hot Takes 🔥

1. My strategy for mitigating AGI risk is that I always tell the chatbot "please" and "thank you." ~Matthew Yglesias

2. Underrated info hazard from AI: when employees learn that they can automate their own work & no one cares. ~Ethan Mollick

Meme of the Day 🤡

me trying to keep up with AI trends

That’s all for today! See you tomorrow with more such AI-filled content.

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!