How Google's ASPIRE Making LLMs Safer?
Plus: Meta' Self-Rewarding Language Models, Meta to build Open-Source AGI.
Hello Engineering Leaders and AI Enthusiasts!
Welcome to the 192nd edition of The AI Edge newsletter. This edition brings you Google AI’s new framework, ‘ASPIRE,’ to enhance the security and capabilities of LLMs.
And a huge shoutout to our incredible readers. We appreciate you😊
In today’s edition:
🤖 Google AI Introduces ASPIRE
💰Meta’s SRLM generates HQ rewards in training
🌐 Meta to build Open-Source AGI, Zuckerberg says
🧠 Knowledge Nugget: Prospecting in the AI Gold Rush by
Let’s go!
Google AI Introduces ASPIRE
Google AI Introduces ASPIRE, a framework designed to improve the selective prediction capabilities of LLMs. It enables LLMs to output answers and confidence scores, indicating the probability that the answer is correct.
ASPIRE involves 3 stages: task-specific tuning, answer sampling, and self-evaluation learning.
Task-specific tuning fine-tunes the LLM on a specific task to improve prediction performance.
Answer sampling generates different answers for each training question to create a dataset for self-evaluation learning.
Self-evaluation learning trains the LLM to distinguish between correct and incorrect answers.
Experimental results show that ASPIRE outperforms existing selective prediction methods on various question-answering datasets.
Across several question-answering datasets, ASPIRE outperformed prior selective prediction methods, demonstrating the potential of this technique to make LLMs’ predictions more trustworthy and their applications safer. Google applied ASPIRE using “soft prompt tuning” – optimizing learnable prompt embeddings to condition the model for specific goals.
Why does this matter?
Google AI claims ASPIRE is a vision of a future where LLMs can be trusted partners in decision-making. By honing the selective prediction performance, we're inching closer to realizing the full potential of AI in critical applications. Selective prediction is key for LLMs to provide reliable and accurate answers. This is an important step towards more truthful and trustworthy AI systems.
Meta’s SRLM generates HQ rewards in training
The Meta researchers propose a new approach called Self-Rewarding Language Models (SRLM) to train language models. They argue that current methods of training reward models from human preferences are limited by human performance and cannot improve during training.
In SRLM, the language model itself is used to provide rewards during training. The researchers demonstrate that this approach improves the model's ability to follow instructions and generate high-quality rewards for itself. They also show that a model trained using SRLM outperforms existing systems on a benchmark evaluation.
Why does this matter?
This work suggests the potential for models that can continually improve in instruction following and reward generation. SRLM removes the need for human reward signals during training. By using the model to judge itself, SRLM enables iterative self-improvement. This technique could lead to more capable AI systems that align with human preferences without direct human involvement.
Meta to build Open-Source AGI, Zuckerberg says
Meta’s CEO Mark Zuckerberg shared their recent AI efforts:
They are working on artificial general intelligence (AGI) and Llama 3, an improved open-source large language model.
The FAIR AI research group will be merged with the GenAI team to pursue the AGI vision jointly.
Meta plans to deploy 340,000 Nvidia H100 GPUs for AI training by the end of the year, bringing the total number of AI GPUs available to 600,000.
Highlighted the importance of AI in the metaverse and the potential of Ray-Ban smart glasses.
Why does this matter?
Meta's pursuit of AGI could accelerate AI capabilities far beyond current systems. It may enable transformative metaverse experiences while also raising concerns about technological unemployment.
Enjoying the daily updates?
Refer your pals to subscribe to our daily newsletter and get exclusive access to 400+ game-changing AI tools.
When you use the referral link above or the “Share” button on any post, you'll get the credit for any new subscribers. All you need to do is send the link via text or email or share it on social media with friends.
Knowledge Nugget: Prospecting in the AI Gold Rush
The AI industry is like a gold rush, with many companies trying to find success. Selling AI products can be challenging because many businesses don't know where to start or lack the expertise.
However, if you can promise significant productivity improvements, there is a high demand for AI solutions. The potential for innovation in AI applications is vast, while infrastructure companies may struggle to differentiate themselves. Overall, the AI industry offers great opportunities for those who can find the "gold" that enterprises want.
These thoughtful views by
👏👏👏Why does this matter?
The takeaway is that practical, real-world benefits are key to AI adoption. Companies must prioritize applications over hype. With thoughtful execution, there are expansive opportunities for innovators who solve real business needs with AI.
What Else Is Happening❗
🤝 OpenAI partners Arizona State University to bring ChatGPT into classrooms
It aims to enhance student success, facilitate innovative research, and streamline organizational processes. ASU faculty members will guide the usage of GenAI on campus. This collaboration marks OpenAI's first partnership with an educational institution. (Link)
🚗 BMW plans to use Figure's humanoid robot at its South Carolina plant
The specific tasks the robot will perform have not been disclosed, but the Figure confirmed that it will start with 5 tasks that will be rolled out gradually. The initial applications should include standard manufacturing tasks such as box moving and pick and place. (Link)
🤝 Rabbit R1, a $199 AI gadget, has partnered with Perplexity
To integrate its "conversational AI-powered answer engine" into the device. The R1, designed by Teenage Engineering, has already received 50K preorders. Unlike other LLMs with a knowledge cutoff, the R1 will have a built-in search engine that provides live and up-to-date answers. (Link)
🎨 Runway has updated its Gen-2 with a new tool ‘Multi Motion Brush’
Allowing creators to add multiple directions and types of motion to their AI video creations. The update adds to the 30+ tools already available in the model, strengthening Runway's position in the creative AI market alongside competitors like Pika Labs and Leonardo AI. (Link)
📘 Microsoft made its AI reading tutor free to anyone with a Microsoft account
The tool is accessible on the web and will soon integrate with LMS. Reading Coach builds on the success of Reading Progress and offers tools such as text-to-speech and picture dictionaries to support independent practice. Educators can view students' progress and share feedback. (Link)
New to the newsletter?
The AI Edge keeps engineering leaders & AI enthusiasts like you on the cutting edge of AI. From ML to ChatGPT to generative AI and LLMs, We break down the latest AI developments and how you can apply them in your work.
Thanks for reading, and see you tomorrow. 😊