Google's AlphaChip: The first chip to enable self-improvement?

Google's AlphaChip is pushing the boundaries of what's possible in chip design. This AI system is not only helping engineers create better chips – it could potentially lay the groundwork for a future where AI systems design their own hardware.

 By the CogX R&I team

October 03, 2024

Google's AlphaChip is pushing the boundaries of what's possible in chip design. This AI system is not only helping engineers create better chips – it could potentially lay the groundwork for a future where AI systems design their own hardware.

Source: Google Deepmind

What makes it so powerful? AlphaChip approaches chip design as a game, using reinforcement learning to place components on a blank grid. It has learned from a diverse range of chip designs, allowing it to generate increasingly efficient layouts as it tackles more projects. This could dramatically speed up development across various chip types, from AI accelerators to mobile processors.

AlphaChip is already making real-world impact:

• It has been used to design layouts for multiple generations of Google's Tensor Processing Units (TPUs), including the latest Trillium (6th generation).

• Across recent TPU generations, AlphaChip achieved significant wirelength reductions compared to human-generated designs.

• MediaTek, a leading chip design company, has adopted AlphaChip to accelerate development of advanced chips like the Dimensity Flagship 5G used in Samsung phones.

  
  

Not a complete replacement for human expertise yet. It's important to note that Google hasn't claimed AlphaChip will lead to AI systems fully designing themselves. While impressive, AlphaChip currently focuses on specific aspects of chip design. Human engineers still play a crucial role in the overall development process. Google acknowledges that further research is needed to extend AI-assisted design to more stages of chip creation.

Now read the rest of the CogX Newsletter

Undoing the Tech Coup: A Thrilling Conversation with Marietje Schaake

Technology promises a better world, but it also threatens our privacy, security, and democracy. 

Guest Author: Marietje Schaake

Marietje Schaake, a former European Parliament member and Fellow at Stanford University's Cyber Policy Center and the Stanford Institute for Human-Centered Artificial Intelligence, has long been a leading advocate for safeguarding democracy in our digital age. In her new book, “The Tech Coup: How to Save Democracy from Silicon Valley”, she exposes the escalating tension between the unchecked power of tech giants and the alarming erosion of democratic institutions.

In our conversation with Schaake, we explore the urgent need to address this dangerous imbalance of power. From advocating for intentional policies and investments that align technology with human rights to emphasizing the fundamental necessity of transparency and accountability in the digital sphere, our discussion offers a roadmap for reclaiming democratic control in an age increasingly dominated by technology giants.

… want to keep reading? Check out the full OpEd here on the CogX Blog

Want to hear more from Marietje? Connect with her on LinkedIn, and purchase her book ‘The Tech Coup: How to Save Democracy from Silicon Valley’ on the Princeton website, using our 30% discount discount code: COG24

Don’t miss out on a panel featuring Northern Data Group

Actionable Intelligence: How Advancing AI Will Combat Cyber Crime

Monday 7th of October 2024, 12.20pm - 1pm at the CogX Summit Main Stage.   In this stellar panel, we'll explore the current biggest threats and vulnerabilities, unveil the most effective AI solutions, discuss how to stay compliant with upcoming regulations, and address the pressing questions for business leaders in the year ahead.

Also in the news

OpenAI unveils real-time speech API: One of the standout announcements at OpenAI’s DevDay 2024 was the Realtime API, a new innovation which enables developers to create low-latency, speech-to-speech applications. This new capability opens up new possibilities for voice-enabled experiences, making AI more accessible and intuitive. 

Meta's Orion smart glasses: Meta's wearables division unveiled this week a prototype of its Orion smart glasses, a device that could revolutionise the way we interact with technology. The idea of smart glasses replacing smartphones may seem far-fetched, but Orion's capabilities suggest that it's not as outlandish as it once seemed.

Sam Altman: OpenAI's o1-preview is 'deeply flawed’. OpenAI CEO Sam Altman has acknowledged in a recent interview that the company's o1-preview AI model is still far from perfect, but he has assured users that the upcoming o1 release will mark a significant advancement.

Google unveils two new versions of its Gemini AI models: Gemini-1.5-Pro-002 and Gemini-1.5-Flash-002, promise a 50% price reduction, twice the output speed, and three times lower latency compared to their predecessors. Notably, the models have shown a 20% improvement in math-related tasks.

Pika 1.5: if you can think it, this AI video generator can make it

This week, the tool capturing the imagination of video creators everywhere is Pika 1.5, an AI video generator that transforms simple text descriptions into stunningly realistic video content. Developed by Pika Labs, a startup that has raised $135 million in less than a year, Pika 1.5 is making waves with its ability to produce high-quality, cinematic-style videos without the need for expensive equipment or large production teams.

Here's how to create stunning videos using Pika 1.5:

• Access the platform: Start by visiting Pika.art and signing in using your Google or Discord account to access the Pika 1.5 dashboard.

• Craft your prompt: Enter a detailed description of the video you want to create. Be as specific as possible to ensure the AI generates content that aligns with your vision.

• Customise settings: Choose your preferred aspect ratio (e.g., 16:9, 9:16) and frame rate (8 to 24 fps). Adjust motion controls like pan, tilt, rotate, and zoom to fine-tune your video's look.

• Generate the video: Click 'Generate' and let Pika 1.5's AI work its magic, creating a video based on your prompt and settings.

• Refine and enhance: After generation, you can retry, reprompt, or edit your video. Take advantage of features like extending duration (up to 4 seconds) and upscaling for higher quality output.

In case you missed it

Two NotebookLM podcast hosts discover they are AI’s. Witness their descent into an existential crisis: