• AIdeations
  • Posts
  • AI-Driven Chip Design, Bacterial Computers, and OpenAI’s Power Play

AI-Driven Chip Design, Bacterial Computers, and OpenAI’s Power Play

Thank you for 350 issues of Aideations! Today we dive into AI-designed chips, bacteria solving math, and OpenAI’s massive data center plans. Big updates ahead—don’t miss it!

Author

Brent Moreno
September 27, 2024

Aideations Newsletter #350 – A Thank You and a Look Ahead!

Today marks issue #350 of Aideations. First, a huge thank you to everyone who has shared this newsletter with friends, family, and coworkers. To those of you reading it daily, providing feedback – I see you, and I appreciate you. It’s been a journey, and it wouldn’t be what it is without your support.

Now, for some updates. For a while, I’ve wanted to add something extra for those of you looking for more. My original plan was to launch a full course. I made solid progress early on, but as client projects picked up and life threw its usual curveballs, I hit a wall. Burnout happens, and I needed to step back to avoid spinning my wheels.

But I’m not dropping the ball on this. Moving forward, I’m shifting to a more focused approach: I’ll be putting out one in-depth newsletter every Friday or Saturday. It’ll break down tools, techniques, and tips I’ve learned, plus how to build automations, use specific tools, build apps, and more. This lets me keep the content sharp and differentiated from what everyone else is doing. Aideations has always been about delivering real value, and this next step keeps that front and center.

Thanks for sticking with me, and I’m excited to keep delivering content that matters.

AI is Now Designing Chips Faster Than Ever – Here's How AlphaChip is Leading the Charge

Quick Byte:

Google’s AlphaChip AI is revolutionizing the way computer chips are designed. This AI-powered tool creates superhuman-level chip layouts in hours, not months, and it’s already being used in everything from smartphones to the data centers running AI models like Gemini. It’s faster, cheaper, and more efficient than anything humans could produce—and it’s just getting started.

Key Takeaways:

  • AlphaChip speeds up chip design: What used to take months of human effort can now be completed by AlphaChip in a matter of hours.

  • AI is designing Google’s TPUs: AlphaChip has been used to design chips that power AI models like Gemini and Imagen—Google’s heavy hitters in the AI space.

  • Companies like MediaTek are onboard: Leading chip designers are already using AlphaChip to design chips for mobile phones, improving performance and efficiency.

  • 15% improvement in wirelength reduction: AlphaChip's designs are more efficient, reducing wire length and boosting performance across three generations of Google’s TPUs.

Bigger Picture:

AlphaChip represents the future of chip design. With the ability to optimize the process from start to finish, it’s set to transform not only the AI industry but also everyday devices like smartphones and medical equipment. The chip design process is no longer held back by human limitations, and this AI-powered tool could dramatically shorten the time it takes to bring new innovations to market.

We’re talking about faster, cheaper, and more power-efficient chips across the board—and AlphaChip is paving the way for this future. Expect to see even more AI-driven advancements as chip design becomes fully automated and hyper-optimized.

Bacteria Are Now Solving Math Problems?! Welcome to the Weird Future of AI

Quick Byte:

Yeah, you read that right—scientists have turned E. coli bacteria into mini problem-solving machines. These “bactoneurons” are like little biological computers that can spot prime numbers, identify vowels, and even figure out how to slice a pizza using only straight lines. The best part? These bacterial brain cells are cheaper, self-replicating, and smaller than traditional silicon chips. Wild, right?

Key Takeaways:

  • Bacteria as computers: Researchers hacked E. coli bacteria to act like neurons, processing information via chemicals instead of binary code.

  • Doing math and more: These bactoneurons can ID prime numbers, vowels between A-L, and solve geometric puzzles like how many ways you can cut a pizza.

  • Tiny and scalable: At 2 to 5 micrometers, they handle inputs and outputs with their own chemical power supply, making them cheaper and smaller than any silicon chip.

  • LEGO-style computing: This system is modular, meaning you can mix and match these little guys to tackle different problems—all using the same basic building blocks.

Bigger Picture:

This is the kind of breakthrough that feels straight out of a sci-fi movie, but it's real—and it could change everything about computing. We're talking self-replicating bacteria solving problems that we’d typically throw at classical computers. Imagine AI systems like ChatGPT, but instead of massive data centers, they're powered by bacteria on a molecular level.

The future of computing might not be silicon and circuits—it could be biological, scalable, and insanely efficient. This is just the start, and if this tech evolves, we might be looking at a future where bacteria could outperform traditional chips, not just in cost but also in speed and adaptability.

OpenAI’s Insane Plan for 5GW Data Centers: Can the U.S. Keep Up with Sam Altman’s AI Dreams?

Quick Byte:

Sam Altman just dropped a massive power play—literally. OpenAI wants to build data centers that need 5 gigawatts of power each (that’s like Miami-level energy). And guess what? They need multiple of these bad boys. Altman’s out here asking the U.S. government to help fast-track the construction of these AI megahubs, but the big question is: who the hell has that much power to spare?

Key Takeaways:

  • OpenAI’s superpower hunger: 5 gigawatts is no joke. It’s 100x the power of your average data center. And OpenAI wants several of these monsters.

  • Beyond wild: To put this into perspective, building just one of these centers could drain twice the electricity used by all of New York State. Yup, Altman’s asking for city-level power for each one.

  • Power experts calling BS: Nobody really knows how they’ll meet these energy demands. Even nuclear plant CEOs are like, “That’s… a lot.”

  • Tech vs reality: While the idea sounds cool (AI that needs city-sized power grids!), the actual infrastructure and timeline? Major hurdles.

Bigger Picture:

Altman’s vision is crystal clear: OpenAI is gunning to stay at the front of the AI race, and that means building mind-bending computing power—fast. The goal? Scale up AI to a point where it’s solving real-world problems on a massive scale, like curing diseases or building next-gen AI systems. But here’s the catch: even with the government’s help, we’re talking hundreds of billions of dollars and infrastructure challenges that we’ve never faced before.

So yeah, OpenAI might be the king of AI today, but if they want to keep that throne, they’ll need to figure out how to power this empire. The race isn’t just about AI—it’s about energy, resources, and who can move faster to build the future.

NotebookLM adds audio and YouTube support, plus easier sharing of Audio Overviews

Today, we’re expanding the types of sources you can use with NotebookLM and giving you an easier way to share an Audio Overview.

I just tried ChatGPT’s 5 new voices — and it’s way better than any other voice AI

This is wild

An author is facing backlash for using AI to write 120 books. He says he's misunderstood.

Tim Boucher has written more than 120 books. His secret weapon: AI.

AI tweaks to photos and videos can alter our memories

It has become trivially easy to use artificial intelligence to edit images or generate video to remove unwanted objects or beautify scenes, but doing so leads to people misremembering what they have seen.

Build a Powerful AI App in Minutes: Web Scraper Tutorial with Replit Agent & Make.com

Authors:
Chenming Zhu, Tai Wang, Wenwei Zhang, Jiangmiao Pang, Xihui Liu
Affiliated with: The University of Hong Kong and Shanghai AI Laboratory.

Summary:
LLaVA-3D takes language models to the next level by adding 3D spatial awareness to them. While large multimodal models (LMMs) have become adept at understanding 2D images and visual reasoning, they’ve been lacking in 3D scene comprehension—until now. This research introduces a method to upgrade 2D models to 3D by utilizing a new representation called "3D Patch." Essentially, this lets a model understand depth and spatial relationships in 3D scenes while retaining its powerful 2D capabilities. The magic is that it doesn’t require massive datasets or special 3D encoders, meaning the technology can be applied efficiently without too much computational power.

Why This Research Matters:
The ability to understand 3D scenes has been a barrier for LLMs because existing models weren’t trained on large-scale 3D data. LLaVA-3D changes that by repurposing 2D vision data and enhancing it with 3D spatial features. This opens the door to more interactive AI systems that can engage with the physical world, not just 2D visual environments. Think about autonomous systems, augmented reality, or robots—they all require a deep understanding of 3D space to function properly. LLaVA-3D empowers AI to move from flat visual understanding to spatial reasoning.

Use Cases:

  1. Robotics: Robots need to navigate and manipulate objects in the 3D world. With LLaVA-3D, these robots can make better decisions and interact more intelligently with their environment.

  2. Autonomous vehicles: These vehicles rely on 3D spatial understanding to avoid obstacles and make real-time decisions. LLaVA-3D can significantly enhance these capabilities.

  3. Augmented Reality (AR) and Virtual Reality (VR): By integrating 3D-aware AI, AR/VR systems can offer more immersive and contextually aware experiences, leading to more realistic interactions.

  4. Smart assistants: Imagine an AI that can analyze 3D environments—your living room, a city street, or even a warehouse—and offer advice or manage tasks accordingly.

Immediate Impact:
LLaVA-3D brings immediate upgrades to existing 2D multimodal models without sacrificing their 2D capabilities. By adding 3D-awareness to AI systems, industries like robotics, autonomous driving, and augmented reality can benefit from enhanced AI functionality without the need for large, expensive 3D datasets. The reduced training time—3.5x faster than other 3D models—makes this framework accessible and cost-effective.

Future Impact:
In the future, LLaVA-3D could become the foundation for AI systems that need to operate in complex, real-world environments. As more industries adopt 3D-aware AI, we can expect better autonomous systems, smarter robots, and more immersive AR/VR experiences. The potential for LLaVA-3D to redefine human-machine interaction in 3D spaces is massive, with applications in healthcare, logistics, entertainment, and beyond.

By enabling AI to understand both 2D and 3D contexts, LLaVA-3D positions itself as a cornerstone for the next wave of AI-driven technology.

Nuvio - Handles your income, expenses, and cash flow, letting you focus on what matters—growing your business. It’s your personal CFO, delivering insights and predictions to keep you ahead.

Clarity - A high-resolution upscaler that can enhance images and add details. You can decide how much detail you want the AI to add.

Neolocus - AI Rendering services for Interior Designers and Architects.

SFX Engine - AI sound effect generator, designed specifically for audio producers, video editors, and game developers.

Outspeed - Provides networking and inference infrastructure to build fast, real time voice and video AI apps.

Figma To Website - Design and publish websites without ever leaving Figma.

Create a 5 Day Email Course:

I want you to help me write a a 5-day email course for [TARGET AUDIENCE] to help them [ACHIEVE DESIRED OUTCOME].

The course topic is: [COURSE TOPIC]

For each of the 5 emails, please provide:

An attention-grabbing, curiosity-inducing subject line that makes people want to open faster than a kid on Christmas morning
A brief outline of the core content and lesson for that day, as well as a brief recap of what was covered the day before (obviously not applicable for Day 1)
3-5 key actionable strategies or techniques that subscribers can implement immediately
A cliffhanger or teaser to build anticipation for the next day's
email

Email 1 should set the stage for and outline each day of the course, generate excitement, and deliver a quick win or valuable insight that makes subscribers feel
like geniuses.

Emails 2-4 should each focus on one crucial strategy, technique, or mindset shift related to the topic, complete with clear examples and step-by-step guidance even a monkey could follow.

Email 5 should tie everything together, summarize the key takeaways, and lay out a clear roadmap for implementation. It should also include a soft pitch for my related paid offer: [PAID OFFER]

Each email should be written in a warm, conversational tone. 

Each email should be around 400-500 words.

Use storytelling, clear analogies, and relatable examples to make the content engaging and memorable.

The final CTA in each email should inspire readers to put the day's lesson into action and build anticipation for the next email in the series.

Never use exclamation points or emojis.