Google Updates Gemma AI Models with Gemma 3

Improved Performance and New Features

A little over a year after releasing two "open" Gemma AI models built from the same technology behind its Gemini AI, Google is updating the family with Gemma 3. According to the blog post, these models are intended for use by developers creating AI applications capable of running wherever they’re needed, on anything from a phone to a workstation with support for over 35 languages, as well as the ability to analyze text, images, and short videos.

Enhanced Capabilities

The company claims that it’s the "world’s best single-accelerator model," outperforming competition from Facebook’s Llama, DeepSeek, and OpenAI for performance on a host with a single GPU, as well as optimized capabilities for running on Nvidia’s GPUs and dedicated AI hardware. Gemma 3’s vision encoder is also upgraded, with support for high-res and non-square images, while the new ShieldGemma 2 image safety classifier is available for use to filter both image input and output for content classified as sexually explicit, dangerous, or violent.

Addressing Misuse Concerns

Last year it was unclear how much interest there would be in a model like Gemma, however, the popularity of DeepSeek and others shows there is interest in AI tech with lower hardware requirements. Despite its claims of advanced capabilities, Google also says, "Gemma 3’s enhanced STEM performance prompted specific evaluations focused on its potential for misuse in creating harmful substances; their results indicate a low risk level."

Open-Source License and Academic Program

What exactly constitutes an "open" or "open source" AI model remains a topic of debate, and with Google’s Gemma, that has focused on the company’s license that restricts what people are allowed to use it for, which has not changed with this new release. Google is continuing to promote Gemma with Google Cloud credits, and the Gemma 3 Academic program will allow academic researchers to apply for $10,000 worth of credits to accelerate their research.

Frequently Asked Questions

Q: What is Gemma 3?
A: Gemma 3 is an updated version of Google’s open-source AI model, designed for developers to use in their applications.

Q: What are the new features in Gemma 3?
A: Gemma 3 includes an upgraded vision encoder that supports high-res and non-square images, as well as a new ShieldGemma 2 image safety classifier for filtering image input and output.

Q: Is Gemma 3 open-source?
A: Yes, Gemma 3 is open-source, but with restrictions on how it can be used, as specified in Google’s license.

Q: Can I use Gemma 3 for free?
A: Yes, Google is offering Google Cloud credits and a Gemma 3 Academic program for academic researchers to use Gemma 3 for free, with $10,000 worth of credits available.

Meta Tests In-House Chip for Training AI Systems

Meta’s Effort to Reduce Reliance on Hardware Makers

Meta is reportedly testing an in-house chip designed for training AI systems, a move aimed at reducing its reliance on hardware makers like Nvidia. The company is piloting a "small deployment" of the chip, manufactured in partnership with Taiwan-based firm TSMC, and plans to scale up production if the test is successful.

A New Approach to AI Chip Design

This is not Meta’s first foray into custom AI chip design. The company has deployed custom chips before, but only to run AI models, not to train them. However, this new effort marks a significant shift in their approach, as the chip is designed to handle AI-specific workloads.

Partnership with TSMC

The chip was manufactured by TSMC, one of the world’s leading contract chipmakers. The partnership is a sign that Meta is serious about its ambitions in the field of AI chip design. TSMC has a strong reputation for producing high-quality chips, and its partnership with Meta could be a significant boost for the social media giant.

Cost Savings and Long-Term Goals

Meta expects to spend $65 billion on capital expenditure this year, with a significant portion going toward Nvidia GPUs. If the company can reduce even a fraction of that cost by shifting to in-house chips, it would be a major win for the company. This move could also have long-term benefits, as Meta would gain more control over its AI hardware infrastructure and reduce its reliance on third-party providers.

Conclusion

Meta’s testing of an in-house AI chip is a bold move that could have significant implications for the company’s future. By reducing its reliance on hardware makers and increasing its control over AI chip design, Meta may be able to gain a competitive edge in the market. As the company continues to pilot its new chip, it will be interesting to see how this development unfolds and what impact it has on the company’s future growth and success.

FAQs

• What is Meta’s goal with its in-house AI chip?
  Meta aims to reduce its reliance on hardware makers like Nvidia and gain more control over its AI hardware infrastructure.
• Who is partnering with Meta on the chip design?
  Meta is partnering with Taiwan-based firm TSMC to manufacture the chip.
• What is the current scope of the project?
  The project is currently in the pilot phase, with a "small deployment" of the chip being tested.
• What are the potential long-term benefits of this move?
  Reduced costs, increased control over AI hardware infrastructure, and potentially a competitive edge in the market.

As the price of solar panels has plummeted in recent decades, installation costs have taken up a greater share of the technology’s overall price tag. The long installation process for solar farms is also emerging as a key bottleneck in the deployment of solar energy.

Now the startup Charge Robotics is developing solar installation factories to speed up the process of building large-scale solar farms. The company’s factories are shipped to the site of utility solar projects, where equipment including tracks, mounting brackets, and panels are fed into the system and automatically assembled. A robotic vehicle autonomously puts the finished product — which amounts to a completed section of solar farm — in its final place.

“We think of this as the Henry Ford moment for solar,” says CEO Banks Hunter ’15, who founded Charge Robotics with fellow MIT alumnus Max Justicz ’17. “We’re going from a very bespoke, hands on, manual installation process to something much more streamlined and set up for mass manufacturing. There are all kinds of benefits that come along with that, including consistency, quality, speed, cost, and safety.”

Last year, solar energy accounted for 81 percent of new electric capacity in the U.S., and Hunter and Justicz see their factories as necessary for continued acceleration in the industry.

The founders say they were met with skepticism when they first unveiled their plans. But in the beginning of last year, they deployed a prototype system that successfully built a solar farm with SOLV Energy, one of the largest solar installers in the U.S. Now, Charge has raised $22 million for its first commercial deployments later this year.

From surgical robots to solar robots

While majoring in mechanical engineering at MIT, Hunter found plenty of excuses to build things. One such excuse was Course 2.009 (Produce Engineering Processes), where he and his classmates built a smart watch for communication in remote areas.

After graduation, Hunter worked for the MIT alumni-founded startups Shaper Tools and Vicarious Surgical. Vicarious Surgical is a medical robotics company that has raised more than $450 million to date. Banks was the second employee and worked there for five years.

“A lot of really hands on, project-based classes at MIT translated directly into my first roles coming out of school and set me up to be very independent and run large engineering projects,” Banks says, “Course 2.009, in particular, was a big launch point for me. The founders of Vicarious Surgical got in touch with me through the 2.009 network.”

As early as 2017, Hunter and Justicz, who majored in mechanical engineering and computer science, had discussed starting a company together. But they had to decide where to apply their broad engineering and product skillsets.

“Both of us care a lot about climate change. We see climate change as the biggest problem impacting the greatest number of people on the planet,” Hunter says. “Our mentality was if we can build anything, we might as well build something that really matters.”

In the process of cold calling hundreds of people in the energy industry, the founders decided solar was the future of energy production because its price was decreasing so quickly.

“It’s becoming cheaper faster than any other form of energy production in human history,” Hunter says.

When the founders began visiting construction sites for the large, utility-scale solar farms that make up the bulk of energy generation, it wasn’t hard to find the bottlenecks. The first site they traveled to was in the Mojave Desert in California. Hunter describes it as a massive dust bowl where thousands of workers spent months repeating tasks like moving material and assembling the same parts, over and over again.

“The site had something like 2 million panels on it, and every single one was assembled and fastened the same way by hand,” Hunter says. “Max and I thought it was insane. There’s no way that can scale to transform the energy grid in a short window of time.”

Hunter says he heard from each of the largest solar companies in the U.S. that their biggest limitation for scaling was labor shortages. The problem was slowing growth and killing projects.

Hunter and Justicz founded Charge Robotics in 2021 to break through that bottleneck. Their first step was to order utility solar parts and assemble them by hand in their backyards.

“From there, we came up with this portable assembly line that we could ship out to construction sites and then feed in the entire solar system, including the steel tracks, mounting brackets, fasteners, and the solar panels,” Hunter explains. “The assembly line robotically assembles all those pieces to produce completed solar bays, which are chunks of a solar farm.”

Each bay represents a 40-foot piece of the solar farm and weighs about 800 pounds. A robotic vehicle brings it to its final location in the field. Banks says Charge’s system automates all mechanical installation except for the process of pile driving the first metal stakes into the ground.

Charge’s assembly lines also have machine-vision systems that scan each part to ensure quality, and the systems work with the most common solar parts and panel sizes.

From pilot to product

When the founders started pitching their plans to investors and construction companies, people didn’t believe it was possible.

“The initial feedback was basically, ‘This will never work,’” Hunter says. “But as soon as we took our first system out into the field and people saw it operating, they got much more excited and started believing it was real.”

Since that first deployment, Charge’s team has been making its system faster and easier to operate. The company plans to set up its factories at project sites and run them in partnership with solar construction companies. The factories could even run alongside human workers.

“With our system, people are operating robotic equipment remotely rather than putting in the screws themselves,” Hunter explains. “We can essentially deliver the assembled solar to customers. Their only responsibility is to deliver the materials and parts on big pallets that we feed into our system.”

Hunter says multiple factories could be deployed at the same site and could also operate 24/7 to dramatically speed up projects.

“We are hitting the limits of solar growth because these companies don’t have enough people,” Hunter says. “We can build much bigger sites much faster with the same number of people by just shipping out more of our factories. It’s a fundamentally new way of scaling solar energy.”

Affiliate Links and Codes: Guidelines and Best Practices

Introduction

To ensure a transparent and user-friendly experience, affiliate links, codes, or cookies must be included only when they provide a direct and transparent user benefit related to the extension’s core functionality.

What Constitutes a Violation?

Some common violations include:

Unrelated Affiliate Links

Inserting affiliate links without providing a tangible benefit to users, such as a discount, cashback, or donation.

Continuous Background Injection

An extension that continuously injects affiliate links in the background without related user action.

Guidelines for Affiliate Links and Codes

To maintain a high level of transparency and user satisfaction, the following guidelines must be followed:

• Affiliate links and codes must be clearly labeled and presented to users.
• Users must be provided with a clear understanding of the benefits of using affiliate links and codes.
• Affiliate links and codes must be related to the extension’s core functionality.
• Affiliate links and codes must not be injected in the background without user action.

Conclusion

By following these guidelines, developers can ensure a transparent and user-friendly experience for their users. Remember, affiliate links and codes should only be used to provide a direct and tangible benefit to users, not to deceive or manipulate them.

FAQs

Q: What are the benefits of using affiliate links and codes?
A: Affiliate links and codes can provide users with discounts, cashback, or other benefits related to the extension’s core functionality.

Q: Can I inject affiliate links in the background without user action?
A: No, it is not permitted to inject affiliate links without related user action and without providing a tangible benefit to users.

Q: What if I provide a discount or cashback for using an affiliate link?
A: In this case, you are providing a tangible benefit to users, and the affiliate link is permissible.