Google, widely renowned for its search engine dominance, has been forced to confront an unexpected reality: its position in the artificial intelligence or AI arms race is no longer solid. In a leaked internal memo between executives, Google admitted that open source AI has surpassed it and other closed-source systems, making it impossible for them to compete.
According to the leaked memo, open source AI is faster, more customizable and just overall better in all areas than Google’s, and even OpenAI’s, proprietary AI.
These developments have caused Google to shift strategies to maintain its edge in the industry; it is now considering releasing its projects on an open source basis to retain control while still benefiting from the advantages of open source technology.
So, what does this mean for the future of AI, Google and open source AI? Let’s take a look at the implications of this leaked Google memo.
It Was Not OpenAI, After All
The memo’s author, Luke Sernau, started the document with this statement:
“The uncomfortable truth is, we aren’t positioned to win this arms race and neither is OpenAI.”
While Microsoft-backed OpenAI and Google are busy one-upping each other in the AI space, Sernau said that open source has been lapping both of them in several respects, including:
• Mobile phone large language models (LLMs)
• Multimodal capabilities
• Responsible release
• Scalable personal AI, like an open source AI chatbot
This started from the leak of Meta’s LLaMA. In itself, LLaMA didn’t have anything special. Just a very large language model. But, it was the first of its kind to be released openly.
It inspired many others to pick up where Meta left off, leading to the open source revolution we’re seeing today.
Sernau said this addressed a perennial problem of proprietary AIs today: scaling. Meta’s LLaMA reduced the barrier of entry to just “one person, an evening and a beefy laptop.”
And this is not the first time big tech lost the open source vs closed source battle.
Sernau noted that what’s happening right now with large language models is vastly similar to what happened to the rise of image generation. OpenAI released Dall-E as the first AI image-generation tool in January 2021. However, its open source counterpart, Stable Diffusion, received all sorts of improvements, such as:
• User interfaces
• Product integrations
Whether the open source AI projects we’ve seen popping up become Google’s “Stable Diffusion moment” remains to be seen. However, “the broad structural elements are the same,” Sernau said. Meaning, open source AI might be Google’s biggest AI challenge yet.
Faster, More Customizable, More Private
The memo also admitted the primary weakness of Google’s and big tech’s large language models: iterations are slow and costly. In contrast, open source generative AI is agile, highly customizable and cheap.
Meta’s leaked LLM already got improved variants, including:
• Instruction tuning
• Reinforced learning from human feedback (RLHF)
• Human evaluations
And all these improvements happened within one month of LLaMA’s leak. Sernau points out that this was made possible by low-rank adaptation (LoRA) and scaling breakthroughs in image synthesis and large language models.
LoRA is an open source technology that allows for faster and more efficient training of large language models to improve their accuracy. Chinchilla and latent diffusion, on the other hand, are two techniques used to scale up neural networks for AI.
Here’s what the memo said about these technologies:
“In both cases, access to a sufficiently high-quality model kicked off a flurry of ideas and iteration from individuals and institutions around the world.”
In the long run, giant models are just slowing big tech down. Projects like an open source AI chatbot proved that “the best models are the ones which can be iterated upon quickly.” This is because the open source community is doing so much more with $100 and 13 billion parameters than Google can do with $10 million and 540 billion parameters. To make matters more interesting, they are doing all these things in a matter of weeks, not months.
Furthermore, Sernau pointed out that the quality of open source vs closed source is so close people are not willing to pay for a restricted model that adds just the same value as free alternatives.
An excellent example of this would be the rise of ChatGPT alternative projects, hurting their paid version even if it already runs on GPT-4. While these open source chatbot alternatives don’t run on GPT-4, the number of iterations they receive from the open source community is phenomenal and is closing any gap between them.
Open Source vs Closed Source AI Models: What’s the Difference?
So, what’s the exact difference between these two AI models? While they all use neural networks, their differences lie in how they interact with data.
Closed source models are highly optimized and tightly guarded intellectual property that can’t be tweaked or improved by anyone outside of the company. Their closed nature “provides greater control and security for those who develop and use the technology,” said Nanette Taripe, Thrive’s Senior search engine optimization (SEO) Strategist.
On the other hand, open source models offer “greater transparency and accessibility, allowing for greater collaboration and innovation in the development process,” Taripe said.
Unlike closed source AI, open source models benefit from a customer feedback loop, meaning users can learn from their mistakes and improve the model iteratively.
However, the leaked Google memo specifies three key areas where open source AI projects are miles better than what they currently have.
Scale, in this context, means different things for Google and the open source community. For Google, scale meant investing in LLMs and billions of parameters. For the open source community, scaling meant “training on small, highly curated datasets,” that are highly flexible and quick in terms of pace of improvement.
This means that even if they have a size disadvantage, the “cumulative effects of all these fine tunings” more than compensate for it. For example, there are already indistinguishable ChatGPT alternatives and variants that operate as an open source chatbot. The scaling of open source AI projects is much better in producing desired results than how Google is running things on its own end.
As pointed out earlier, open source AI projects primarily operate within LoRA, latent diffusion or Chinchilla. These cheap methods allow engineers to iterate on previous work instead of starting from scratch, like what Google does with their LLMs.
This makes developing robust projects like a ChatGPT alternative easier within just a few weeks without wasting too many resources and energy. And if you need to work with other engineers worldwide, most open source AI projects allow you to do that.
Since ordinary individuals know more about the best use case of AI development in open source, the quality of their work is higher than what Google can offer with their LLMs. This opens up opportunities for new and creative applications that weren’t available before.
“The fact that this technology exists is underexploited inside Google, even though it directly impacts some of our most ambitious projects,” Sernau said in the memo.
Even if the fine-tunings in open source AI projects are “low rank, their sum need not be, allowing full-rank updates to the model to accumulate over time. This means that as new and better datasets and tasks become available, the model can be cheaply kept up to date, without ever having to pay the cost of a full run,” Sernau stated.
On the other hand, Google’s generative AI models rely on constant retraining, pretraining and throwing away iterative improvements, which cost resources and time.
What’s the Future of AI? Google Offers a Solution
The future of AI, as Google sees it, is open source, but not OpenAI. Hence, Sernau suggested Google cement itself as a “thought leader and direction setter” of the open source innovation, “shaping narratives on ideas that are larger in itself.”
This way, they can essentially “own” the platform where most innovations happen. This meant treading an uncomfortable direction of relinquishing control of some parts of their proprietary AI, such as ULM variants and model weights.
However, this strategy is not unfamiliar to Google as they have handled such matters before with their Android OS and Chromium.
Google finds itself in a unique position of leading another open source community, but this requires losing their tight control of their models. In effect, Google expects other open source alternatives to lose their attractiveness and allow Google to hold the reins of the community.
What You Can Do: Three Tips From SEO Experts
Unlike image generation, the rise of generative AI has profound implications for the future of SEO and marketing in general. There’s a lot of information going out of big tech today, like the ones from Anthropic and Google, but how do you make sense of it all?
Stay in the Loop About Open Source Innovation
You can start by educating yourself on the latest open source AI trends, such as GPT-4 and OpenAI’s recent advancements.
This leaked Google memo “emphasizes the quick pace of innovation and the necessity of staying updated about new trends and technology, even though it may not have an immediate impact on SEO and digital marketing experts,” Taripe said.
A great way to stay in the loop is to follow reliable sources that update you on the latest news and breakthroughs. Some great digital and search marketing resources include:
• Search Engine Journal
• Thrive Internet Marketing Agency
• SEO Roundtable
• Search Engine Land
These websites offer great insights into the latest trends in SEO and open source AI. Keep yourself up to date with what’s happening in the industry, so you can make informed decisions when it comes to your franchise and business marketing strategies.
Expect Google’s Pivot Towards Open Source
Google is obviously gearing to enter the open source community as a move against their primary big tech AI competitors, OpenAI and Microsoft.
With the help of Anthropic, Google is positioning themselves as leaders in responsible AI and is expected to be a significant player in the open source AI race soon.
As a result, Taripe recommends paying closer attention to open source programs and technologies since they “offer affordable and creative alternatives that enhance work procedures and results” on top of a “strong community of developers and contributors that are constantly upgrading and improving these programs.”
Be prepared for Google’s strong presence and industry-leading strategies as they launch their products in the open source AI space. These strategies and tools might come in the form of pay-per-click (PPC) solutions or a comprehensive platform for machine learning development.
Wisely Choose Your AI Solutions
We are spoiled with endless options for AI today. However, this doesn’t mean that we should just choose whichever we like.
Don Gruspe, Thrive’s Demand Generation Specialist, listed a few things to keep in mind when investing in AI programs and software.
If you require frequent code modifications and access to a community of developers, then open source AI is the way to go. However, “Open source AI may be riskier because they depend on volunteer development,” Gruspe said.
That’s why “AI systems like Bard and ChatGPT, which provide better security, dependability and customer support, may be preferred by companies with sensitive data or those with less technical know-how,” Gruspe explained.
While each business has different needs, Gruspe still strongly recommends using proprietary AI because of safeguards and security protocols employed by reputable companies like Google and Microsoft.
Stick to Current SEO and AI Best Practices and Guidelines
Much of the future of SEO and AI is up to speculation. So, the best way forward is to stick with the tried and tested SEO best practices and AI guidelines.
Google’s recent internal memo might have provided insight into their plans. Still, it does not necessarily mean these will be their only long-term strategies. Be mindful of the changes in Google’s algorithms and the developments in open source AI innovations.
That’s why we recommend the same things, such as optimizing your website’s technical SEO so search engines would easily understand what your website is all about. If you’re a local business, then you should definitely optimize for your geo-location to increase your visibility to local customers.
These optimizations would never fail despite the rise of AI. After all, they are the basics of SEO, and would always be relevant no matter how advanced the technology is.
Expect More Open Source AI Products in the Market
Gruspe noted that “open source AI solutions are expected to continue to expand in both development and application” as their benefits are acknowledged by companies and professionals in several industries. “To keep up with open source AI development, major corporations like Google may adopt this trend and distribute codes as open source.”
As we said, keeping track of these advancements helps stay ahead of the curve and efficiently allocate resources as new technologies arise.
Get Ahead With Digital Marketing With Thrive
The future of SEO and AI is open source, and Google’s move into the space signals a shift in how we interact with technology. Just as an engine needs fuel to run, so does AI need data – but it also requires careful consideration from digital marketers willing to stay up-to-date on industry trends and best practices.
With Thrive’s help, you can navigate the ever-changing waters of SEO and digital marketing with the knowledge to keep your strategies ahead of the curve.
So don’t be left behind: join our Growth Insider list today and start future-proofing your digital marketing strategies.