The OpenAi irony. Is Meta the real “Open” AI.
In April, Meta teased a groundbreaking open-source AI model to rival the best private models from companies like OpenAI. Today, they deliver with Llama 3.1, boasting 405 billion parameters, outperforming GPT-4o and Claude 3.5 Sonnet on several benchmarks.
Key Highlights:
- Performance: Llama 3.1 outshines major competitors and is now available in more countries and languages.
- Scale: Trained with over 16,000 Nvidia H100 GPUs, costing hundreds of millions of dollars.
- Accessibility: Meta offers Llama 3.1 under an open-source license, promoting its use like Linux in the OS market.
- Data Use: Meta used synthetic data for training, sidestepping detailed disclosures to avoid potential copyright issues.
- Advanced Capabilities: Llama 3.1 integrates with search engines and tools, showcasing agentic behaviors like executing Python code from web searches.
- Security: For the first time, Meta included cybersecurity and biochemical use cases in red teaming tests.
- Meta AI Assistant: Llama 3.1 is integrated into Meta’s platforms, including WhatsApp, Instagram, and Facebook, with a feature to generate personalised images.
Meta predicts Llama 3.1 will become the most-used AI assistant by the end of the year, surpassing ChatGPT’s 100 million users. The race in AI development is heating up, with Meta positioning open-source AI at the forefront.
In comparison, the current industry go-to from OpenAi (creators of ChatGPT) started as an open-source, nonprofit organisation with a mission to ensure that artificial general intelligence (AGI) benefits all of humanity. However, as of recent years, OpenAI has transitioned to a more closed and profit-oriented model to secure funding and resources necessary for advanced AI research and development. With GPT-3 and GPT-4 now considered ‘closed’, where the underlying code and model details are not publicly accessible and the full functionality can only be accessed and utilised via paid services.
What we are seeing is a battle of open versus closed software similar to ‘Linux versus Microsoft’ over the years with the internet. Where one is looking at creating a profitable model by closing off certain aspects or selling the whole model as a one-time or subscription purchase versus pure open-source software, it’s the age-old conflict of closed, controlled, and profitable versus open, free, and for the good of developing and progressing the industry and technology. With adoption potential being much greater with open source options but monetization potential becomes more complicated.
While Linux thrives on collaboration, transparency, and flexibility, Microsoft leverages control, consistency, and comprehensive support. Both models have their unique strengths, and the landscape continues to evolve as they adapt to new technological trends and market demands.
Recent years have seen Microsoft embrace open source in almost a u-turn, as Microsoft is now allowing users on Azure (Microsoft’s cloud computing platform) to run Linux-based applications and services on their cloud platform, alongside other open-source options like MySQL. This shift shows their commitment to being more flexible and open to different software solutions, which is important in to reflect on when comparing to the approach taken by Meta and OpenAi.
We can see Meta taking a similar approach to Linux in that they are heading very clearly into an open-source model which although may typically be associated with community and less so with commerciality - will benefit Meta in the long run through their owned platforms of Facebook, Instagram and Whatsapp.
New Icons take on Metas move.
New Icon being huge supporter of open source software and its pure potential when developing for clients, we hope that the project form meta succeeds and opens more opportunities to develop bigger and better ai-related and ai-integrated projects - with true customisation at the heart.
Don’t get us wrong - we’re also big fans of the ChatGPT api its really good and inexpensive and its a platform with huge potential yet to be explored. Being based on Large language models (LLM) they are trained on billions of pieces of text and content to be able to perform natural language processing.
LLMs can recognize and generate text, interpret human language, and understand how words, characters, and sentences work together. They can also analyze unstructured data and distinguish between content without human intervention. So it’s a huge step forward for ai and a new foundational component that we can build interesting things with. It’s genuinely a significant shift forward and will eventually impact billions of lives in a similar way that the internet has.
What it means for our customers.
The rise of customizable LLMs (Large Language Models) presents a wealth of opportunities for our customers. While these advanced models can supercharge your apps and solve a variety of challenges, the real advantage lies in the ability to train your own LLM on top of a pre-trained base model. This customization opens up a world of innovation, allowing you to tailor the model specifically to your unique needs and business environment.
One significant benefit is the ability to address scenarios where data exceeds current context window limits. By embedding specific knowledge directly into the model, we can create a powerful tool that excels in niche or complex situations. For example, imagine a version of ChatGPT that is not only fluent in your industry’s terminology but also perfectly captures your company’s tone of voice. This level of expertise and customization can transform the way you interact with customers, handle internal communications, and manage specialized tasks.
The landscape of AI is evolving in a way that encourages healthy competition among companies. This competition is crucial because it prevents any single entity from monopolizing the market, ensuring that innovation continues to thrive. For businesses, this means more choices and better options when it comes to selecting the right AI solution.
When considering whether to go for a custom-trained LLM or an off-the-shelf solution, it’s important to weigh up your specific requirements, budget, and timeline. Both options have their merits, but the increasing availability of open-source models is a game-changer, making advanced AI accessible to a broader range of businesses.
In conclusion, the ability to customize LLMs offers tremendous value, enabling businesses to innovate and stay ahead of the curve. The presence of competition in the AI space ensures that this technology will continue to evolve, benefiting companies of all sizes. Whether you choose a custom solution or an off-the-shelf model, the key is to leverage these tools to meet your unique needs and drive your business forward.