(Edit: This blog has been updated to include the latest news 06/05/2024)
In the ever-evolving landscape of AI, 2024 has witnessed a remarkable surge in the capabilities and accessibility of Large Language Models (LLMs). Cutting-edge AI systems have revolutionised the way businesses operate, offering unprecedented opportunities for innovation, efficiency, and competitive advantage. LLMs have cut down on customer service tickets, streamline document/data analysis, and generate high-quality content at scale. With a diverse array of LLM offerings available, it’s the perfect time to harness the power of AI.
Legend
- API Available: This indicates whether developers can access and integrate each AI model into their software applications through a programming interface.
- Online Interface: This shows whether each AI model has a web-based or app-based interface where users can directly chat or interact with it.
- ELO Ratings: This provides a numerical rating that reflects the relative performance or capability of each AI model compared to others. These are based on open-source metrics and also incorporate user feedback.
- Multimodal: This specifies whether an AI model can understand and generate different types of data beyond just text, such as images or audio.
- Context Window: This shows the maximum amount of text (in pages, 1000 tokens = 1 page) that each AI model can consider. This is not 100% accurate as it depends on a lot of variables, however can be used as an approximation of how much custom data could be used with these models.
- Open Source: This indicates whether the code and data used to create each AI model are publicly available or proprietary (closed-source).
- Censorship: This rates the level of content filtering or censorship applied to each AI model’s outputs, ranging from low to high.
A quick prompt comparison can show a lot about how different models work:
What services could an AI company offer? How could it help a regular business? Give a line or two on this.GPT-4 | An AI company can offer services like predictive analytics, customer service automation, and personalized marketing. These services can help a regular business by optimizing operations, enhancing customer experiences, and driving targeted marketing strategies, ultimately leading to increased efficiency and growth. |
Claude 3 | An AI company could offer services like natural language processing, computer vision, predictive analytics, process automation, and conversational AI. These could help businesses automate tasks, gain data insights, and improve customer experiences – driving efficiency and revenue. |
Gemini | Absolutely! Here’s a breakdown of some AI services and their impact on businesses: Customer Service Enhancement: AI-powered chatbots can provide 24/7 support, answering common questions and resolving simple issues, freeing up human agents for complex inquiries. Data-Driven Decision Making: AI algorithms can analyze vast datasets, uncovering patterns and insights that would take humans much longer to identify, improving strategic business choices. Process Automation: AI can streamline repetitive tasks like data entry and report generation, increasing efficiency and saving employees valuable time. … |
Mistral Large | An AI company can offer services such as natural language processing, chatbots, predictive analytics, image and speech recognition, and personalization. These services can help a regular business by automating tasks, providing insights from data analysis, and creating personalized experiences for customers, leading to improved efficiency, reduced costs, and increased customer satisfaction. |
Falcon LLM | An AI company can offer a wide range of services to businesses, including chatbots for customer service, natural language processing for data analysis, and machine learning algorithms for predictive analytics. They can also provide personalized recommendations, fraud detection, and image or speech recognition services. |
LLama3 90B | An AI company could offer a wide range of services, including data analysis and visualization, predictive modeling, natural language processing, automation of repetitive tasks, and more. By leveraging these services, a regular business could improve operational efficiency, gain valuable insights from their data, and enhance customer experiences, ultimately leading to increased productivity and revenue growth. |
Only Google’s Gemini wrote a lot more than “a line or two”.
The Frontrunners of the LLM Race
GPT-4 (OpenAI)
There’s no denying OpenAI started the race, but that doesn’t mean it’s still the front-runner. Much of GPT-4’s popularity is due to its initial momentum and longer time on the market, allowing it to establish extensive support systems. This does make GPT-4 the easiest to manage and use as a business, but it does fall behind when it comes to utilising larger amounts of custom data and its high levels of censorship. These limitations have left room for competitors to also gain a foothold.
Claude 3 Opus (Anthropic)
Anthropic’s Claude 3 Opus has quickly become a favourite in the LLM arena. Scoring higher in ELO rankings, Opus has become the AI enthusiast’s favourite LLM experience. Renowned for its exceptional language understanding and generation abilities, this model offers a staggering context window of 200K tokens for online interactions and an impressive 1M tokens for private API access. While still containing censorship, at least Anthropic is more transparent in its processes and takes user input into account. – Claude 3 has secured a strong spot in the LLM race.
Gemini Pro (Google)
Google’s Gemini Pro is a powerhouse in the LLM realm, integrating with Google services. Boasting a context window of 100K tokens for online interactions and a staggering 1M tokens for preview API access, Gemini Pro’s marketing sets it up for promise. Yet, it is still lagging behind GPT-4 and Claude 3 Opus in real world performance. All of these things considered, its only real value is for businesses that want to stay within the Google ecosystem, as services integrate well. Additionally, Google’s approach has faced criticism over excessive censorship and concerns about quality control in the models they release. Its long-term value will depend on how Google addresses these issues and refines its approach moving forward.
Mistral Large (AWS)
Mistral Large has garnered significant attention – with its low censorship and integration with AWS services, Mistral Large offers businesses a flexible and customizable AI solution. While its context window may be more limited compared to some competitors, its open-source nature allows for greater transparency and adaptability. The model aims to enable developers to design their content moderation policies allowing companies to choose what to restrict, although the exact level of filtering is not specified. As a European company, Mistral AI is closely tracking the development of the upcoming EU AI Act, making it an attractive choice for European companies.
Falcon LLM
Falcon LLM, an open-source model developed by researchers at Tsinghua University, has emerged as a promising contender in the LLM space. It lags behind most metrics and also requires some technical experience to even try out. However, its advantage comes from its open-source availability, meaning that you have the flexibility to use it in a development project, even for fine-tuning. While this is not a task most businesses would find useful, it is a niche for those who want to not just observe the LLM race but be a part of it.
LLaMA 3 (Meta AI)
LLaMA 3’s release gathered widespread hype within open-source communities, and we see why. Not only do their flagship 90B models perform exceptionally well, performing just as well as GPT-4 when it was first released – it achieves this with a model close to 19 times smaller. Adding insult to injury, this size difference is not just an engineering accomplishment but directly translates to their pricing. For processing 1000 pages of input, you’d pay only $0.9 with LLaMA 3, while with GPT-4, the cost would be around $15. Remarkable.
Another important note is that with the release of their flagship model, various smaller open-source models also perform highly, relative to bigger models. While not state of the art, they are an attractive choice for businesses interested in maintaining their own AI infrastructure and possibly custom fine-tuned models.
ERNIE 3.0 Titan (Baidu)
Baidu’s ERNIE 3.0 Titan is a formidable LLM, combining language understanding with multimodal capabilities and a secure, highly censored environment through Baidu’s API. While details on context window and API access are scarce, rumours of Apple’s interest in potential partnerships could signal ERNIE’s rising prominence. As of now, due to limited knowledge of both the model and what service from Baidu would look like, it’s not a safe choice for most users.
Grok-1
Elon Musk’s Grok-1 is an open-source LLM that has garnered significant interest from the AI community. With its low censorship and user-friendly online interface (Grok Chat), Grok-1 provides businesses with an accessible and transparent AI solution. However, its limited context window and lack of API access restrict its capabilities for most applications. Nevertheless, it’s a contender you should keep an eye on.
Making the Right Choice for Your Business
Choosing the right LLM for your business is no easy feat, but it’s a game-changer. With a diverse range of models out there, it’s crucial to evaluate your specific needs, budget, and use cases to make an informed decision that aligns with your goals. Factors like context window, censorship level, interface options, and open-source vs. proprietary solutions all come into play. Some models even excel at processing different data types like images and audio. The key is striking the right balance between capabilities, cost, and ease of integration – there’s no one-size-fits-all solution.
Need help finding what works for you? Look no further than AI Heroes. Our scouting resources are tailored to help you find the perfect model that aligns with your unique business needs.
