Llama-2 vs GPT-4 - who wins?

Good morning and happy hump day! Did you know that more than 2 million people in the Philippines help feed data to LLMs like ChatGPT? Neither did we. This issue goes out to all of those unnamed people who help push AI forward.

Here's what’s cookin’ today.

• 📈 OpenAI Set to Surpass $1 Billion in Annual Revenue

• 🦙 Llama-2 Matches GPT-4 in Factual Summarization Accuracy

• 🛍️ Alibaba Unveils Advanced, Open-Source AI Models

If you’re ready, let's get into it!

OpenAI Is Not Going Anywhere

📈 OpenAI To Surpass $1 Billion in ARR

Overview

Whoever started the rumor that OpenAI was going bankrupt is completely wrong. Although they are spending $700,000 per day, OpenAI, backed by Microsoft, is projected to exceed $1 billion in revenue over the coming year from the sale of AI software and its computing power. This figure is notably above prior revenue projections shared with its shareholders. The significant increase in revenue is largely due to the success of its ChatGPT chatbot and its growing adoption by app developers and companies.

Key Points

• OpenAI's anticipated $1 billion revenue represents an enormous leap from its revenue last year, which stood at $28 million before the introduction of ChatGPT.

• The ChatGPT chatbot is driving up OpenAI's earnings as businesses and app developers find innovative ways to utilize its conversational text technology for revenue generation and cost reduction.

• The growth and success of OpenAI are being tracked by Microsoft, Google, and other businesses invested in similar technologies, signaling potential competitive influences in the sector.

Why It Matters

OpenAI's projected revenue growth underlines the financial potential of AI technology and its substantial role in today's market. Its success could influence the strategies of other companies in the sector, indicating a potential shift towards increased investment in AI technology. Furthermore, OpenAI's commercial triumph could contribute to the broader development and progress of artificial intelligence, elevating its potential in different industries.

Source: TheInformation

Never Count Mark Out

🦙 Llama-2 Matches GPT-4 in Factual Summarization Accuracy

Overview

Researchers compared numerous language learning models (LLMs) including open source LLM Llama-2 and OpenAI's gpt-4 on their factuality in summarization. Using a set of 373 news reports, they determined that Llama-2-70b is almost as accurate as gpt-4 and significantly better than gpt-3.5-turbo. The study identified and overcame issues such as LLMs not following instructions and ordering bias.

Key Points

• Llama-2-70b and gpt-4 demonstrated superior factuality in summary generation, nearly matching human accuracy levels, significantly outperforming gpt-3.5-turbo.

• Llama-2-70b, despite its longer tokenization than ChatGPT, was found to be 30 times cheaper than GPT-4.

• They conducted the experiment using Anyscale Endpoints, revealing a significant ordering bias in gpt-3.5-turbo and smaller Llamas models.

Why It Matters

The experiment signifies a promising development in the field of language learning models, with an open-source LLM matching the performance of proprietary models. This achievement not only makes accurate summarization technology accessible to a larger range of users but also underlines the potential cost efficiency. However, the study also highlights potential pitfalls, like ordering bias, that can affect task performance, emphasizing the need for thorough testing and evaluation.

This is why no one should ever count Mark Zuckerberg out. He will always find a way to compete at the highest level. We suspect Llama-2 will be amongst the big three LLMs.

Source: AnyScale

Our Latest AI Tutorial

🎥 How To Relight Images Using AI - No Photoshop Required

Want to learn how to relight images using a free AI tool? In our latest YouTube video, we explain how to do exactly that. Check it out here.

AI and E-Commerce Make The Perfect Match

🛍️ Alibaba Unveils Advanced, Open-Source AI Models

Overview 

Alibaba launched two new artificial intelligence models named Qwen-VL and Qwen-VL-Chat, designed to comprehend images and carry out complex conversations. These models not only supersede the capabilities of Alibaba's previous AI products, but they will also be open-source. This move allows worldwide researchers, academics, and companies to implement and develop their own AI applications without the need for system training.

Key Points

• Alibaba's AI models, Qwen-VL and Qwen-VL-Chat, can handle open-ended queries about different images, generate photo captions, and carry out complex interactions like comparing various image inputs and responding to multiple rounds of questions.

• These models are built on the large language model named Tongyi Qianwen that Alibaba launched earlier this year. This model's unique attribute is that it supports the training of the AI model on vast amounts of data.

• As part of its strategy to gain more users for their AI model, Alibaba made the model open-source, which allows for its utilization without licensing fees, particularly beneficial for its cloud division as it prepares for public listing.

Why It Matters 

The introduction of these advanced open-source AI models by Alibaba indicates a significant stride toward global leadership in AI technology. Making the AI models accessible to researchers and companies worldwide not only underscores the opportunities for innovation but also saves time and resources. It signifies the potential expansion of AI applications in various sectors, offering more user-friendly and intricate interactions.

Source: CNBC

Tool Highlight From Our Database

EchoFox

The Funding Corner

💰The Companies Of Tomorrow

1. ConverSight Secures $9M to Intensify Data Analytics Tools

2. Healthcare Robotics Firm Mendaera Gains $24M Series A Funding

3. Lex Raises $2.75M for AI-Integrated Writing Tool

That’s it for today! Thanks for reading. If you enjoyed this issue head over to https://bestofai.com to find more tools, tricks, insights, and news.

Until next time, stay safe.

-Best of AI Team

Note: This issue was written in collaboration with ChatGPT, and Bing Chat and edited by a human. Welcome to the future.