What is the difference between Google PaLM2 and OpenAIGPT-4, and who is more powerful?

Tech 2023-05-17 11:39:58 Source: Network

Google launched the next generation Pathways Language Model (PaLM2) on Google I/O 2023 on May 10, 2023. Its new Large Language Model (LLM) has many improvements over its predecessor (PaLM) and may eventually become the GPT-4 ready to face its biggest competitor, OpenAI

Google launched the next generation Pathways Language Model (PaLM2) on Google I/O 2023 on May 10, 2023. Its new Large Language Model (LLM) has many improvements over its predecessor (PaLM) and may eventually become the GPT-4 ready to face its biggest competitor, OpenAI.

But how much improvement has Google made? Is PaLM2 the differentiation manufacturer Google hopes for, and more importantly, with so many similar features, how is PaLM2 different from OpenAI's GPT-4?

Google

PaLM2 and GPT-4: Performance Overview

PaLM2 has new and improved features compared to its predecessor. One of the unique advantages of PaLM2 compared to GPT-4 is its smaller size, making it suitable for certain applications that do not have as much onboard processing power.

All these animals of different sizes have their own small models, called geckos, otters, bison, and unicorns. Geckos are the smallest, followed by otters, bison, and finally the largest unicorn.

Google also claims that WinoGrande and DROP have improved reasoning abilities compared to GPT-4, with the former leading slightly in ARC-C. However, there have been significant improvements in PaLM and SOTA across the board.

According to Google's 91 page PaLM2 research paper [PDF], PaLM2 is also better in mathematics. However, the way Google and OpenAI build test results makes it difficult to directly compare these two models. Google has also omitted some comparisons, possibly because the performance of PaLM2 is not as good as that of GPT-4.

In MMLU, GPT-4 scores 86.4, while PaLM2 scores 81.2. The same is true for HellaSlag, where GPT-4 scored 95.3, but PaLM2 could only achieve 86.8, and ARC-E, where GPT-4 and PaLM2 achieved 96.3 and 89.7, respectively.

The largest model in the PaLM2 series is PaLM2-L. Although we do not know its exact size, we know that it is much smaller than the largest PaLM model, but more training calculations are used. According to Google, PaLM has 540 billion parameters, so the "significantly smaller" PaLM2 should have 1 to 300 billion parameters. Please remember that these numbers are based solely on the assumptions made by Google in the PaLM2 paper.

If this number is close to 100 billion or less, the parameters of PaLM2 are likely to be smaller than GPT-3.5. Considering that a model that may be less than 100 billion yuan can keep up with GPT-4 and even defeat it in certain tasks, this is impressive. GPT-3.5 initially blew everything out of the water, including PaLM, but PaLM2 has fully recovered.

Differences in training data between GPT-4 and PaLM2

Although Google has not yet announced the size of the PaLM2 training dataset, the company reported in its research paper that the new LLM training dataset is much larger. OpenAI also adopted the same method when launching GPT-4, without making any statement about the size of the training dataset.

GooglePaLM2

However, Google hopes to focus on a deeper understanding of mathematics, logic, reasoning, and science, which means that most of the training data for PaLM2 is focused on the aforementioned topics. Google stated in its paper that the pre trained corpus of PaLM2 consists of multiple sources, including online documents, books, code, mathematics, and conversation data, which has been comprehensively improved, at least compared to PaLM.

The conversational skills of PaLM2 should also be at another level, considering that the model has been trained in over 100 languages to provide better contextual understanding and translation capabilities.

As for the confirmation of the training data for GPT-4, OpenAI informed us that it has trained the model using publicly available data and its licensed data. The research page of GPT-4 states that "data is a network scale dataset that includes correct and incorrect solutions to mathematical problems, weak and strong reasoning, contradictory and consistent statements, and represents a variety of ideologies and ideas

When GPT-4 is asked a question, it can generate various answers, but not all of them are related to your query. In order to align it with the user's intentions, OpenAI uses reinforcement learning and human feedback to fine-tune the model's behavior.

Although we may not know the exact training data for these models, we know that the training intentions are very different. We will have to wait and see how this difference in training intent distinguishes these two models in actual deployment.

PaLM2 and GPT-4 chat robots and services

The first portal to access these two LLMs is to use their respective chat bots, PaLM2's Bard and GPT-4's ChatGPT. That is to say, GPT-4 is behind the paid wall of ChatGPTPlus, and free users can only access GPT-3.5. On the other hand, Bard is free for everyone and available in 180 countries/regions.

This does not mean that you cannot access GPT-4 for free. Microsoft's BingAIChat uses GPT-4, which is completely free, open to everyone, and second only to Google's biggest competitor in the field, Bing Search.

Google I/O2023 is filled with announcements about how PaLM2 and generative AI integration will improve the Google Workspace experience through AI features, which will appear in almost all services provided by Google Docs, Sheets, Slides, Gmail, and this search giant. In addition, Google has confirmed that PaLM2 has been integrated into over 25 Google products, including Android and YouTube.

In contrast, Microsoft has introduced AI functionality into the Microsoft Office program suite and many of its services. At present, you can experience these two LLMs in similar products from two competing companies that are facing each other in the AI battle.

However, due to the early emergence of GPT-4 and its careful avoidance of many of the mistakes Google made on the initial Bard, it has actually become a third-party developer, startup, and almost any other LLM that wants to merge. So far, there are capable AI models in their services.

This is not to say that developers will not switch to or at least not try PaLM2, but Google still needs to catch up with OpenAI in this regard. In fact, PaLM2 is open source and not locked behind paid APIs, which means it may be more widely adopted than GPT-4.

Can PaLM2 compete with GPT-4?

PaLM2 is still very new, so whether it can compete with GPT-4 remains to be answered. However, given everything Google has promised and the radical ways it has decided to spread it, it seems that PaLM2 can indeed rival GPT-4.

However, GPT-4 is still a highly capable model and, as mentioned earlier, has defeated PaLM2 in many comparisons. That is to say, the multiple smaller models of PaLM2 give it undeniable advantages. Gecko itself is very lightweight and can work on mobile devices even when offline. This means that PaLM2 can support completely different categories of products and devices that may be difficult to use with GPT-4.

The competition for artificial intelligence is intensifying

With the launch of PaLM2, the competition for AI dominance has heated up, as this may be the first valuable competitor to compete with GPT-4. With an updated multimodal artificial intelligence model called "Gemini" also being trained, Google has shown no signs of slowing down in this regard.

Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])

Mobile advertising space rental

Tag: is and What the difference between Google PaLM2 OpenAIGPT-4