Everyone is talking about ChatGPT – writers, technology experts, coders, CEOs, even lawyers, and academia. And why not? The latest AI ChatBot from OpenAI has proven to be a sensation for its extraordinary human-like conversational responses. But it has agitated a giant too – Google.
We all know Google rules the internet world. Interestingly, Google possesses a similar technology that many experts unanimously agree is far superior to ChatGPT – it’s called LaMDA. All the more reason that Google is feeling the brunt. After all, despite the muscles and power and an already perfected technology, ChatGPT has taken all the limelight to the point that it’s being called Google Killer.
The question is; will LaMDA again establish Google’s supremacy in the online world, or is ChatGPT on its way to becoming the new king? Let’s try to gauge the possibilities with a side-by-side comparison of Google’s AI LaMDA Vs OpenAI’s ChatGPT.
Google’s AI LaMDA Vs OpenAI’s ChatGPT: Key Takeaways
- LaMDA, or Language Model for Dialogue Applications, is a conversation technology trained specifically to hold human-like conversations.
- Although Google perfected LaMDA a while back, it was hesitant to release it to the general public because the responses are not always correct, which might impact Google’s reputation.
- Also, Google still has to find a way to fit LaMDA into its revenue model, which is mostly ad-based.
- ChatGPT is OpenAI’s Large Language Model, trained on the GPT 3.5 module. In a way, it’s similar to Google’s LaMDA, as both serve the same purpose of human-like conversations.
- Unlike Google, OpenAI is a non-profit organization and, therefore, more flexible in testing its AI chatbot more rigorously and in the real world.
- The free availability of ChatGPT, coupled with its extraordinary caliber to generate crisp responses, is the essential lever that pushed it to unsurmountable popularity.
Google’ AI LaMDA: Brief Overview
LaMDA stands for “Language Model for Dialogue Applications.” The system is composed of a large artificial neural network, a mathematical model mimicking actual neurons. LaMDA was trained on in-house Transformer type neural network architecture developed by Google and made open source in 2017.
The technology is similar to BERT or OpenAI’s GPT3. Incidentally, transformers reflect functionality similar to Recurrent Neural Networks and can process massive amounts of sequential data. Consequently, they are a perfect model for natural language processing.
Interestingly, LaMDA has been trained in millions of dialogues covering diverse topics. In short, LaMDA is an AI technology trained specifically to engage in conversation with the ability to be aware of the context as well as the specificity of the dialogues.
OpenAI’s ChatGPT: Brief Overview
ChatGPT is an AI chatbot designed by the AI research firm OpenAI. It’s the variant of the Generative Pre-trained Transformer or GPT model. Incidentally, ChatGPT is trained on the latest GPT 3.5 model and fine-tuned to mimic human conversational style.
ChatGPT was fed a massive dataset to attain fluency in human-like conversations. On top of that, ChatGPT also exploits the supervised learning model similar to Google’s LaMDA. It means apart from a large data set; actual human trainers were also employed to train ChatGPT to understand the prompts and reply accordingly.
Google’s AI LaMDA Vs OpenAI’s ChatGPT: Capabilities
The capabilities of both LaMDA and ChatGPT stem from the technology behind these tools. To decipher their capabilities, it’s imperative to delve deeper into their working and gauge their capabilities in that context;
Capabilities Of LaMDA
LaMDA is trained on 137B parameters and fed with 1.56T publicly available words, dialogue data, and documents on the internet. On top of that, LaMDA is also fine-tuned to always mold its response on three key parameters – Safety, Quality, and Groundedness.
The dialogue-specific training is its main strength and gives LamDA an edge over other language models. Incidentally, Riley Goodside from Scale AI has pointed out that LaMDA’s responses are more authentic and closer to human speech than ChatGPT. It makes it easier for LaMDA to integrate with applications like Google Assistant, Workspace, and even the search engine itself.
LaMDA is trained to abide by the three matrices in all contexts (Safety, Quality, and Groundedness), which is also its major strength. It makes LaMDA more trustworthy and less prone to errors or presenting fictitious, made-up responses.
Besides, LaMDA emphasizes more on quality and relies on SSI parameters – sensitivity, specificity and interest. This ensures that answers make sense as per the prompt, follow specifically the context and are not generic replies applicable to multiple contexts. On top of that, the interest parameter ensures that the responses are actually insightful and witty – not boring or artificial. It translates to a better experience and a feel for actual conversation, which is the ultimate goal.
Additionally, LaMDA doesn’t rely on predefined responses. It generates replies instantly, understanding the context, the actual demand of the query and how it relates to previous prompts. It allows LaMDA to converse seamlessly in human-like speech without losing the context. It was efficiently demonstrated by Sundar Pichai, CEO of Alphabet, during the Google I/O conference, where LaMDA replied by playing as Pluto – the dwarf planet.
In short, if we speak about conversational style, LaMDA truly aces it. It undoubtedly has a style and perspective and the ability to better adapt to the intricacies of language.
Capabilities Of ChatGPT
ChatGPT exploits the GPT-3.5 architecture, which is based on three models – code-DaVinci-002, text-DaVinci-002, and an additional base model to understand codes. Additionally, the text-DaVinci-002 model was further strengthened by human trainers checking the quality of generated responses.
Besides, GPT 3.5 was further reinforced by reinforcement learning with human feedback or the text-DaVinci-003 model. It is a reward-based training module that allows ChatGPT3 to learn from its mistakes and correct its replies if the same question is asked again. These two factors alone bring ChatGPT to the level of LaMDA. Incidentally, LaMDA is still not fine-tuned to generate codes – so ChatGPT is one step ahead for now.
ChatGPT is also trained for task-specific corpus. You can use it for language translation, summarisation, text improvement, and more. On top of that, ChatGPT is trained over a massively larger dataset than LaMDA, making it more robust for tasks like text generation, translation, drafting content, and more.
Additionally, Microsoft is reportedly looking to integrate OpenAI’s ChatGPT into its Bing search engine. If that happens, ChatGPT might get access to the internet at large, exponentially increasing its capabilities and expanding its use case. It will pave the way for AI Bot to integrate with products and apps, making it a rival to Google. Besides, it will help Microsoft radically improve the user experience and expand its market share, which is meager compared to Google in terms of search engines.
See more to this comparison and discover more of their capabilities when we compared Bard and ChatGPT with Bing AI
Google’s AI LaMDA Vs OpenAI’s ChatGPT: Limitations
Despite their potential, LaMDA and ChatGPT are far from perfect. They have limitations, some more far-reaching than others. In this section, we will explore the limitations that Google’s LaMDA and OpenAI’s ChatGPT have to overcome;
Limitations of LaMDA
As I mentioned earlier, LaMDA is explicitly trained in dialogue data. While it allows it to be more precise in natural language processing, it limits its functionalities. The training explicitly focussed on dialogue-based interactions makes LaMDA highly specialized in generating responses closer to actual human speech; it might restrict it to having a broader understanding of topics.
For now, LaMDA comes with limited flexibility. It converses like a human, sure, but it might lag in other natural language processing tasks, such as language translation and text summarization. Likewise, LaMDA has yet to be publicly available. It restricts the more extensive access of LaMDA and prevents testing of its capabilities. Public access would have allowed researchers and developers to test it rigorously and point out any shortcomings.
Limitations of ChatGPT
ChatGPT has its own set of problems to deal with. One of the most significant issues many have pointed out is that ChatGPT produces irrelevant, shallow, and truly fictitious responses. The ChatBot has also attracted much flak regarding factual inaccuracies in its replies.
Google’s Cassie Kozyrkov garnered much attention when she pointed out how ChatGPT failed to provide correct responses to the prompts about how it relates to GANs. She commented that the response was shallow and needed additional verification.
Also, despite claiming to be a robust model capable of human-like conversations, ChatGPT still needs to catch up to LaMDA. Experts like Riley Goodside have pointed out that ChatGPT’s responses are often monotonic and devoid of emotions.
ChatGPT is still far behind in terms of holding open-ended conversations. It can fetch information based on fact pretty accurately, but it fails when the prompt can have multiple answers. In contrast, Google LaMDA excels in holding open-ended conversations.
On top of that, ChatGPT has limited expertise to continue the conversation where it was left off, generating responses built on the previous context. Besides, it is significantly limited in understanding different conversational styles and language structures and therefore fails to create varied and more interesting responses.
Google’s AI LaMDA Vs OpenAI’s ChatGPT: Availability
In this segment, we will talk about the accessibility to these platforms.
Availability of LaMDA
LaMDA is not as freely available to test as its counterpart – ChatGPT. Earlier in August, Google rolled out the AI Test Kitchen app that enabled users to test and interact with LaMDA and provide feedback. However, access to AI Test Kitchen is only allowed to limited groups of people and in small batches. To test Google’s LaMDA, you can register at https://aitestkitchen.withgoogle.com/. If your request is approved, you will receive an invite that will allow you to access the AI test Kitchen.
Availability Of ChatGPT
OpenAI, playing a master stroke, made ChatGPT publically available in November last year. It’s, I think, a major contributor to its popularity. People from all walks of life have tried it ( you might have heard that ChatGPT crossed 1 million users within a week of its launch ) and got hands-on experience with its amazing human-like responses and the expertise to generate rational responses to complex questions.
There’s no information on how OpenAI will monetize ChatGPT, but for now, it’s freely available. You can visit the OpenAI website and sign up to try ChatGPT. Once you have registered with an email, the ChatGPT interface will be available to you to test the astonishing power of artificial intelligence and the amazing capabilities of ChatGPT.
- ChatGPT Chatbot: How To Use This AI Chatbot?
- Why Is ChatGPT Better Than Alexa And Siri
- How To Do Keyword Research Using ChatGPT?
- Chatbot Statistics
- GPT-4 vs GPT-3.5
- ChatGPT vs Claude
Conclusion: Google’s AI LaMDA Vs OpenAI’s ChatGPT
In summary, while both LaMDA and GPT-3 are capable language models, LaMDA takes things more seriously regarding responses. The built-in mechanism to abide by Safety, Quality, and Groundedness ensures that LaMDA regurgitates more responsible replies, not nonsense. Likewise, LaMDA is heavily trained on dialogue-based interactions, making it closer to human-like responses. Moreover, LaMDA is lightweight and more adept at being fine-tuned. Therefore, while it might not generate code at the moment, it can be easily trained to understand and reply to code-related prompts.
ChatGPT, on the other hand, is based on GPT -3’s large training data and sophisticated architecture. It allows it to generate more nuanced and diverse responses and perform better on more complex tasks like drafting content, translation, code snippets, and more. At this point, ChatGPT is marred with inaccuracies, factually incorrect responses, and an inability to understand the context in some cases. However, the text-DaVinci-003 model that includes reinforced learning allows it to improve and rectify its errors. Therefore, it’s not too exaggerated to assume that ChatGPT will only improve its capabilities in the coming future.
That said, both LaMDA and ChatGPT are language models designed for varying language processing tasks – but each has its use case. Therefore, while ChatGPT can be used in various applications, whether it will compete with Google depends on the actual context in which the technology is applied.