The New Age of AI, Language & Chat: Cohere AI, ChatGPT & Google’s Bard
Authors:? Sonam G. & Varun Kumethi (repost with authors permission original post)
Large Language Models (LLMs) and conversational bots have been hot topics on the internet. A new chatbot or LLM is being introduced daily, making it tedious to stay current. So, Varun and I decided to compare some of them. Let’s start with what precisely an LLM is (without getting into the technicalities).
What are LLMs?
Large Language Models are artificial intelligence (AI) that uses machine learning algorithms to generate human-like text. These models are trained on massive amounts of text data and can generate coherent and contextually relevant sentences by predicting the next word in a sequence. They can also answer questions, write essays, summarize texts, translate languages, and generate poetry or creative writing.
Why So Large?
These models are called “large” because they have an enormous number of parameters and are trained on vast amounts of data. For example, Cohere AI’s command models and OpenAI’s GPT-3 were trained on hundreds and hundreds of gigabytes of text from the internet.
However, it’s important to note that while LLMs can generate impressively human-like text, they do not understand the content as humans do. The models don’t have beliefs, emotions, or desires, and any output they generate is purely a result of the statistical patterns they’ve learned during training. Despite their limitations, LLMs represent a significant step forward in natural language processing. They have many potential applications, from chatbots and virtual assistants to content generation.
Image 1 below provides a basic understanding of the process where humans input a query; the query is then transformed into text embeddings (a way of representing the text into valued vectors) that machines understand, get processed by deep neural networks, and then output the answer. LLMs could be used for various tasks, such as summarizing any given text, generating text, and more. Please refer to?Cohere’s Introduction to LLMs?for deeper technical details on how LLM works. Since the boom in LLMs and generative AI, various companies and individuals have produced several applications, and some of the most popular ones are OpenAI’s ChatGPT, Google’s Bard, and HuggingChat. We discuss ChatGPT, Bard, and Cohere’s AI in this article.
What’s ChatGPT like?
ChatGPT is a computer program that learns data from various internet sources and responds to your queries in the most human-like language. Imagine two human beings conversing with each other and responding with the knowledge they have gained over time. In this case, OpenAI trained a computer program in the form of a chatbot on the data from 2021 to converse with you. In technical terms, ChatGPT uses GPT3.5 architecture, a smaller version of GPT3 regarding the parameters used in the model. The GPT architecture is based on the prevalent transformers model with attention. To understand transformers, please refer to the paper, ‘Attention is All You Need.’ ChatGPT can correct itself, correct the grammar of the given text, question-answering, summarize, write, debug codes, translate language, and more. In the following image 2, you see my conversation with ChatGPT, where I asked them to tell me the range of the BLEU score, and it generated a decent answer with not just the content but explained what a BLEU score is. With the recent updates to ChatGPT, we can like/dislike the response as part of giving feedback on the answer quality and turning off the conversation history.
Societal Implications of ChatGPT
ChatGPT’s arrival has brought about seismic shifts in the societal and business landscapes, challenging conventional norms and redefining the interfaces of human-machine interactions. Society benefits enormously from this technology as it democratizes access to information, transcending geographical and socio-economic boundaries. It is an empowering tool for individuals, aiding education by providing detailed explanations on complex topics or helping them learn new languages. It could be a valuable assistant for the differently-abled, enhancing their access to digital platforms.
However, these benefits also come with challenges. The potential for misuse, such as the propagation of misinformation, bias, or offensive content, necessitates careful handling. Also, as AI becomes more integrated into society, data privacy and security issues become increasingly critical. Balancing these concerns with the potential benefits of ChatGPT is a crucial task for developers, regulators, and society.
Business Implications of ChatGPT
ChatGPT is ushering in a new era of efficiency and productivity across various business sectors. The technology holds significant potential for automating repetitive tasks, enhancing customer service by providing rapid, round-the-clock responses, and aiding in content creation, from drafting marketing materials to coding. It can streamline processes, boost customer engagement, and improve customer retention.
Businesses must tread carefully, as there are potential challenges with implementing such AI technology. As per OpenAI, ChatGPT may occasionally produce incorrect or biased content, and privacy and cybersecurity concerns should be addressed due to its internet-based nature. Despite these hurdles, the transformative potential of ChatGPT is evident, and its applications are expected to evolve and improve over time.
Moreover, ChatGPT is limited to the amount of information fed into its system by OpenAI. Using ChatGPT generally to improve your operational workflows, such as automating copywriting tasks, content production, general email replies, etc. However, one major limitation of such a robust chat system is that should you or your businesses need to improve specific operational tasks that are run under the rules of your organization, ChatGPT merely becomes a template to get started. Fine-tuning tasks can somewhat eliminate this limitation; however, fine-tuning jobs are out of the reach of the masses, as they would require further information about parameter adjusting, data collection, training, and, most importantly, coding with OpenAI’s platform. (Source:?Forbes)
What’s Cohere AI like?
Cohere AI, a leading AI startup, has leveraged the prowess of LLMs to create a text generation, classification, and processing system that surpasses many of its contemporaries. Technically speaking, Cohere AI provides an innovative platform that seamlessly empowers developers to incorporate LLMs into their projects. Its APIs and SDKs are versatile and compatible with various stacks, facilitating tasks such as text generation, embedding, and classification.
What sets Cohere AI apart is its user-friendly playground, a feature that not only enhances accessibility to advanced NLP models but also fosters an environment of learning and exploration. This commitment to democratizing NLP technology is a testament to Cohere AI’s vision of making AI accessible to all, regardless of machine learning expertise. The following image 3 is an example of Cohere’s playground for text generation:
Societal Implications of Cohere AI
Cohere AI is reshaping society’s relationship with AI. Its advanced NLP models democratize access to AI, making it a handy tool for everyone, even those without a deep understanding of AI or machine learning. This advantage makes it a game-changer in sectors like education, where Cohere AI can simplify complex topics or aid in language learning.
Nonetheless, Cohere AI also recognizes and is committed to addressing the challenges associated with AI, such as potential misuse and propagation of bias, reiterating its dedication to ethical AI use.
领英推荐
Business Implications of Cohere AI
Cohere AI is pushing boundaries in the business world, driving productivity and efficiency to unprecedented levels. Its platform can automate various tasks, streamline operations, and enhance customer service — all while outperforming other models like ChatGPT. Furthermore, its presence on the Amazon Marketplace amplifies its accessibility, making it a practical choice for businesses seeking to integrate AI.
One of Cohere AI’s standout features is its easy fine-tuning on the platform. This allows businesses to tailor the model to their specific needs, a significant advantage over models like ChatGPT, where fine-tuning can be more complex and less accessible.
While Cohere AI is a powerful tool, it doesn’t shy away from the responsibility that comes with it. The company is proactive in addressing data privacy and security concerns and is committed to preventing the propagation of bias or misinformation. This forward-thinking approach, coupled with Cohere AI’s transformative potential, is set to redefine the business landscape and how we perceive AI in business.
What’s Google Bard like?
Just like ChatGPT, Google released its “rival bot” (as people like to call it) called Bard. It is another generative AI program that helps answer, summarize, and solve coding questions, among other tasks. Bard is a lighter version of the LaMDA (Language Models for Dialogue Applications) model. In simple terms, LaMDA is another seq-to-seq model based on the transformer architecture trained on human dialogues/conversations and stories. The goal of this model was to have human-like open-ended discussions. Given the complexity of LaMDA, which has versions of the models using 2B, 8B, and 137B parameters, Google released Bard to be lighter and more optimized. In the?Google Bard Report, they explained the technical details in more depth, as well as the limitations this model poses, for example, bias in training data, inaccurate and incorrect answers to questions, false positives/negatives, etc.
I asked Bard to generate a code to reverse a given string in Python, and it returned several solutions, as seen in the following image 4:
It generates different drafts/answers for the same question following Google’s policy of no one answer to any question. Towards the bottom, it also cites the sources and related topics for other searches, as seen in image 5.
All of these chatbots and LLMs are competitive with each other. Let’s learn more about the business implications.
Societal Implications of Bard:
As we drafted this blog, Google I/O 2023’s unveiled an update to the Bard chat tool, powered by the new PaLM 2 model, which could significantly impact societal communication methods. Bard’s ability to support more languages and surface images in responses signifies a potential revolution in communication, facilitating more inclusive and interactive conversations. Moreover, introducing AI-centric tools such as Sidekick, which offers contextual suggestions in Google Docs, could redefine how people engage with digital platforms. However, this also brings challenges. For instance, the risk of AI-generated misinformation could lead to societal issues around the trust and authenticity of information (Source:?Google,?TechCrunch)
Business Implications of Bard:
From a business perspective, introducing advanced AI technologies has profound implications. PaLM 2 in Bard can enhance customer interactions and improve content creation in various industries, offering competitive advantages. Launching AI-powered tools like Codey, designed explicitly for code completion, can boost productivity in the tech sector. Additionally, businesses can anticipate more streamlined operations with Google’s Workspace suite incorporating AI to auto-generate tables and images. However, these tools also necessitate robust data privacy measures to ensure user trust.
While these features are up-and-coming, it’s worth noting that they’re yet to be available globally. For example, countries like Canada still await access to these innovative tools. Previous shortcomings, such as Bard’s failure to outperform Bing’s integration with ChatGPT in communication tasks, might have contributed to this limited initial release. It will be interesting to see how Google addresses these past issues to ensure Bard’s global success and maintain a competitive edge in the dynamic AI landscape. (Source:?Google,?TechCrunch)
Limitations and Failures
ChatGPT: Among all the cool things that ChatGPT can do, such as solving a Wharton MBA exam, summarizing long articles for you, or even correcting the grammar of your text, there are, however, some limitations to it. Some of the failures and inaccuracies in the results from ChatGPT that we have seen include circular or repetitive answers to the same question, ethical issues in the solutions in terms of biases, coding errors, the absence of citations for sources it uses, and one of the most important ones: factual errors in some answers.?Borji, A. (2023)?provides many examples of ChatGPT failures in their paper.
Cohere AI:?Despite its strengths, its primary limitation lies in its lack of brand awareness compared to more established models.
Google Bard:?Initially, when Google released Bard, it was filled with errors and inaccuracies. You might come across news articles and blogs from around February 2023 talking about it. Recently, they refined their Bard model and added the capability of solving coding questions. However, this bot’s limitation is sometimes forgetting the context of the previous conversation and facing factual errors. For instance, a?Bloomberg article?about Google employees facing ethical concerns about Bard’s responses or misinformation returned to prompts mentioned in?Fortune.
One thing to note is that none of these bots can let you upload images or generate any. They are purely text-based.
Final Thoughts
Since the introduction of ChatGPT, people around the globe have been going crazy with the plethora of applications, the tool getting banned, people running scared it will replace human jobs could have harmful implications, and many other theories. At a recent Women in Data Science meetup, an OpenAI employee said humans need to be in the driver’s seat when handling these LLMs and other AI models because there are only certain things the bots can do. In one of the?deeplearning.AI?webinars with Andrew Ng and Yann LeCunn, they said that just like how the FDA intervenes for drug approval, the government could regulate the use of such models and bots. “AI is part of the solution and not part of the problem.”
We live through some beautiful technological breakthroughs, especially in AI. These chatbots are amazing, but we must use them responsibly since they come with limitations. Some of the incorrect uses could involve plagiarizing, spreading misinformation, etc. Let’s use them to make our lives easier; for instance, ask the bots to give ideas for a fun project, help you get ideas to write a blog, create a trip itinerary, and more. We want to end our blog with the lines from Cohere’s CEO, Aidan Gomez, “A.I. will make people more effective, not displace them.”
P.S.:?Now eagerly waiting and wondering if?Cohere?will introduce something of its own.
References
Founder at SixThirty Group
1 年Great article! Would love to hear your current thoughts on Cohere and its place in the industry right now.
Stay Positive!!
1 年Thanks for sharing it's such a resourcful information