The pace at which increasingly powerful AI Chatbots and models are released to the public has accelerated in recent times. We just witnessed the release of Claude 2, which is likely the second-most capable AI system available to the public.
Before that, OpenAI launched Code Interpreter, the most sophisticated artificially intelligent model to date. And even earlier, certain intelligent systems gained the ability to process visual content.
I can offer some orientation on the current state of this technology, even though it might just be a fraction of the full story. As an educator, I have been compiling and sending a ‘Getting Started Guide to AI’ for my students and interested readers every month.
However, the rapid advancements in the field have necessitated frequent modifications, and the past couple of months have been particularly mind-boggling.
Current State of AI
It’s important to note that when we talk about Artificial Intelligence today, we are typically referring to Large Language Models (LLMs). The majority of intelligent applications rely on LLMs, with a limited number of foundational Models developed by a few organizations.
Each company provides direct access to its models through chatbots. OpenAI’s GPT-3.5 and GPT-4 power ChatGPT and Microsoft’s Bing. Google has a range of models labeled as Bard, while Anthropic offers Claude and Claude 2.
There are other LLMs not covered in this discussion. One such example is Pi, a chatbot created by Inflection. Pi excels in conversations but is not very useful for other tasks.
Additionally, various open-source models exist that anyone can utilize and modify, although they may not be accessible or practical for casual users at present. Future guides may incorporate these models.
GPT-3.5 and GPT-4, developed by OpenAI, are the primary Artificial intelligence systems available. GPT-3.5 initiated the current craze in November, while GPT-4 premiered in the Spring with significantly enhanced capabilities.
A variant of GPT-4 allows plugin integration with the internet and other applications. While many plugins may not be particularly useful, users are encouraged to explore them when necessary.
Code Interpreter, an incredibly powerful version of ChatGPT capable of running Python programs, is also available. If you haven’t paid for OpenAI, you have likely only experienced GPT-3.5.
Apart from the plugin variation and a temporarily suspended browsing feature in GPT-4, these models are not connected to the internet. Microsoft’s Bing utilizes a combination of GPT-4 and GPT-3.5, often being the first to introduce new features within the GPT-4 family.
Bing can create and view images as well as read documents within a web browser. It is connected to the internet, although it can be a bit challenging to use despite its capabilities.
Google has been improving its Artificial Intelligence model, referred to as Bard. Bard utilizes a variety of Foundation Models, including the recent PaLM 2. However, Google’s AI offerings have been somewhat underwhelming, although recent improvements indicate ongoing work to enhance the underlying technology.
Anthropic’s Claude 2 stands out for its expansive context window, allowing it to retain a vast amount of information. It can store an entire book or multiple PDFs in its memory. Moreover, Claude has been designed to exhibit less malicious behavior compared to other Large Language Models, resulting in occasional admonishments.
Now, let’s explore some practical applications of these AI systems:
Creating Images
Nice platform: Adobe Firefly
Open-source tool: Stable Diffusion
Free Tools: Bing or Bing Image Creator (utilizing DALL-E), Playground (enabling multiple models),
Quality images: Midjourney
Below are four notable image generators that are available to the public:
- DALL-E: Integrated into Bing (in creative mode) and Bing Image Creator. It offers reliable performance but is inferior to Midjourney in terms of image quality.
- Adobe Firefly: Built into various Adobe products, it lags behind DALL-E and Midjourney in terms of image quality but is more transparent about image sources, using only those with proper rights.
- Stable Diffusion: This is open-source and can be run on high-end computers. It provides excellent results when prompts are crafted effectively.
- Midjourney: Considered the best system as of mid-2023, with a low learning curve and impressive results. Usage is as simple as typing something you want to see. It requires a Discord connection.
It is important to be aware of biases within these systems, as they are trained on internet data and may exhibit certain predispositions. Transparency regarding the training data and the legal implications surrounding generated images are also issues to consider.
Generating Ideas
Best free tool: Bing
Paid option: ChatGPT 4.0, although Bing’s internet connections often make it the better alternative
Artificial Intelligence is ideal for idea generation, as it can produce a high volume of ideas and be forced to exhibit high creativity through effective prompting. Engaging Bing in creative mode and exploring unusual idea-generation techniques can yield intriguing results.
Random prompts or favorite superheroes can also inspire unique ideas.
Writing
Best free tools: Bing and Claude 2
Paid tools: ChatGPT 4.0/ChatGPT with plugins
Currently, GPT-4 remains the most capable digital tool for writing. Users can access it for free through Bing by selecting “creative mode” or by subscribing to ChatGPT for $20 per month. Claude, however, is a close competitor, and it offers a limited, free option.
These powerful digital writing tools are also being built into popular office applications. Microsoft Office is even incorporating a copilot powered by GPT, while Google Docs integrates suggestions from Bard. These innovations have significant implications for the field of writing.
Here are some ways to utilize Artificial Intelligence in writing:
- Generating drafts for various types of content, such as blog posts, essays, speeches, and more, by providing prompts
- Enhancing the quality of your writing by requesting improvements or making suggestions tailored to specific audiences
- Assisting with tasks like writing emails, creating sales templates, and generating your next business plan
- Overcoming writer’s block and gaining momentum through AI-generated inspiration
Also Read: IoT Made Easy: A Simple Breakdown of the Internet of Things
However, it is crucial to be cautious of potential issues. Artificial Intelligent systems may generate false or incorrect content, even though it appears plausible.
Verification of the information is necessary. Additionally, These systems don’t provide genuine explanations for their outputs but rather generate text that mimics the explanation process.
Understanding biases within these systems is challenging, despite their likely existence. Lastly, ethical responsibility lies with the user to ensure appropriate and responsible use of these tools.
Creating Videos
Good animation tools: Runway v2 is used for creating videos from text, while D-iD is for animation
A good tool for voice cloning: ElevenLabs
Artificial intelligence enables the generation of videos featuring entirely AI-generated characters, voices, scripts, and animations. It is even capable of deep-faking people.
However, caution must be exercised to ensure ethical usage. Runway v2 is a recently released text-to-video tool that produces short 4-second clips, showcasing potential developments in this field.
Obtaining Information and Learning
Best free tool: Bing
Other useful tools: Children can use Khanmigo from Khan Academy
While using these Chatbots as a search engine is not recommended due to potential hallucinations and the limited internet connectivity of most models, recent pilot studies indicate that they can often provide more useful answers than traditional search engines when used properly.
Bing, in particular, proves valuable information in areas where search engines may fall short, such as tech support, restaurant recommendations, or seeking advice. However, caution is still advised.
Handling Documents and Data
For data and code-related tasks: Code Interpreter
For documents: Claude 2 for large documents or Bing Sidebar for smaller documents and webpages
Code Interpreter, a mode of GPT-4, enables running code, uploading files, and downloading AI-generated outputs. It can execute programs, perform data analysis, and create various files, games, and web pages.
While concerns have been raised regarding untrained individuals using Code Interpreter for analysis, many experts have been impressed with its capabilities, suggesting it may revolutionize the field of data science.
Claude 2 excels at working with text, including PDFs. It has shown impressive results when processing entire books, and the new model is even more powerful.
Users can paste complex academic articles and request summarized results, followed by further inquiries. However, verifying results for accuracy is still necessary.
Conclusion
Artificial Intelligence holds promise for educational applications, including self-guided learning. Artificial Intelligent models can explain concepts effectively and serve as an automated tutor. Although there is a risk of hallucinations, users can cross-check critical information against reliable sources.
The field of Artificial Intelligence is evolving rapidly, and the tools discussed here are likely among the least advanced options you will encounter. It is essential to recognize that AI is a tool, and it is not always the appropriate tool for every task.
Consider the limitations and ethical implications associated with these systems. The responsibility lies with users to employ AI ethically. Don’t forget to share any additional use cases or ideas you may have in the comments section.
