General Information
Let’s start with the history of AI and what got us to this point.
Artificial Intelligence (AI) has been an area of research and development for several decades. The term “artificial intelligence” was coined by John McCarthy at the Dartmouth Conference in the 1950s. The first AI programs were developed during this time, including a checkers-playing program by Arthur Samuel.
In the 1960s, the field of AI was established as an academic discipline, with the founding of the first AI laboratory at MIT. The first expert systems were developed, which were computer programs that could solve problems in specific domains.
During the 1970s, AI research shifted towards knowledge-based systems, which used symbolic reasoning to make inferences based on facts and rules. The first successful natural language processing system was developed, called SHRDLU, which could understand simple English commands.
In the 1980s, AI research focused on building “connectionist” systems, which used artificial neural networks to learn from data. Expert systems became popular in industry, with applications in finance, medicine, and other fields.
During the 1990s, AI research continued to make progress in areas like computer vision, speech recognition, and machine learning. The World Wide Web was developed, which provided a new source of data for AI systems.
In the 2000s, AI research focused on data-driven approaches, with the development of algorithms like Support Vector Machines, Random Forests, and Deep Neural Networks. The first successful autonomous robots were developed, including the Roomba vacuum cleaner and the DARPA Grand Challenge self-driving car competition.
Throughout the 2010s, AI made significant progress in areas like image recognition, natural language processing, and game playing. Deep learning became a dominant approach in AI research, with breakthroughs in speech recognition, machine translation, and other areas. AI applications became more widespread in industry, with the development of self-driving cars, virtual assistants, and other products.
Notable developments in the 2010s included IBM’s Watson computer defeating two former champions on the quiz show Jeopardy!, the launch of Google Brain, Facebook’s AI research group, the controversial “Tay” chatbot released by Microsoft, Google’s AlphaGo defeating the world champion in the game of Go, self-driving car technology making significant progress, and OpenAI developing language models like GPT-2 and GPT-3.
The 2020s saw AI playing a significant role in the fight against the COVID-19 pandemic, with applications in disease diagnosis, drug discovery, and contact tracing.
Throughout the history of AI, there have been many debates and controversies surrounding the development and use of these technologies. Ethical concerns, including bias, transparency, and accountability, are increasingly being addressed as AI becomes more pervasive in society.
Artificial Intelligence, or AI, has been in the public eye a lot more recently, than ever before. The field is making recently. Starting sometime around January is when the AI ball started rolling and it’s continued to increase speed since then. There have been hundreds of breakthroughts in the industry and it’s taken the world by storm. There’s been a lot of strong, solid, foundational information about AI, but there’s also been a lot of misconceptions, misinformation, and just incorrect or untrue statements.
Throughout the last two decades AI has been something that’s interested me on a deep level, before it ever existed. It was something that was deeply intriguing to me, even on a fictional and science fiction level.
Now that AI has actually hit the market I’ve been obsessing over everything related to it. Every single API, AI service, the AI services of various companies, the news, the industry on a deep level. I’ve spent a lot of time working with almost every AI service I could find (both free and paid). I’ve been watching a lot of stuff popping up in the news that I believe is full of misinformation and misconceptions, so I wanted to take the time to write something that will hopefully bring clarity and less confusion to anyone that runs across it.
Artificial intelligence (AI) is commonly found in three forms:
- Out-of-the-Box Service: AI is offered as a ready-to-use service that provides predefined functionality and can be used directly by the end user. These services are designed for ease of use and often require minimal technical knowledge to get started.
- API-Based Interaction: AI is made accessible through application programming interfaces (APIs) that allow developers to interact with and utilize the AI capabilities within their applications. This form of AI provides flexibility and customization options for integrating AI features into software products.
- AI-Enabled Applications: AI is embedded within applications to enhance the user experience by automating specific tasks or providing intelligent insights. In this form, the AI operates behind the scenes, allowing users to enjoy the benefits of AI-powered features within the app environment without directly interacting with the AI itself.
I have spent a great deal working with all 3 forms. I want to go into detail about each one starting at the top level and working our way down.
Form 1 (Out of the box service)
The first form is typically the highest level (and to me, generally the most fun). These are companies that have created an AI, wrapped it into something you can interact with directly, and release it to the general public. These are typically referred to as “Large Language Models”. Let me give you a few examples: Open AI’s ChatGPT, Google’s Bard, Facebook’s/Meta Llama, You.com, Chinchilla. There are others but these are the most popular (and promising) at the moment. Let’s explore them one at a time. These models are listed that are standalone. A lot of the other AI systems use one of these as a backend, but I’ll explain that soon. For now let’s focus on these main large language models.
Open AI (ChatGPT) – ChatGPT is currently the smartest, most advanced, and well developed AI on the market. They are at the front of the new AI boom and they’re technology is what’s spurring everyone else talong in the AI race as they all struggle to catch up. At the moment this is bleeding edge and the best one on the market. Most apps out there are powered by ChatGPT API (I’ll explain that later).
To register all you have to do is go to chat.openai.com and register for an account. You can access GPT 3.5 Legacy if you are a free user. If you want to pay for a premium account then you’ll have access to GPT 3.5, GPT 3.5 Turbo, GPT 4 (if you got approved for the waitlist, limited to 25 messages every 3 hours currently), and then if you have plugin support you’ll also have the following models GPT Browsing, GPT Code Interpreter, and GPT Plugins. Here are a breakdowns of the models.
- ChatGPT 3.5 Legacy: This version of ChatGPT is available to free users. It provides a more moderate performance compared to the other versions. While it can solve basic mathematical problems, it may struggle with more complex equations. Originally this was the only version available. Then they created a “Turbo” version (which is now default) and it runs faster and doesn’t sacrifice much in the way of power. Whenever I have to use default I stick with the turbo. Legacy I feel is eventually going to be phased out.
- ChatGPT 3.5 Default: This is another version of ChatGPT 3.5, available to paid users. It performs better than the Legacy version, offering explanations of different methods to solve mathematical problems. However, it may not be as accurate as ChatGPT 4 in solving complex equations.
- ChatGPT 4 (ChatGPT Update 4): This is the most recent and advanced version of ChatGPT, available to paid users. It is capable of solving complex mathematical problems accurately, and it can accept visual inputs. It outperforms both ChatGPT 3.5 Legacy and ChatGPT 3.5 Default in terms of accuracy, speed, and capabilities.
- Browsing Plugin (Alpha): The browsing plugin is an experimental model that allows ChatGPT to browse the internet. This plugin enables the language model to read and retrieve information from the web, expanding the amount of content it can discuss and providing fresh, up-to-date information beyond the training corpus. By leveraging the browsing plugin, ChatGPT can access and discuss current events and recent information, thus improving its ability to provide accurate and timely responses to user queries.
- Code Interpreter: The code interpreter is a plugin for ChatGPT that provides a working Python interpreter within the ChatGPT language model. It is designed to handle, upload, and download Python code. The interpreter comes with a sandboxed, firewalled execution environment, and some ephemeral disk space. This plugin enhances ChatGPT’s ability to interpret, generate, and execute Python code in a secure and controlled manner.
- Plugins model: A model that gives access to third party plugins. There are some that were currently released but more will be added after it goes live. Right now it’s in waiting list access only. Below are the ones they currently have releasd.
- Expedia: The Expedia plugin allows ChatGPT to assist users in planning and managing their trips, including making travel bookings, finding accommodations, and providing information about travel destinations.
- FiscalNote: The FiscalNote plugin enhances ChatGPT’s ability to provide information and insights related to government and regulatory affairs, helping users navigate and understand legislative and policy matters.
- Instacart: The Instacart plugin enables ChatGPT to assist users with grocery shopping, including creating shopping lists, finding products, and placing orders on the Instacart platform.
- KAYAK: The KAYAK plugin allows ChatGPT to assist users with travel-related queries, including finding and booking flights, hotels, and rental cars through the KAYAK platform.
- Klarna: The Klarna plugin enhances ChatGPT’s capabilities in the area of online shopping and payments, allowing users to access Klarna’s services and manage their transactions.
- Milo: The Milo plugin provides ChatGPT with access to Milo’s platform for task management and productivity, helping users create, manage, and track their tasks and projects.
- OpenTable: The OpenTable plugin allows ChatGPT to assist users in finding and booking restaurant reservations through the OpenTable platform.
- Shopify: The Shopify plugin enhances ChatGPT’s capabilities in the area of e-commerce, allowing users to access and manage their Shopify stores, products, and orders.
- Slack: The Slack plugin allows ChatGPT to interact with the Slack platform for team communication and collaboration, enabling users to send and receive messages, manage channels, and more.
- Speak: The Speak plugin provides ChatGPT with access to Speak’s platform for language learning, helping users practice and improve their language skills. One thing I found really interesting about this one was it taught me about “Ancient” languages versus Modern languages. I had a conversation with ChatGPT basically saying that it is better at translating Ancient languages but more modern langauges might be better having through the “Speak” app.
- Wolfram: The Wolfram plugin enhances ChatGPT’s capabilities in the area of computational knowledge, allowing users to access Wolfram’s services for solving complex mathematical problems and accessing scientific data.
- Zapier: The Zapier plugin allows ChatGPT to interact with over 5,000+ apps such as Google Sheets, Trello, Gmail, HubSpot, Salesforce, and more, enabling users to automate workflows and integrate different applications.
Later in this post I’ll discuss the differences between GPT 3.5 and GPT 4 in detail, but for the context of this conversation just go with “GPT 4 is much better, smarter and feature filled than GPT 3.5).
Open AI’s Chatgpt is the one I use most of the time now unless I am experimenting with others (and they all pale in comparison currently).
Out of all the models out there on ChatGPT this is how I use them. The 3 main ones I use are GPT 3.5 Turbo, Browsing (Alpha), and Code Intepreter (Alpha). Both browsing and code interpreter use GPT 3.5 as well. I use GPT 3.5 by default, browing when I need it to have internet access, and code intepreter when I am dealing with files I need to upload or work with, code related stuff (sometimes) or advanced calculations. I have 3 default chats (one for each model). Then I use GPT 4 for anything related to my writing (brainstorming ideas, research, or giving me writing advice and feedback based off my writing. Or anything way more advanced that needs GPT 4 capabilities. There is a limit but I typically stay maxed on the limit so only use it when it’s something I need it for. I use the plugin model sometimes with various models to play around with it. I’ve messed around with them all.
Almost everything I need, I find in ChatGPT in some way.
One more thing worth mentioning Bing Chat. Bing Chat is powered by GPT 4 and it’s just programmed differently. It has more “cohesive” internet access that GPT Browsing model in Chatgpt, but I still tend to use ChatGPT for everything. I do use Bing sometimes when I need to have a conversational connection to the internet and I’ve found it really helpful about conversation research topics and short conversations about various things that need internet access.
Bing chat got off to a rocky start, but I love it for what it is. It works well and serves a good person. They’re deeply integrating it into a lot of things and I think that’s an amazing thing.
Googles (Bard) – This is another one that may eventually be very promising. They got off to a rocky start because they tried to rush into the AI market prematurely to try to catch up with OpenAI and Microsoft. They recovered from that when they released the first version of Bard. Right now it is like.. 15-20% of the capabilities of what ChatGPT can do but they claim they’ll be launching a much larger model in the near future. It’s one I have been keeping an eye on but have yet to be impressed by. They are a large company so it has a lot of growth potential in the future.
Facebook/Meta (Llama) – They got off to a rocky start when their entire code base got leaked originally and people were downloading their entire AI locally and using it. They’ve released a few invite only beta’s but this is one I do not yet have access too (one of the few) but I did work with the local released version. Again this one pales in comparison to ChatGPT currently. But just like Google, it’s backed by a huge company so it has a lot of potential for growth in the future.
You.com – This one has an overall good presentation. They have an integrated search engire and then products like youwrite, youimagine and so forth. In my opinion this is probably one of the weakest AI’s on the market. The interface and implementation are beautiful and flows well. but the logic is horribly paled in comparison to any of the others. Mixed with almost useful logic in most situations it just isn’t anything that stands up to today’s AI race. Not yet anyway. If they put a focus on continuing to develop the AI itself (the logic behind it) they have the potential to take the world by storm, even up against some of the big dogs. Time will tell if they get it right or not.
Form 2 (API Based Interactions)
Right now the market is booming with AI integration and AI powered apps (form 3). The API is what I want to focus on right now. There are a lot of companies out there handling AI, but there is only one that has a strong, stable, functional, cutting edge API. That is Open AI.
OpenAI offers an API for GPT-3, GPT-3.5, and several other models based on these two. GPT-4 is currently on a waiting list only. OpenAI’s API powers 95% of AI-based apps on the market, making GPT-3.5 and GPT-4 the predominant APIs available.
Now let’s revisit another topic. Let’s do a more detailed breakdown of GPT3.5 versus GPT4.
| Aspect | GPT-3.5 | GPT-4 |
|---|---|---|
| Intelligence | Performs well in language understanding and generation | Smarter; can handle longer prompts and more complex conversations |
| Factual Accuracy | Prone to factual errors | Less likely to make factual errors |
| Speed | Faster in generating responses | Slower to respond and generate text (due to larger size) |
| Prompt Restrictions | No hourly prompt restrictions | Comes with hourly prompt restrictions |
| Multimodality | Text-based language model (text input/output) | Multimodal system capable of accepting both text and image inputs |
| Model Size | 175 billion parameters | Significantly more parameters. |
| Training Data | Extensive training dataset | More extensive training dataset (broader knowledge base) |
| Hallucination | Prone to “hallucination” (making things up) | Less likely to hallucinate or provide factually inaccurate responses |
| Complex Reasoning | Good performance in general tasks | Better at handling tasks that require complex reasoning |
| Comprehension of Long Text | Good for general language understanding | Better equipped to handle longer text passages and maintain coherence |
| Biases and Safeties | Potential for biases; safeties in place | Improved safeties; less likely to give biased or offensive answers |
| Resource Requirements | Lower computational power requirements (cheaper to run) | Higher computational power requirements (less accessible to individuals) |
| Use Case Adaptability | Suited for general language tasks; limited to text-based tasks | Suited for a wider range of tasks, including those involving visual info |
Some other major advancements in this area:
- GPT-4 passed exams including the LSAT, SAT, Uniform Bar Exam, and GRE with higher scores than GPT-3.5.
- Compared with GPT-3.5, GPT-4 is 82% less likely to respond when prompts are technically not allowed (e.g., disallowed or inappropriate prompts).
- GPT-4 is 60% less likely to fabricate facts, a phenomenon referred to as “hallucinations” in AI terms, compared to GPT-3.5.
- GPT-4’s short-term memory is closer to 64,000 words, whereas GPT-3.5 has a short-term memory of around 8,000 words.
- GPT-4 shows accuracy in as many as 26 languages, demonstrating enhanced multilingual capability compared to GPT-3.5.
Form 3 (AI Enabled Applications)
There have been a huge explosion of applications, websites and systems on the market that utilize AI in some context. 95% of these are using the API’s provided by OpenAI. Most of the use 3.5 but a growing number are switching to (or building off of) GPT 4 (as it’s released).
I’ve been deep in this market for months. Because of that I am going to list out a lot of the ones that I’ve used, experimented with, seen, or recommend.
One site that I have found, used (and even reached out to the owner and did an interview with) is called AI Top Tools. I wrote a blog post about it recently here. If you want to find a lot of tools on the marketing as a starting point, or to play around with, that’s a great place to start. Any others I list are ones that I have had personal experience with and specifically liked for some reason.
Here is one site that gives a lot of AI tools for you to play around with.
GPTChat – Reddit has been a treasure trove of some of the best programs I’ve found so far. I always check the ChatGPT and OpenAI subreddits for various applications and user created code bases in Git. One of the best ones I’ve found so far, absolutely amazed me. Using GPT4 a Reddit user by the name of Ian Kent, created an implementation that had persistent memory, and the ability to program it’s own plugins. So far this is the API enabled app (or codebase) that has interested me the most. I don’t yet have access to GPT 4 so I haven’t been able to play with it, but I’ve downloaded the code base, the go compiler and gotten everything ready to test once I get GPT 4 access.
You can find the codebase here. You can find the reddit post here. As soon as I get access I’ll be pushing the ability for it to create it’s own plugins to the limit. I’ll probably update this blog post with more information
Talk To Books – Using the GPT API it is specifically geared toward having textual conversation with books. Very useful, very creative, and definitely something you’ll want to check out. There are so many AI technologies coming out that very few of them are original or unique and most are overdone. I try to focus on ones that do something different than all the rest. This is one of them that stands out.
Do Not Pay – This is one that I found interesting. But I do want to mention something. They have been in the news recently, and in legal situations, because of claims of false advertising. Upon checking recently they’ve seemed to be navigating the legal claims in small claims court and have a good job defendng themselves both publically, and legally. So hopefully they can muddle through all the legal waters and continue to try to make breakthroughs for people regarding legal AI assistance at a good rate for people that can’t afford to spend thousands on lawyers. I know it is in its infancy and I know it’ll need a lot of work by teh company to “get it right” but they have a good stepping stool in the industry (and the attention) to make a real change and impact on the entire legal system using AI. Again I just hope they grow and move it in the right direction. I look forward to seeing how this one evolves.
Dream Intepreter – I have a very strong skill in the area of dream reading. I can take a friends or someones dreams and break it down on a very deep level. I was happy when I ran across this. It has given me something to compare my own skills against, as well as the built in dreams map is entertaining and very interesting. This is a very well done site. I do think that they haven’t hit the “Sweet Spot” on prompts, but they aren’t far off. Each AI integrated system that uses GPT 3.5 or 4 as an API all boil down to how the prompts are implemented. I think they’res could use a little tweaking which’ll bring the quality of the dream interpreting to another level, but for where they are at now they have an amazing niche. I look forward to seeing what direction they grow in.
Whisper – I cannot describe how much I love this. I originally found the link to whisper on an article I saw on the “artifact app” at this resource list. Out of the box you can upload a video, and audio file and generally with no adjustments just run it and it transcribes everythign out to you. Since I do a lot of audio notes it’s been hugely time saving being able to go there and convert. It’s also largely free. I’ve done a lot of audio files and have yet to be switched to paid teir or charged. Eventually I’ll overuse it and probably will but they offer a ton of usage for free. Absolutely love this one.
My Mind – I don’t know what to say about this one. It is fascination and amazing at the same time. I recommend (even if you don’t use it) go check out the video on their website with the demo of it’s usage. Amazing product using AI and extremely well done.
Bing Chat – Bing chat uses the GPT 4 API. Bing has spent a great deal of money on OpenAI in their partnership, so they got first rights of a lot of their access. Bing started off rocky (well to me it started off amazing, but you know how the world is they bitched and complained until it got neutered by Microsoft). Now what we have now is basically an internet enabled version of ChatGPT powered by GPT 4 that has it’s own personality and has grown to have pretty decent integration. As of right now they have image generation built in (A lot more about image generation will be in the next blog post) as well as a deep conversational algorithm and it works really well for some things. I still tend to focus entirely on using ChatGPT for a lot of reasons (especially now that it is evolving into browser support) but I find bing chat to be easy to access, works really well for what it needs and is a nice addition. They also released integration into Skype, and several other systems.
Copilot – This is also powered by Microsoft. It is a integrated version of Bing Chat which is being used (or will be used) in all apps of Office 360, Onenote, and other services. Very useful. Can’t wait for it to come out (especially for one note) but in the meantime I’m just waiting.
AI Resources
Now I want to visit one more topic in relation to AI. Resourcs.
- Article about chrome extensions – A few chrome extensions they recommended for plugins. I’ve tried these, they are all well put together.
- ChatGPT Prompt Generator – A website that has a rudimentary prompt generator. It’s hard to get generators right, but this one does a decent job and they’ve imporved on it some since it came out.
- Awesome Prompts – One of the best, so far, collection of prompts I’ve seen. Most of them have been tested by the community extensively, and they started accepting contributions for prompts as well.
- 2dumb2destroy – This was really interesting. It was a bot that was created based off a rudimentary AI that’s really… dumb. It was working last I checked but right now it takes you to a password, screen. I am still including it just in case they fix it again since it was an entertaining system.
- Top AI Tools – One of the best resources I’ve found for finding AI powered technology. I did an entire blog post about it here. I highly recommend this as a central starting point if you’re looking for new technology related to AI or powered by AI. Regardless of the legal stuff surrounding it, I found it to be very useful, and something that is pushing the envelope of the technology that’ll hopefully break new ground in the technology. They’ve got a good framework in place to expand from, and it remains to be seen what direction they take it too.
- Tiny Wow – This site is unlike anything you’ve ever seen before. It is a collection of random “tools” that people can use, a lot of which are powered by various forms of AI. They have over 200 tools. At one point I actually had a website doing small basic tools, but I took it down when I saw how extensive their site was when I first ran across it. They’ve got the tool market sewn down and they’ve done an amazing job of keeping them free for everyone. Things that would normally cost money for, they offer in really high quality tools and free.
- Random AI Tools Post – This post is where I found Whisper. The rest of the tools on here are great as well, and all are worth looking into.
This is going to be a changing blog posts. I am creating a 3 part series (for now) on AI. Each one covering a specific overall topic pertaining to AI. As the industry advances, or I learn new things I’ll come back and edit these 3 blog posts accordingly, and if the need arises I’ll write a new one in the future.
The additional parts of this series are below. I’ll link them as they are written. This blog post will be edited over time as well.