Artificial Intelligence (AI) is more than just a technological marvel; it's a powerful tool that has reshaped the fabric of our digital lives. As I've traversed through this landscape, I've experienced firsthand the nuances that set each AI model apart.
ChatGPT: The Creative
OpenAI's ChatGPT can truly be a creative maverick. It weaves responses with an uncanny charm that makes it seem almost sentient. But beware, this AI is also a confident liar, cooking up responses that may not always be rooted in reality. While this flaw might not be a deal-breaker, it certainly highlights the importance of discernment. It's vital to have a good grasp of the subject matter you're querying so you can separate facts from fiction. This fundamental flaw is why ChatGPT will never replace a real expert.
ChatGPT is a great pattern matcher, and I have found that it excels at transforming information from one format into another. I commonly use it to transform raw information into haikus. Or I use it to give feedback on short stories I write in my free time. Or I'll feed it a raw brain dump of tribal knowledge I have about parts of idem along with pages of documentation and tell it to turn it into a professional blog post. 😏
Bing's AI: The Tutor
Created by Microsoft, one of the largest funders of OpenAI, and based on GPT-4, Bing's AI breathed life into the search engine. Remember when Microsoft was practically bribing people to use Bing? Well, with the advent of Bing's AI, the engine is finally earning its keep. The AI scours the internet for the latest information, presenting responses meticulously woven together from diverse sources. It provides accurate, reliable, sourced information. It's like a digital scholar at your disposal, and I find myself leaning on it as a tutor while studying for my Master's degree.
Bard: The App Enhancer
Compared to its AI siblings, Google's Bard feels a step behind in terms of fluidity and creativity. And, surprisingly, it doesn't seem to tap into Google's search prowess for up-to-date information. But don't write it off just yet. Bard is not based on GPT, Google developed their own LLM Called LaMDA. Give Bard the clues to a scavenger hunt and watch it shine, pinpointing real-world locations from your hints. Plus, its seamless integration with Google Suite apps gives it an edge in streamlining your email, calendar, and other productivity services.
LLaMa: The Off-roader
Developed by the creators of Facebook, Meta, LLaMa is a unique offering in the AI world; capable of operating completely offline. There's something liberating about this model, perhaps because it offers a sense of privacy often lacking in other AI tools. An open-source wrapper named Alpacca.cpp lets you access LLaMa easily, offering a viable option for handling sensitive or classified data on your private server. The model leaked not long ago, but a courteous request to Meta will get you legitimate access quickly.
I foresee a future where LLaMa, or a similar offline model, could enable AI-as-a-service. It could be hosted on-prem and be trusted with classified or sensitive corporate information, proprietary code, and internal documentation. This model has the potential to really shine in corporate environments. Other AI like ChatGPT have leaked proprietary code, leading to Samsung and Apple getting hacked because of developers asking it questions about proprietary code and those prompts being used as training data. An offline model hosted on-prem and fed proprietary data could enable developers in a strict corporate environment to fully leverage AI without leaking data.
HuggingFace: The Polyglot
HuggingFace, an open-source AI model, is on a mission to make artificial intelligence accessible across all languages. Traditional models, like GPT-4 and AI platforms such as Bing's AI and ChatGPT, tend to overrepresent English in their input datasets. HuggingFace breaks this mold, striving to make AI equally accessible and usable in all spoken languages worldwide.
HuggingFace's strength lies in its potential for localization. This feature, particularly its superior translation capabilities, has led to its adoption in projects like idem-ai and pop-ml during our innovation sprint. With HuggingFace, we were able to tap into a vast multilingual database, thereby expanding the versatility of our projects.
In a world increasingly unified by technology, HuggingFace has an important role to play in diminishing language barriers, and it’s certainly a model to watch in the expanding AI landscape.
LangChain: The Glue
LangChain is a python libary that provides a seamless wrapper around ALL the AI mentioned above and many more. It also has detailed documentation that describes how to write your own personal assistant or make your own app using the various AI APIs (Say that ten times fast). It serves as a powerful resource for consuming AI on your own. Give it a try!
The Price of AI: It’s Not All Free Lunch
The power of AI doesn't come without a price. Yes, numerous mobile apps offer access to online-hosted AI models, but more often than not, they come with a subscription fee. Larger models, like GPT-4, demand significant resources, making them practically impossible to run on private servers. Every API request to an AI model brings along a considerable processing fee, translating to those pennies that quickly add up with each request. Hence the prevalence of subscription-based models or ad-laden free AI apps.
The catch with these 'free' AI applications, like ChatGPT and Bing's AI, is that they learn from you. Your prompts serve as fodder for their training data. As you interact with these models, they refine their responses, drawing from the wealth of information users provide. In essence, you contribute to its learning and growth, becoming an integral part of the product. It's a trade-off; while you enjoy the benefits of AI, the AI learns from you, gradually evolving with each interaction. When something is free, you are the product, and with AI, your interactions provide them with training data.
Companies such as Ray, known for training large AI models like GPT-4, feed on this data to enhance the AI. The prompts and interactions from 'free' AI applications become the training data for these AI models. Remember, when something is 'free,' it's often because you are the product. Your interactions, prompts, and data are invaluable in training and refining these AI models.
The essence of AI pricing is understanding this dynamic. Paid AI apps offset the high costs of API requests through subscriptions or ad revenue. On the other hand, 'free' AI apps leverage user interactions as valuable training data, using this to enhance the AI's capabilities. As AI users, being aware of this symbiotic relationship is essential to navigate the fascinating yet complex landscape of artificial intelligence.
The Road Ahead: Navigating Future AI Challenges
As we look to the future, the road ahead for AI presents a mixture of promise and caution. Currently, we are in the golden age of AI. AI models like ChatGPT, Bing's AI, and HuggingFace are showcasing what's truly possible with this technology, from enhancing our online experiences to making our work and personal lives more efficient.
However, there's a potential downside that looms in the near future. Imagine a scenario where AI, in its effort to improve, could become bogged down by "SEO-like" responses. You ask ChatGPT to write you a short story, but due to corporate interference in the training data, stories about trucks are always centered around a "Ford truck". This deliberate poisoning of training data, akin to aggressive marketing tactics, has the potential to turn the AI landscape into an ad-filled dump, taking away from the rich experience that AI has the potential to offer.
This foreseeable future could result in a few years where AI seems to be moving backward, becoming less user-focused and more commercial. This is a rather bleak outlook indeed.
However, I firmly believe this potential slump will not be the end of AI's progress. Competition, innovation, and the relentless pursuit of improvement will push AI out of this rut. I predict an unexpected resurgence, a sudden leap in the capabilities and potential of AI, akin to a technological renaissance.
Scientists have also been able to fully map the brain of a fruitfly on a computer. This has the potential to massively increase the capabilities of AI. Fruitflies have threat response, generational learning, reward systems, and so many other organic responses that would be interesting to simulate inside an LLM to make it more of an AGI. Human brains have trillions upon trillions MORE neurons and connections between them than a fruitfly -- perhaps The Singularity and a world where complete digital simulation of a human mind will become more than fiction. Side note, I'm really sad that the show Pantheon got canceled, as it explored the implications of that future really well. There are also non-invasive Brain Interface Devices that could augment human cognition with AI. It's exciting to think about all the possibilities of the future of AI. So, as we stand on the cusp of this golden age of AI, let's buckle up for an exciting, unpredictable journey into the future of artificial intelligence. It's a ride you don't want to miss.