The future of AI is Thought
“What a piece of work is man! How noble in reason, how infinite in faculty!” -- Shakespeare's Hamlet
In this edition:
My thoughts
Notable headlines
Learning, Tools, and Experiments
AI MISTAKES
My Thoughts
I find it fascinating how AI is transforming our work and creativity. Seeing how a simple prompt can lead to rapid and insightful responses is incredible, allowing me to extend ideas easily. However, not everyone finds AI so enchanting; we'll explore that further in this issue.
It’s still magical to type something in and receive a nearly instantaneous response that answers my questions or brings new discoveries right to my screen. Even more valuable is that I can keep going, keep iterating, and keep investigating and growing my knowledge at an incredible pace.
Shakespeare's sentiment was right. It would be wrong to forsake what we have been so blessed to be given. But not everyone wants to use AI because it helps learning.
What piece of work is a man, how noble in reason,
how infinite in faculties, in form and moving,
how express and admirable in action, how like an angel in apprehension,
how like a god!
For this reason, we should seek to make man better, not replace him. We should strive to improve our faculties rather than allow them to fade away with lack of use.
We have myriad examples of what happens when we disengage from the thought process and let the machine do all the work without proper human attention (here’s my favorite if you’re interested).
So, I encourage you not to allow AI to remove your skills, whatever they may be, but to use AI to sharpen them.
Write better prose
Create better software
Speak to your customers
Convey your value more clearly
This is only the beginning of how AI can help us learn and improve. The problem is, what if someone is just using AI to do all the work? It’s already impossible to discern what is written by a large language model like GPT-4 from what is written by a human. Worse yet, we often think that human writing is better than AI until we discover that AI wrote it; then, it’s not as good as a human can write.
In a recently published paper, GPT-4 is judged more human than humans in displaced and inverted Turing tests. This shows that humans and AI aren’t particularly good at discerning between AI and humans. Humans don’t even get a 50/50 chance; they’re just wrong most of the time. Even the AI does a terrible job of discernment.
So, instead of trying to compete against each other, let’s make AI the foundation of improvement. Make AI MISTAKES, so you can improve and understand what’s possible. That way, when the time comes, you’ll be better, as will your choices about when to use AI.
Notable Headlines
SearchGPT: A limited release of their new search experience.
Llama 3.1: Open models are making great strides. What’s new? Improved multilingual support, longer context length, and better reasoning abilities
Google Research gets silver in math: Google's DeepMind AI systems, AlphaProof and AlphaGeometry 2, achieved a silver medal level in the 2024 International Mathematical Olympiad
Grok-2 makes big gains: Elon Musk's xAI has significantly improved Grok-2 and Grok-2 mini's performance. Grok-2 now ranks on the Lmsys Chatbot Arena leaderboard
Apple’s model is open!?: Apple has released several open-source large language models (LLMs) called OpenELM, designed to run on devices rather than through cloud servers. These models, along with complete training frameworks and evaluation tools, are now available on the Hugging Face Hub.
GPT-4o Mini changes the pricing game: OpenAI has introduced GPT-4o mini, a more affordable version of its powerful GPT-4o model. At 15 cents per million input tokens and 60 cents per million output tokens, it's 60% cheaper than GPT-3.5 Turbo.
FLUX makes a splash: Flux.1, developed by Black Forest Labs, is a new AI image model that challenges industry leaders like Midjourney. With 12 billion parameters, it surpasses competitors in image quality and performance.
Learning, Tools, and Experiments
Fine-Tuning your Model (for free with OpenAI for the next month)!
OpenAI has announced free fine-tuning for GPT-4o mini through September 23, 2024, allowing developers to customize AI models for their specific use cases.
Fine-tuning is particularly effective for tailoring AI models, but I can already hear you asking, “Why!?” Ok, sure, so here are some common use cases:
Let’s say you want to create an AI model that can write just like you. You can make a model that sounds like you in style, tone, and vocabulary. But there’s plenty more. A fine-tuned model can help you improve the reliability of your desired outputs. This allows for correcting failures to follow complex instructions or niche-specific situations that don’t work well with the large models alone.
The process involves preparing a dataset of example conversations, uploading it to OpenAI's API, and creating a fine-tuning job. You can use many sites and videos to learn how to do it, and I’m hoping to show off how I do it in a video, so watch the channel for more details!
You may not want to try this, but now you know it’s possible—which also means anyone can do it! While this free option is available now, it might be the best time to explore it. Regardless, you can bet that there will be more opportunities in the future as the costs continue to decline and the technology improves.
GPT-4o mini is available for free fine-tuning until September 23, 2024, with each organization receiving 2M tokens per 24-hour period at no cost. This initiative allows developers to explore and leverage advanced AI capabilities without immediate financial constraints.
AI MISTAKES
AI Detection is incredibly difficult. This month, let’s be less concerned about whether or not humans produced work and more about how we’ve learned and improved from the process.
Recent research indicates that over 80% of workers use AI without telling their employers. In the classroom, where students must apply their intellect to their assignments, it’s not much of a stretch to imagine that they, too, are using AI to get their work done without anyone else knowing.
In the classroom, though, it’s about more than plagiarism. It’s about ensuring students get value from a hard-earned education. The question is, did they learn anything in the process? If no learning took place, was the exercise even worth it? I don’t know about you, but who wants to grade papers just to grade papers? Not me. As a grad student, I was a TA, and it’s not a fun activity.
For learning to be truly valuable, struggle must be at least part of the process. It’s not about roll calls, grading papers, or filling out scantrons (which I just learned ended last year). It’s about what students learn.
The problem is that shady software vendors out there tell you they can reliably detect AI writing, and it’s just not so. AI detection tools are often inaccurate and unreliable. They can produce both false positives (incorrectly flagging human-written text as AI-generated) and false negatives (failing to detect AI-generated content). This unreliability can lead to unfair accusations against innocent students or failure to identify actual instances of AI use. The real problem here is that relying on AI detectors can damage the trust between teachers and students. It creates an atmosphere of suspicion and surveillance, which can negatively impact the learning environment and student-teacher relationships.
The AI tech is rapidly evolving, so it’s also expected that while their detection capabilities will likely improve, they will struggle to keep up. This could lead to an endless cycle of trying to outsmart increasingly sophisticated AI systems, diverting focus from actual teaching and learning. It’s just not the direction we should go.
AI Dectors are more of a problem than a solution.
AI Detection Impacts more than Scores
AI Detection Doesn’t Solve Cheating
AI Detection is a Problem for Teachers
So what to do instead?
Since AI detectors offer little meaningful educational value regarding the use of AI, educators should consider the following approaches:
Foster Open Dialogue by engaging students in discussions about AI use, its implications, and ethical considerations.
Create an environment where students feel comfortable discussing their use of AI tools. Below, I’ve provided a list of resources to help with this.
Please provide a clear set of guidelines and expectations with practical ways to use the technology to help them improve what they turn in for grading without eroding the value of the process. This will help students, faculty, and staff as the entire community work effectively with this technology going forward.
Resources
There are also many resources available to help you navigate this emerging situation. While I’ve already created some helpful content, these resources align with your mission and problems. Please use these to better guide your next steps so we can skip over the bad MISTAKES and get to the heart of learning faster.
Khan Academy's AI for Education Course:
https://www.khanacademy.org/college-careers-more/ai-for-educationEthan Mollick's Resources:
https://interactive.wharton.upenn.edu/teaching-with-ai/Code.org's AI 101 for Teachers:
https://code.org/aiGoogle's "Generative AI for Educators":
https://grow.google/ai-for-educators/Microsoft's "Empower Educators to Explore the Potential of Artificial Intelligence":
https://learn.microsoft.com/en-us/training/courses/empower-educators-explore-potential-artificial-intelligenceWharton Interactive's "Teaching with AI" Resources:
https://interactive.wharton.upenn.edu/teaching-with-ai/EdTechTeacher's "AI Essentials for Educators":
https://edtechteacher.org/ai-course/
Practically every school district worldwide has to derive a policy on where to take things next, and in my research, I came across the Plano ISD page on AI (https://www.pisd.edu/ai). They’ve already collected some great resources as well.
As I close out this edition, let’s avoid the harmful AI MISTAKES, the disastrous, damaging mistakes that could leave long regrets in their wake. It’s essential to encourage healthy experimentation, growth, and curiosity, but do it in a way that improves the mind. Go, try new things, start getting your hands dirty, and discover how this technology can help us grow and improve.
I want to hear from you; I read all your messages, just hit reply, and it comes directly to me. I’m here to help—learning, teaching, consulting, and helping others to learn from MISTAKES with AI.