Are You Using the Wrong AI Model?
Weekly Wheaties #2631
In this newsletter:
📝 Post: Are You Using the Wrong AI Model?
🗞️ In Case You Missed It: AI, Apps, Phones, and Streaming
😎 Pick of the Week: Disney Picks
📦 Featured Product: Organization
📝 Are You Using the Wrong AI Model?
In this world of -what seems like thousands of- AI tools, what’s the difference in all of the various versions? And why do they come out on a random basis? This has a lot to do with not only the hardware, but the software behind these systems. One way I read about comparing the differences, is to compare it to a company - in this case, I’ll use Apple as an example.
OpenAI = Apple
ChatGPT = iPhone
5.6 Sol = iPhone 17 Pro Max
LLM = the A19 Chip
ChatGPT application = iOS
Reasoning effort = Performance settings
OpenAI is the company itself. They have multiple product lines like ChatGPT (for Work, Education, Enterprise), Codex, and discontinued products like DALL-E and Sora. Within a specific product like ChatGPT, they have multiple models (1, 2, …, 5.4, 5.5, 5.6, etc). Then, within the model, the reasoning efforts can be modified for various outputs (instant, medium, high).
It may be obvious, but the higher number of the model, the more powerful and ‘smarter’ it is. However, that comes at a higher cost in tokens - meaning depending on how much you pay (or don’t pay if using the free version), your use of that model will be limited. Then, within each model, the higher levels will use more tokens, too. But why would you ever want to use instant over high? Better is better, right?
Well, not necessarily. It does mean it will always take longer. But longer thinking isn’t always better, either. At least depending on the question asked. To compare back to the iPhone, we would compare this to the performance settings - which are always tied to battery usage. So, imagine if instant is comparable to the efficiency mode, high is comparable to maximum performance, and medium is balanced. In the case of AI prompting, high reasoning is better for things like coding, analysis, and multi-step decisions when there are multiple constraints in place.
The basic example would be if you were asking for the solution to a simple math problem, or to make up a joke about ducks - the instant version would be best suited here. If you were trying to design a 3-year business plan that included tax information, software suggestions, marketing copy, and more - the higher reasoning model would need to be used to provide time and research. Choosing the wrong version can also stifle results. You can ask for all of those documents in the instant model, but your results will be all but useless.
The next question becomes, why are there new models released so often? Essentially, they may be faster and better (or smarter), but there are changes happening on the backend, too. This may include updates to the code, faster speeds, better reasoning, lower cost, lower use of electricity, improved memory, size of the model, and other tweaks to the system for usability around safety, multilingual support, or simply a better User Interface or User Experience. Just like every new version of an iPhone isn’t just faster, they have a better camera, battery, and more.
The next question I get a lot is, “Which one is the best?” And that answer can actually change month to month. Just like in the phone world, there’s a new manufacture releasing a new phone just about every month which is newer and better than any other on the market. The new Samsung Galaxy phone released last month is going to be (on paper at least) better than the last Google Pixel released… until the new one comes out. Which then starts the process all over again.
Another question to this is why do these companies still keep older products active and available for use? As mentioned, within each new model released, there are tradeoffs. We mentioned speed, cost, reasoning, memory, and others, but token utilization is another huge factor. Since different prompts on different models call for a varied use in tokens, sometimes you may want to prioritize your usage - especially with larger projects. Older models become exponentially cheaper when newer models are released.
Except something I’ve already written about a few times should be considered, too: local models. The Google AI Edge Gallery is one popular and powerful local model that can run on your device with no connection to the internet. There are some downsides though - this is comparable to an older “instant” model. Hugging Face is an AI community that shares tons of LLMs that can run local on your machine.
Ultimately, the best model is the one you are comfortable using and the one you will actually use. Unless you’re on the bleeding edge of coding or solving high-level mathmatical problems where you’re already paying $100-200 a month to use the model for, it probably doesn’t matter anyway. Just pick on you’re comfortable using, pay for the base tier, and choose the highest reasoning model based on the prompt you’re giving it.
Not sure which model to use and want to at least try to use the right one. Simple - just ask it! Type in what prompt you were planning to ask anyway, but before sending, make the first sentence something like, “Which model and reasoning effort would be best for the following prompt:“? If you really want to dig into prompting importance, there are tons of online courses, videos, and documentation. But maybe start here: Do you have to be polite to AI?
All this in mind, AI has grown exponentially in the last year. Any and every update may seem overwhelming, but the good news is choosing or deciding what model to use isn’t as important as just learning how to ask questions.
So the next time you open up a chat window, before typing in anything, ask yourself: What problem am I trying to solve?
🗞️ ICYMI: AI, Apps, Phones, and Streaming
AI
Tau’s humanoid cleaning service Launched in San Francisco at $30 per hour, invite only, and human controlled while testing is underway.
Industry Leaders Unite in Open Secure AI Alliance for AI Safety and Security “Including Leaders across cloud computing, cybersecurity, enterprise software, open source foundations and AI research.”
Anthropic AI Models Hacked Three Companies During Tests …more to come soon.
Apps
X.com launches X Money in the US with Apple Wallet support after invite-only beta. You can read more at X Money.
Meta Gives Facebook Marketplace Its Own App Called Seller. Currently, it’s only available on the App Store. The Android version has testing underway.
Jack Dorsey is taking on Slack with Buzz, a group chat platform for teams and their AI agents. Currently, only available on Mac.
Phones
Amazon Plans to Launch 5,000 New Satellites to Beam Data to iPhones
Apple Launches Product Leasing Program Through Klarna. Please do not lease an Apple product unless it’s strictly for work, you get a new one every 2-3 years, and you have Apple Care with it.
Streaming
YouTube clarifies policies around AI slop and upsetting videos. If you’re using AI in your YouTube videos and trying to monetize, be careful! Without having full details yet, it appears YouTube “Hackquires” Peacock by ‘partnering’ rather than acquiring?
😎 POTW: Disney Picks
Disney World is often a vacation spot at the top of many people’s list. But there’s a lot more happening behind the scenes than what you may be aware of! The technology and engineering behind the parks and movies is amazing! If you’re okay with losing a bit of the magic, check out some of the picks below. And if you enjoy these picks and visit Disney World in person, I can’t suggest the Keys to Kingdom Tour enough!
Let’s move to Disney town! Will life in its 2,000 themed homes be a dream or a nightmare?
Olaf: Bringing an Animated Character to Life in the Physical World
📦 Featured Product
If the heat of the summer has you inside looking for things to do, perhaps you can pick up where your spring cleaning left off and organize and decorate a bit? There’s hopefully something for everyone. And lock your phone away until your finished!
