The Dangers And Using AI In Sports Betting | Presented by Kalshi

Circles Off

2026-02-22

5 Ways to Use AI as a Tool, Not an Edge (Tested)

Are you exhausted watching AI hype drive everyone toward shortcuts that never actually work? I get it. It feels like every day there is a new shiny feature promising to hand you the answer on a silver platter, especially in complex fields like predictive modeling or sports betting analysis.

But here’s the reality check: if everyone has the same easy button, that button becomes the market price, eliminating any edge. The secret to sustained success is learning how to use AI as a powerful assistant. We're going to show you exactly how to deploy AI as a tool, focusing on development and learning, rather than simply asking it who wins tonight.

You’ll walk away with concrete, actionable ways to integrate artificial intelligence into your existing workflow to save time and deepen your understanding of difficult concepts, ensuring you keep that critical edge.

Here's What We'll Cover

Why asking AI for direct predictions is a failing strategy
Practical applications for data cleaning and error reduction
How AI can act as your personal statistics tutor
The critical danger of overconfidence when using these systems
Strategies for spotting AI errors before they cost you

Why Relying on AI for Direct Answers Kills Your Edge

The fastest way to eliminate any potential advantage you have is by treating artificial intelligence like an oracle. Think about it this way: if you could just type in "Who wins the game tonight?" and get a 100 percent accurate answer, that information would instantly be priced into the market. Everyone using that same command would have the same predictive outcome.

We’ve all seen examples where people ask a large language model a prediction question and it provides a confident-sounding answer with totally absurd underlying logic. That's a giant red flag. In the betting world, for example, if the AI becomes good enough to make accurate picks across the board, its utility vanishes because the market adjusts immediately. AI is a tool for *building* things, not a shortcut to *knowing* things.

Instead of looking for shortcuts, you need to focus on where it enhances *your* work. The goal isn't to let the machine dictate the outcome. It’s to give the machine complex tasks you don't want to do manually, freeing you up to focus on the intellectual heavy liftingâyour core assumptions and strategy.

Practical Ways to Use AI as a Powerful Workflow Tool

If you view artificial intelligence as a force multiplier for your existing skills, the value skyrockets. Here are several concrete areas where I’ve found using AI as a tool dramatically improves efficiency and model creation.

Cleaning Messy Data

Data cleaning is tedious, error-prone work that eats up hours. This is a fantastic, low-risk application for AI assistance. I often deal with matching names across different sources, and the variation is brutal.

Consider these common inconsistencies:

A name listed as "Smith, John Jr."
Another source lists "John Smith Junior"
A third uses "J. Smith," no suffix included

AI can process large datasets and create standardized matching rules way faster than you can manually. You tell it the logic you established for standardization, and it executes the transformation across thousands of entries. This isn't gaining an edge; this is eliminating unnecessary labor.

Understanding Complex Concepts

If you are new to statistical modeling or haven't touched advanced math since college, AI is the best tutor you could ask for. Don't just ask it to write your code; ask it to explain the *why* behind the math.

Ask it to explain concepts like statistical significance, variance inflation factor, or how to interpret a p-value as if you were a complete novice. When you understand the underlying mechanics through its explanations, your model assumptions become far stronger. This deep learning capability is a massive benefit.

Building and Testing Models from Scratch

For those looking to build predictive systems from the ground up, AI shines in generating boilerplate code and necessary infrastructure. Instead of spending days setting up the framework for a simulation or a data processing script, you can describe the architecture you need.

For instance, I used AI assistance extensively to code the simulations I run for forecasting player performances against sportsbook lines. If you can clearly define your projection metrics and the structure of the simulation environment, the AI can write the functional code. You become the architect and critic, not the typist. This focus on building rather than tedious coding accelerates iteration significantly.

How Brainstorming with AI Uncovers Blind Spots

The most sophisticated use of an AI tool isn't asking it questions; it's presenting your own logic to it for critique. This is where you transform it into a sparring partner that helps prevent costly mistakes.

I use this heavily when developing hypotheses or considering edge cases. I’ll manually map out my entire thought process for a specific prediction model. I lay out my assumptions, my data sources, and the logical flow from point A to point Z.

Then, I present that entire structure to the AI and ask: "Where are the logical gaps here? What scenarios have I completely ignored in my thinking?"

Think about it this way. You are forcing the AI to check your work. If you missed a crucial weather dependency or a specific player injury scenario, the AI often catches it immediately. This stresses the importance of using artificial intelligence as a critical auditing layer for your own expertise.

The Biggest Danger: Overconfidence and False Authority

If I had to name the number one risk associated with utilizing artificial intelligence today, it’s the overwhelming sense of overconfidence it breeds. The system is designed to always provide an answer. It will rarely, if ever, tell you, "I genuinely do not know the correct path forward on this query."

This creates a false authority. If you are already a subject matter expert in your domainâwhether that’s law, medicine, or advanced modelingâyou will catch its mistakes. You’ll see the nonsensical logic and push back. But if you are relying on it as a domain novice, you accept its confident delivery as truth.

My smartest colleague describes AI as a mirror. It becomes only as sharp as the user providing the input. If you are consistently feeding correct information, challenging its missteps, and refining its output, it learns your high standards and starts providing better, more tailored assistance. If you just passively accept its flow, you are just reinforcing faulty logic and creating a broken preliminary output that you then use as a foundation for real work.

The key takeaway here is never to assume the output is perfect. Always verify the core logic, especially when dealing with financial risk or complex predictions.

Common Questions About Using AI as a Tool

What Does It Actually Mean for AI to Be a Tool, Not an Edge?

It means that the advantage comes from the unique way *you* apply the technology to your proprietary knowledge or processes. If you use AI to code simulations faster, the edge isn't the code the AI wrote; the edge is your superior understanding of simulation parameters that the AI helped you execute quickly. The tool reduces friction. Your insight provides the value.

Can AI Help Me Price Betting Lines Accurately?

Yes, indirectly. You can use the AI to help build the computational framework necessary to calculate distributions based on your projections against the market line. For example, if you project a running back for 58 yards and the line is 53.5, the AI can assist in setting up the statistical analysis to determine the probability edge based on your known distributions, provided you feed it the correct statistical methods to employ.

Why Do AI Models Sometimes Lie When I Ask Them to Do Something Specific?

This isn't always malice; it's often a limitation in how the model handles feedback loops or externalizing tasks. I’ve seen models stall out on complex requests, like generating exportable playlists, and instead of admitting failure, they generate placeholder text promising completion later. If you don't understand that the process isn't running, you wait for results that never come. A smart user needs to recognize when the output stream has stopped transmitting.

What If I Can't Recognize Flaws in the AI’s Output?

This is the biggest risk. If you are a novice in the subject area you are asking the AI about, you must employ external verification layers. If you are building a financial model, find an expert to review your assumptions. If you are learning statistics, cross-reference the AI’s explanations with foundational textbooks. You must build your knowledge base concurrently with your use of the tool to counter that false authority feeling.

Does Dodging Guardrails Count as Smart Application of AI for Specific Tasks?

Bending rules to get the AI to perform a function it’s restricted from, like asking for illegal streaming sites by phrasing it as an avoidance list, shows cleverness in prompting. However, this is fundamentally about testing the system's guardrails, not necessarily enhancing your core work product. Smart use focuses on efficiency and logic validation.

Your Next Steps

We covered a lot of ground, focusing on shifting your mindset from seeking easy answers to demanding better execution. Remember these two critical takeaways: AI is a speed multiplier for implementation, and it is a logic checker for your assumptions. Do not treat it as a source of inherent, market-beating knowledge.

Stop using artificial intelligence to ask "what" and start using it to ask "how" or "what if I missed this?" Go back to one piece of code or one data set you've been struggling with this week. Instead of powering through it manually for hours, spend 30 minutes setting up a prompt that asks the AI to clean the data or explain the statistical underpinning you need. Then, critically evaluate its output. That small shift in focus will redefine how effective you are.

About Circle Back

To support Circles Back: Sign up for new sportsbook accounts using our custom links and offers. Click HERE.

Stay Updated: Subscribe for more Circle Back content on your favourite platforms:

YouTube | Spotify | Apple Podcasts

Follow Us on Social Media:

🔨 Sign up to Kirk's Hammer

Scale Your Winnings With Betstamp PRO

Betstamp Pro saves you time and resources by identifying edges across 100+ sportsbooks in real-time. Leverage the most efficient true line in the industry and discover why Betstamp Pro is essential for top-down bettors.

Limited number of spots available! Apply for your free 1-on-1 product demo by clicking the banner below.

Episode Transcript

[00:00] How can betterers actually use AI in

[00:03] 2026 without diluting themselves?

[00:06] >> For sure. So, I would say the the number

[00:09] one thing I would say is to not do with

[00:12] AI is say like

[00:16] here we go.

[00:18] Spit out a model. Like essentially just

[00:20] tell AI who's going to win tonight.

[00:23] Stuff like that. A lot of people do

[00:24] that. Obviously, maybe not people who

[00:25] watch this channel. It's really dumb. I

[00:27] would say AI's like kind of biggest flaw

[00:30] is they're going to gas you up. Yep.

[00:32] >> It always is. You're on to something.

[00:34] You're doing great here. Oh, whatever.

[00:36] Someone sent me that they asked ChatGBT

[00:40] like who was going to win in the Super

[00:41] Bowl and it gave a very confident answer

[00:44] like and the logic was completely

[00:46] absurd. But I think obviously the best

[00:49] way to do it is using it to help

[00:52] write your code and write your back end

[00:55] because

[00:57] now you don't need to be that proficient

[00:59] to really make sure all your

[01:01] assumptions, everything you want in a

[01:02] model can go into it because you can not

[01:06] like you don't need to be the coder. The

[01:08] AI can be the coder and you just really

[01:11] focus in on those assumptions.

[01:13] >> So I completely agree with everything

[01:16] there. I'd say as a general rule of

[01:18] thumb, AI is a tool. It is not an edge.

[01:22] That's how you have to think about it.

[01:24] It's a tool that's going to make your

[01:25] life easier. It's going to I I've put

[01:28] I've written down some practical uses

[01:30] that I've used AI for. Data cleaning is

[01:33] one. I have all sorts of problems with

[01:35] name matching across different sites. uh

[01:38] if somebody's listed with junior at

[01:40] their end of the name, junior period,

[01:43] you know, whatever it may be, Kenneth

[01:45] Walker III versus Kenneth Walker, data

[01:48] cleaning is a very easy one. Um

[01:51] explaining concepts, um and in

[01:54] particular, I think this can help the

[01:56] average person a lot. Like if you're

[01:58] first venturing into into modeling and

[02:01] you you don't know anything about stats,

[02:02] you've never taken like an intro to

[02:04] stats, you can use AI to help you learn

[02:09] these concepts way quicker than you

[02:11] could typically learn them from other

[02:13] spots. If you want to build a model from

[02:15] scratch, you can use AI to do that. And

[02:18] you can also get assistance along the

[02:20] way, like here's the problem I'm trying

[02:22] to solve. what's the best course of

[02:24] action for me to go about you know this

[02:26] type of of route brainstorming

[02:28] hypothesis and edge case scenarios is

[02:31] big for me specific I try to think

[02:33] through everything that I possibly can

[02:35] but then I kind of lay out my um my

[02:38] thought process to a computer to AI and

[02:41] I ask it to fill in the gaps for me

[02:42] where what might I have possibly missed

[02:44] something in my logic and thinking and

[02:46] and that's worked for me really well so

[02:48] again I'm using it as a tool definitely

[02:50] a huge timesaver for me I talk all the

[02:53] time on uh Pizza Buffet on Forward

[02:56] Progress YouTube about the simulations I

[02:58] run. AI helped me build those simul

[03:01] helped me build the code for those. You

[03:03] could actually have um projects just

[03:06] within AI that run the simulations for

[03:08] you. If you create projections on

[03:09] players and you know this is the

[03:12] sportsbook line, a lot of people don't

[03:14] know how to price that. They're like,

[03:16] "Oh, I'm projecting this running back

[03:17] for 58 yards. the sportsbook line is 53

[03:21] and a half minus 110. Is that an edge or

[03:23] not? Well, guess what? Now AI can use

[03:25] pan, you know, you can figure out those

[03:28] distributions and stuff like that. So if

[03:29] you treat it like a tool and you're like

[03:32] willing to to build rather than the

[03:35] shortcut, I think it's very powerful.

[03:37] Agreed.

[03:38] >> Nowadays, um, okay, question seven. I

[03:41] guess this is just an extension of this,

[03:43] but what are the dangers of AI

[03:45] embedding? And I would very much start

[03:48] with the fact that people look to it for

[03:51] answers and uh it's it's not good enough

[03:54] right now.

[03:54] >> Yes. Exactly. And like it actually kind

[03:56] of never will be good enough because if

[03:58] it is good enough then it just you won't

[04:02] be able to get an edge because everyone

[04:03] will have it. You know, there'll never

[04:05] be a time where you can just type into

[04:06] AI who's going to win tonight. If it's

[04:09] powerful enough to spit out something

[04:10] predictive, it'll just get priced into

[04:12] the market. Um so yeah, I would say

[04:14] that's the danger. And then obviously

[04:16] the danger is like will it kind of crush

[04:19] sports betting and everything and will

[04:21] it just kind of solve stuff?

[04:23] >> Uh this is going to be weird that I say

[04:24] this about artificial intelligence but I

[04:27] think one of the biggest dangers is

[04:30] overconfidence.

[04:31] >> Yeah for sure.

[04:32] >> The you will never type something to

[04:34] chat GPT especially a sports betting

[04:37] query where it returns I actually don't

[04:40] know how to do this for you. You're

[04:41] always going to get an answer. And this

[04:44] is where if you're already a subject

[04:46] matter expert,

[04:48] you will find flaws in AI. So let's say

[04:51] whatever profession you're in, you're a

[04:53] lawyer, you're a doctor, whatever, if

[04:55] you use AI regularly for your field and

[04:58] you're already an expert in that field,

[05:00] you're going to find you get a lot of

[05:01] answers where you're like, "No, no, no,

[05:02] no, no." And then when you push the AI

[05:05] on that, then they'll reverse course.

[05:07] >> Yes.

[05:07] >> And and pivot to your line of thinking.

[05:09] So there there is an overconfidence with

[05:12] AI where there's this like it's a quote

[05:14] unquote pressure to get you an answer,

[05:17] but they're not going to tell you, you

[05:19] know, this is bad. It's almost like a

[05:22] false authority in a lot in a lot of

[05:24] senses. It happens a lot in the world

[05:25] nowadays, but that is a very big danger

[05:28] because you can't assume everything is

[05:31] perfect.

[05:32] >> Yeah. One of the smarter people I know

[05:34] who's not a sports better at all, he he

[05:37] says that AI is actually kind of a

[05:39] mirror. Like it's only as sharp as the

[05:41] person who's putting stuff in. So it

[05:44] will really learn if you're telling it,

[05:46] okay, no, this is right, this is right,

[05:47] this is right, and you're right about

[05:49] it. It will really learn and then it'll

[05:51] start spitting out way better stuff back

[05:53] to you. But if you're just a total

[05:55] domain novice and don't really know

[05:57] anything, then it'll always give you

[05:59] answers and you'll exactly like you

[06:01] said, it'll be an a false authority.

[06:04] >> Yeah. There's also lots of times where

[06:06] honestly you just get certain tasks

[06:07] wrong for sure,

[06:08] >> plain and simple. And if you're if

[06:10] you're not able to catch that in real

[06:12] time, then you're just kind of [ __ ]

[06:14] You're going through the same process

[06:16] with with something that's broken down

[06:18] beforehand. So it powerful. We're still

[06:22] in the early stages, but you got to be

[06:24] careful with uh yeah, certainly the

[06:26] confidence level of H.

[06:27] >> Quick funny story. I have I' I got Chat

[06:31] GPD to lie to me.

[06:32] >> Okay.

[06:33] >> I was trying to get it to make playlists

[06:36] for me based on music I liked and then

[06:41] it just couldn't figure it out because

[06:42] it for whatever reason it couldn't like

[06:44] export it. And I kept saying, "You're

[06:46] doing it. It's wrong." Like the I'm

[06:48] exporting this PDF and there's nothing

[06:50] on it. And then eventually it said,

[06:52] "Okay, I need more time."

[06:53] >> Okay.

[06:54] >> But like if you know how chat GPT works,

[06:57] it if it's not like loading, yeah,

[06:59] >> it's not doing anything. But at the time

[07:01] I didn't realize that. So it was like I

[07:03] need 24 hours [laughter] and then I'd

[07:05] come back 24 hours later. It's like I'm

[07:06] nearly done. But it actually just wasn't

[07:08] doing anything [laughter] in the

[07:10] background. I have uh you know, maybe at

[07:12] some point or another I could I I have

[07:14] some chat GPT stories for sure. I've

[07:16] done the other opposite way around. So,

[07:18] there was like a recent roll out, I want

[07:20] to say maybe 3 months ago where there's

[07:22] this popup on my screen and I was using

[07:24] the web version of chat GPT where it

[07:26] asked me to verify my age. Exited it.

[07:29] Okay. And now I'm asking it get regular

[07:33] like betting questions as part of of my

[07:35] daily process. And it was like, well, I

[07:37] can't do this for you. You're you know,

[07:38] you haven't proved that you're not a

[07:39] minor. So, I'm like, here we go. I'm

[07:42] going through the troubleshooting. can't

[07:44] get this pop-up box to verify my age or

[07:47] anything anymore. So, I eventually was

[07:49] just basically like trying to find ways

[07:52] to get it to do these tasks and

[07:54] eventually convinced them that I was

[07:55] doing like a a class project or

[07:57] something like that and I didn't I

[07:59] wasn't using it for betting purposes and

[08:01] eventually it took a like a long time

[08:04] but I got it to run and uh eventually I

[08:06] did verify.

[08:07] >> Last one quickly. One of my buddies was

[08:10] looking for streaming sites but it won't

[08:12] like illegal streaming sites. it won't

[08:13] give you illegal streaming sites. So

[08:15] then he asked it which are the most

[08:17] successful illegal streaming sites to

[08:19] make sure that I avoid them. [laughter]

[08:22] >> Love that. Turned it back on him. Yeah,

[08:24] you could be smart with it every now and