[00:00] What's going on everyone? Welcome back

[00:01] to Circles Off. I'm Rob Pazola. Joined

[00:03] to my right, Kirk Evans. How are things?

[00:06] >> Things are good.

[00:07] >> Things are good. We're doing something a

[00:08] little different for today's episode.

[00:10] So, few months back, we put out a call

[00:13] for a Q&A. And uh I told you guys like

[00:16] drop your questions down below, email

[00:18] us, whatever, DM on Twitter, whatever

[00:21] you got.

[00:23] I don't like to hold back. I'm a pretty

[00:24] honest guy. I'm going to be honest here.

[00:28] A lot of the questions were not great.

[00:31] Were not great. A lot of uh I'm not

[00:34] exaggerating. Who's the lock this week?

[00:36] Seriously. Uh one of them is was minus

[00:40] 250 cursed now. Uh a surprising number

[00:45] of people asking if their parlays were

[00:47] good.

[00:49] I get it. You know, maybe some people

[00:52] are new to finding this content. They

[00:54] don't know what we're about. We're not

[00:55] giving out any sort of daily pick.

[00:57] Instead of doing a surface level Q&A,

[01:01] uh, I wanted to do something that

[01:03] actually helps people who are serious

[01:05] about getting better. So, here's what I

[01:07] did. The little pivot moment. I took a

[01:11] year's worth of comments from our

[01:13] channel on every single video that we

[01:15] did. All the DMs that we got, all the

[01:18] questions that came into our circles off

[01:20] email, I fed them into AI to chat JPT to

[01:24] be specific. I preferred of the AI even

[01:26] though I use all of them. I told it a

[01:28] very simple question said generate the

[01:32] questions that would be worthwhile for

[01:33] our audience based off of their feedback

[01:36] on other pieces of content.

[01:38] And the list it gave back was what I

[01:41] sent to you, Kirk. It's honestly pretty

[01:42] good. I thought it was a great list of

[01:44] questions. That that was the only like I

[01:47] didn't know any of that backstory. I

[01:48] just saw the the list of questions and

[01:51] three minutes ago I said I think the

[01:53] questions are really good.

[01:54] >> Agreed. Spit out some good ones. So, uh

[01:56] today instead of answering what you

[01:58] asked,

[02:00] we're answering what AI asked, but based

[02:03] off of what you asked in some capacity.

[02:05] So, we got 10 questions today myself and

[02:07] Kirk are going to go through. Hopefully,

[02:08] you find some value in it. If you do,

[02:10] smash that like button down below. If

[02:12] you're not subbed here on circles off

[02:13] just yet, consider subbing. Goes a long

[02:17] way in the algorithm.

[02:20] Here we go.

[02:22] Question number one.

[02:24] At what point should someone accept that

[02:27] they are not a winning better?

[02:30] It's a great question.

[02:32] I think it's tough because obviously it

[02:35] kind of depends on what actually the

[02:38] person is and and how you know not to to

[02:42] be mean but like how kind of bright they

[02:44] are.

[02:44] >> Yep.

[02:45] >> But cuz like I can kind of you can

[02:48] always scale back what type of betting

[02:51] you're doing, right? Like if you're, you

[02:54] know, 500 bets in to originating

[02:58] >> and your XROI is in the red and your ROI

[03:02] is in the red, yeah, you should at the

[03:04] very least step back. I think like for a

[03:08] given strategy, if you're if you think

[03:10] you're going to win, if you have 50 bets

[03:13] and clearly are not beating the close or

[03:16] like if your X ROI is maybe negative 1%

[03:19] negative 1 and a half% then you're

[03:21] probably are on to something. But if

[03:23] it's just, you know, negative 5%, you're

[03:25] just essentially betting the best line

[03:28] on market, but it never moves with you.

[03:31] 50 to 100 bets I would say you step back

[03:34] and take a look at what you're doing but

[03:36] at the same time that is you know even

[03:39] if you're just not a good originator you

[03:41] can just go back and try you know being

[03:45] more of a line shopper trying to snipe

[03:48] blind so I don't know if there's really

[03:50] ever a point where you should just be

[03:52] like I cannot I am just not going to win

[03:55] betting but also obviously if it's just

[03:57] taking up so much time and it's not

[04:00] worth it that's kind of a different

[04:01] question, but I think at the easiest

[04:04] level of gambling, you generally most

[04:07] people could figure out some way to win

[04:09] a little bit if that's like really the

[04:11] goal, right?

[04:12] >> But obviously life factors play in a lot

[04:13] as well.

[04:14] >> Yeah. So like I don't think that there's

[04:15] such thing as someone who can never win.

[04:17] I agree with you. But there might be

[04:19] stuff that people are doing where if

[04:21] they continue doing that, they are not

[04:23] going to win. Uh I think first and

[04:26] foremost sample size is very important,

[04:29] right? If you place 50 bets, that's

[04:32] really not going to tell you anything.

[04:34] >> Well, I think 50 bets tells you a lot on

[04:38] how the market reacts to you.

[04:39] >> Sure.

[04:40] >> Sure.

[04:40] >> It won't 50 bets tells you nothing in

[04:42] terms of ROI. Sure.

[04:43] >> Like if I pulled a 50 bet sample of any

[04:45] bets you placed over time, I could glean

[04:48] some things probably by looking at the

[04:50] closing line value on that, but overall,

[04:53] the larger the sample, the more

[04:55] indicative it's going to be of your

[04:57] performance, right? So if you go

[04:59] through, you know, 50 bets versus 500,

[05:03] those are going to tell you very, very

[05:05] different things. The 500 sample in

[05:08] 99.9% of cases is going to be a better

[05:12] factor than 50. The only time that it

[05:15] wouldn't be is if that 500 stretched

[05:17] maybe like 20 years or something like

[05:18] that. You know what I'm getting at? So I

[05:20] think sample size is very important and

[05:21] people run into that a lot. If you've

[05:23] been gambling for many, many years and

[05:26] you're not winning, you're you're doing

[05:28] something that is wrong and you're going

[05:31] to have to change that. But generally

[05:34] speaking, I know that there's debates on

[05:37] closing line value and depending on the

[05:39] markets, it's still a very good

[05:40] indicator in most markets of where

[05:43] you're at right now. So just next to

[05:46] your actual ROI or your units or your

[05:48] win loss or whatever you're tracking,

[05:50] you should also be tracking whether or

[05:52] not you're beating the market and by how

[05:54] much. For sure. And I'm such a CLV

[05:57] believer and truther, not only because I

[05:59] think it's a really good indicator,

[06:03] but also because like you said, sample

[06:05] size is so important, but we're not

[06:07] talking about, you know, it we're not

[06:10] talking about 500 bets and 50 bets as if

[06:12] that's the same thing, right? If you bet

[06:14] 500 bets and you're a losing bet and it

[06:17] took you 500 to find out, you lose a lot

[06:19] more money than if you bet 50 bets and

[06:21] you find out you're uh a losing better

[06:24] and you'll have way more time to adjust

[06:26] after 50 bets. So that's why CLV is such

[06:29] a good tool for trying to understand if

[06:31] you're a good better and I think people

[06:33] conflate things of is CLV perfect versus

[06:37] is it valuable. So, in pretty much any

[06:40] market, unless you think you have a

[06:42] really specific niche edge that no one

[06:46] else knows about, fine. Maybe it won't

[06:48] be perfect. Your X ROI. And just to

[06:51] clarify here, X ROI would be where your

[06:54] bet closed versus the let's say whatever

[06:59] number you would want, the pin number or

[07:01] like a mix of the numbers of the market

[07:04] verse the number you got. So, you know,

[07:06] if you played uh 2.0 and the market

[07:10] closed at it was 55%, your XROI would be

[07:13] whatever that would be like plus a few

[07:15] uh what is that? 5%. No, I don't know. I

[07:18] don't know how to calculate that off the

[07:19] top of my head, but that's a general

[07:21] idea. Y

[07:22] >> So, you can get

[07:25] if you bet 50 times, unless it's so

[07:27] niche, even if you're a prop bet

[07:30] or betting uh straits, totals, whatever,

[07:33] fine. Maybe it won't be perfectly

[07:36] reflective, but it will go in your favor

[07:39] if you're going to win to some degree.

[07:41] Maybe your XR is a small negative, but

[07:44] the line will move. You're not going to

[07:45] be a winning better if every time you

[07:47] bet the line stays stagnant or goes

[07:49] against you, you're just not going to

[07:50] win that way. So even though I agree

[07:53] closing line value isn't perfect

[07:55] especially but for this question it is

[07:58] such a good indicator and really what

[08:01] should be your guiding light because of

[08:03] how much quicker it'll be because even

[08:05] 500 bets you to me there would be

[08:09] debatably I would need to see the math

[08:11] on it more uncertainty of your ROI after

[08:14] 500 bets verse your CLV after 50 bets

[08:18] like 50 bets is a lot for if you're

[08:21] going to close the uh just one practical

[08:23] thing I'll say and then we'll move on to

[08:24] question number two is um I also think

[08:27] that you know if you have an edge in

[08:30] sports betting you should be able to

[08:32] explain to other people what that edge

[08:34] is and I'm not saying actually publicize

[08:36] it to the world like that's not I'm

[08:37] saying give away your edge but if you

[08:39] can't actually explain what your edge is

[08:43] then you don't have one and if your edge

[08:45] is you know I spend three hours a day

[08:48] researching it's probably not an edge

[08:50] overall. So, I think that's just like

[08:53] where people kind of go wrong is like

[08:56] they think that they're handicapping

[08:58] stuff a lot of times and you know,

[09:01] you're you're in a competitive market

[09:03] and whatever you're betting, you got to

[09:05] be able to explain why what you're doing

[09:07] is going to be better than other people

[09:09] in market at the time that you're

[09:11] betting it is what I would get to. Uh,

[09:13] question number two. Uh, if you had to

[09:16] start from scratch with $5,000 today,

[09:20] what would you do?

[09:21] >> It's a great question. I think I would

[09:25] likely, it's hard. I'm not completely

[09:28] sold, but I do think very likely what I

[09:31] would do is either I would just become a

[09:34] a Kali Poly Market Grinder. Y

[09:37] >> unique markets,

[09:39] >> politics, all like the Spotify streams,

[09:42] YouTube. And what I would really do is

[09:45] build a database as big as possible.

[09:47] There is just nothing more beneficial

[09:50] than having baselines to bet. You know,

[09:54] what was this bet before? A good example

[09:56] is the Bad Bunny

[09:59] um YouTube. There's a a market h will

[10:03] how many views will his uh halftime show

[10:05] get in the first week? I can't bet that,

[10:08] right? because I don't know how many

[10:11] YouTube views Kendrick Lamar got or how

[10:14] many Kesha got or how many all the shows

[10:16] before got. So if you have that data,

[10:20] you can it is so much easier and you are

[10:23] ahead of 99.99%

[10:25] of people, right? So just collecting

[10:27] stuff. Another good example is like the

[10:29] three point contest. I have historicals

[10:31] on all the three-point contests going

[10:32] back for 20 years.

[10:35] >> And I like some of that was found by

[10:38] actually just watching the videos on

[10:41] YouTube. Now I can get an edge on the

[10:44] three-point shooting contest in two

[10:46] minutes because I have that. But if I

[10:49] didn't have that historical, you cannot

[10:50] get it because you don't have a

[10:51] baseline. So that that's what I would do

[10:53] a lot of grinding on. Or maybe futures

[10:57] markets, awards, stuff like that is I

[11:01] still think probably the softest market

[11:03] there is. And I think you can spend a

[11:05] lot of time on that. And it's not so

[11:07] math-based. It's a lot of It's a very

[11:09] kind of inviting market. But the thing

[11:12] about that is if you had 5,000 bucks,

[11:14] you're deploying a lot of money on

[11:16] futures, so it wouldn't turn quite as

[11:19] quick.

[11:20] >> This is a really tough one for me to

[11:21] answer. I I've never really thought

[11:23] about this question all that much if I'm

[11:25] if I'm being completely honest, but I I

[11:27] I prefer to think of it in like um sort

[11:29] of a different day, uh a different way

[11:31] of I guess what my regrets were when my

[11:34] bankroll was smaller. I always think

[11:37] back to regulation of uh sports betting

[11:39] in Ontario and if I could go back I

[11:42] would do things very very differently.

[11:44] So I think a lot of the advice that's

[11:45] often given to people who have low bank

[11:46] rolls is like grind the bonuses and you

[11:50] know really use that to get all the free

[11:51] money upfront and there's some value in

[11:54] that but it it depends what level of

[11:56] sports better you want to be and if you

[11:59] really want to grow it over the years. I

[12:01] think that the biggest mistake that I

[12:03] made was treating accounts like I was

[12:07] going to have infinite accounts, right?

[12:10] And not realizing that, okay, once I

[12:12] burn this book, it's going to be hard

[12:15] for me to get another one of that and so

[12:17] on and so forth. So, I I would honestly,

[12:22] you don't want to stretch yourself too

[12:23] thin with $5,000. you want to focus on a

[12:25] a couple of accounts, but if you really

[12:27] believe that you have some sort of

[12:28] long-term edge and you want to build it,

[12:30] I would look at trying to sustain those

[12:32] accounts for longer in the early going

[12:34] rather than completely nuking them in

[12:37] the early going. There's actually

[12:39] nothing wrong than with taking a lot of

[12:41] longot parlays in the early going of an

[12:43] account. Stuff that might still be plus

[12:45] EV that you know might lose, but if you

[12:49] win and you hit that, you'll be like,

[12:51] "This is very worthwhile for me on this

[12:53] account. I don't care if I get limited

[12:55] um and things like that. But you

[12:56] mentioned a good point about the

[12:57] prediction market stuff as well because

[12:58] we're entering like a new era here. For

[13:00] sure.

[13:00] >> For sure. And I thought that was a

[13:02] really good point of like if the

[13:04] question just completely depends on are

[13:07] you starting with $5,000 and want to

[13:09] turn that into $10,000 or $100,000 or a

[13:12] million dollar. Those the answer to all

[13:14] those questions would be very different.

[13:15] Oh,

[13:15] >> 100%. It would be I I I would have a

[13:18] different answer for what markets I

[13:19] would avoid. Uh what I'd want to

[13:22] specialize in, whether I would be a

[13:24] bottom up or top down better, how I'd

[13:27] track my results probably stays very

[13:28] similar across all of those no matter

[13:30] what. But um things I would absolutely

[13:33] not waste time on. Maybe that's a good

[13:35] idea for a standalone video at some

[13:37] point of just if I I'll I'll think that

[13:39] through a little bit more and get back

[13:40] to that. Of course, Kirk did mention

[13:42] Koshi. They are presenting sponsor here

[13:44] on circles off. If you do want to sign

[13:45] up for Kali, link is down in the

[13:47] description below. Um, we're using Kali

[13:49] more and more as are our peers nowadays

[13:52] and uh, it's pretty useful tool

[13:54] nowadays, just like a completely new way

[13:55] that you can go about attacking

[13:57] different markets. So, if you want to

[13:58] check it out, do so in the description

[14:00] down below. All right, number three, get

[14:03] into some modeling questions here. I

[14:05] think these are often asked. Uh, what's

[14:07] the biggest mistake people make building

[14:09] their first model?

[14:11] >> It's a great question. I don't know if I

[14:13] would necessarily say this is like a

[14:14] first model problem necessarily, but I

[14:16] talk about this a lot with my friends in

[14:18] betting. To me, the biggest mistake

[14:22] novice betterers make is they don't

[14:26] recognize how much they don't know. So,

[14:29] the example I I really like to give is

[14:31] halftime basketball betting. If you made

[14:34] an unbelievable halftime model that was

[14:37] very predictive and very, you know,

[14:40] you're you're a PhD mathematician and

[14:43] you do a really good job predicting, you

[14:46] know, how a player how a team's tempo

[14:49] goes, how their pace is in the first

[14:50] half versus second half, all these

[14:52] things, you still might get [ __ ] in

[14:54] the NBA halftime markets if you just

[14:56] don't realize that where the team is

[14:58] shooting, which direction is

[15:00] unbelievably important. So, I think when

[15:04] you're building really bottom up, like

[15:05] fresh bottom up, without really taking

[15:08] the market into account at all, there's

[15:10] just things, little intricacies that you

[15:13] just don't realize that can matter a

[15:15] ton. I would say not testing against

[15:20] market is one of the biggest mistakes I

[15:22] see people make because your goal is for

[15:26] if you're building a model for betting

[15:28] is to bet it. So typically, you know, I

[15:31] get people all the time are like, "Oh,

[15:33] my, you know, model can predict this. I

[15:35] can predict the winner of the game this

[15:36] percentage of time." It's like, well, is

[15:39] that good enough for you to actually win

[15:40] money? Because you're competing with

[15:41] other people in the space, right? You

[15:43] know, it's it's table stakes to to just

[15:46] build a model using data. Anyone can

[15:48] take, you know, the most important um

[15:50] data for a specific sport, spit out

[15:52] something. Nowadays with AI, you can get

[15:54] this done fairly quickly. But if you're

[15:57] not testing against market and

[15:59] specifically when you're going to be

[16:00] betting that stuff into market, you

[16:03] could be going in completely blind. I've

[16:05] seen people make that mistake many of

[16:06] times before. It's built into everything

[16:08] that I do from a a modeling perspective

[16:10] nowadays. Um, question number four. How

[16:14] do you know if your model actually has

[16:16] predictive power?

[16:18] >> Yeah. Well, I think you you kind of just

[16:20] perfectly explained it. either if you

[16:23] wanted to back test it, which I think is

[16:25] actually a much worse way of going of

[16:27] like back testing it against the close

[16:29] prior, but I think by far the best way

[16:31] is just paper betting or betting really

[16:34] small early, seeing where the market is

[16:36] going on it and if you have a really

[16:39] really big sample, seeing what your ROI

[16:41] is.

[16:42] >> Yeah. So, when Kirk mentions paper

[16:44] betting, he's talking about uh

[16:46] essentially fictionally placing the

[16:48] bets. So whenever you had a bet pop up

[16:50] in the model, rather than actually

[16:52] placing the bet, you would just log it.

[16:53] You track it for a certain period of

[16:55] time. Personally, I do like back testing

[16:57] myself, but you have to have the proper

[17:01] infrastructure. Some people don't

[17:03] necessarily do it properly. Like they

[17:05] back test against opening lines or

[17:06] whatever. That's not going to serve you

[17:08] any purpose. Um CLV tracking very

[17:12] important in this market as well. Um,

[17:14] >> yeah. I don't think back testing

[17:16] necessarily is bad, but I think testing

[17:19] against where the market goes is much

[17:21] more important than back testing.

[17:24] >> Yeah. So, you that's a good point.

[17:26] Sometimes people get carried away with

[17:27] their ROI, but

[17:30] >> betting is very different than just like

[17:33] tracking against what lines were at a

[17:36] time. Like the act of actually betting

[17:38] that into market. Sometimes lines

[17:40] disappear quickly. you know, there's all

[17:42] sorts of complications that arise when

[17:43] you're actually betting a game. So, I

[17:45] would agree with you.

[17:47] >> So, I think a good easy example here,

[17:49] let's say you built an NBA playoff model

[17:52] over, let's say, the last 15, 20 years.

[17:56] Your model picks up that zigzag is very

[18:00] important. Team loses hope uh game one,

[18:04] they're much more likely to win game

[18:05] two. Your model picks that up. You do

[18:07] that against 20 years of data. you crush

[18:11] in your back test because for let's say

[18:14] 10 years the the market didn't realize

[18:17] how how important that was but if

[18:21] obviously you could see this in a back

[18:23] test but you could miss it of that's

[18:26] already priced out into the market so

[18:28] you would have no future predictive

[18:30] value even though your model maybe would

[18:31] have been really good in 2004

[18:34] but if you tested that against CLV now

[18:38] you would see that actually your model

[18:39] has no predictive value. So that's why I

[18:41] think testing against where the market

[18:43] goes is much more valuable than testing

[18:45] prior.

[18:46] >> All right, the next one is a real

[18:48] question from a viewer. AI just said

[18:50] this was a real question. I'm going to

[18:52] use it. Okay. Beating the closing line

[18:55] by 1% over 2,000 bets, but I'm losing

[19:00] money. What do you do

[19:02] >> by 1%? Exactly.

[19:04] >> I'm beating This is the exact question.

[19:06] I'm beating the closing line by 1% over

[19:09] 2,000 bets but losing money. What do you

[19:12] do?

[19:14] >> Yeah. Well, beating the closing. So,

[19:16] when he says by 1%, he said he it kind

[19:19] of depends on what he means. Is he

[19:20] beating is his XR or Y 1%. Because if

[19:23] you're beating the closing line by 1%,

[19:26] you're probably not actually just

[19:27] beating the market. It could just be

[19:29] makes sense that he's losing. Oh,

[19:30] >> or it it actually could just be it's

[19:32] totally plausible that it could just be

[19:34] variance even on 2,000 bets. If your CLV

[19:36] is just 1%

[19:37] >> for sure, such a tight Yeah. If your if

[19:39] your XY is that tight anyway, you

[19:42] probably need to kind of go back. You're

[19:44] probably on to something. Especially if

[19:46] your XO is 1% rather than beating the

[19:48] closing line by 1% because

[19:50] >> to me, beating the closing line by 1%

[19:53] means that you're like shifting the odds

[19:57] from like small negative to just smaller

[19:59] negative. But if your if your expected

[20:01] win is 1%. It's just too tight of an

[20:04] edge to actually bet. Especially if you

[20:07] have 2,000 bets and can see that your

[20:09] XRY isn't you don't have like more

[20:12] expected win there. So I would say

[20:14] you're on to something. But just try

[20:17] going back into the lab and getting

[20:19] that, you know, 1% to, you know, two to

[20:23] 3% and then I feel like you could be

[20:25] even 2% is pretty goddamn tight. You

[20:28] definitely want to be three to four

[20:29] percent. Agreed.

[20:30] >> And and also if you're beating the

[20:32] closing line, you clearly have

[20:34] something. So it is definitely plausible

[20:38] that you could go back. Also, if it's

[20:39] 2,000 bets, I don't know over what time

[20:41] frame that is. Maybe be more selective

[20:43] and try finding the best bet that your

[20:45] model spitting out.

[20:46] >> Agreed. So that was one thing I was

[20:47] going to point out. if if a find out

[20:50] what's not getting closing line value

[20:52] and what's dragging that average

[20:53] percentage down and try to get rid of

[20:55] that completely is what I would say um

[20:58] to that for me the way that CLV is

[21:00] measured can differ as well. So I I

[21:03] don't know what's going on with that

[21:05] specifically. Um definitely other people

[21:07] disagree with me on this but this is

[21:09] just my own personal opinion. And it's

[21:10] something that I've I don't want to say

[21:11] I've always done, but definitely at

[21:12] least in the last 5 years. Um I value

[21:16] CLV, track it, but also try to come up

[21:18] with a secondary metric to determine

[21:20] whether or not my bet was if I got

[21:23] lucky, unlucky. Um so if I'm betting

[21:26] hockey, for example, uh based off of the

[21:29] shot quality data, I can kind of derive

[21:31] like an expected win percentage for each

[21:34] team. There are some public sites that

[21:35] do this not super accurately, but hey,

[21:39] you can use those as a gauge. Like if

[21:41] you're a hockey better, there's a site

[21:42] called Money Puck, for example, that'll

[21:43] give you uh what the deserve to win

[21:46] meter or whatever.

[21:47] >> Fine. Not the perfect way to do it for

[21:50] football. Um, I mean, we're going to get

[21:53] into some AI questions in a second, but

[21:55] I have my own expected points

[21:57] calculation for every single game based

[21:59] off of how the box score played out. But

[22:01] nowadays, you can just you can feed

[22:03] play-by-play data into

[22:06] any AI model and basically say based off

[22:09] of this play-by-play data, what should

[22:12] the expected final score have been in

[22:14] this game? Again, it's not a perfect

[22:16] measure, but let's say you do that with

[22:18] Grock, chat, GPT, and Gemini. You

[22:21] aggregate, you know, you you you

[22:23] calculate the average. You can get some

[22:25] sort of idea of what the score should

[22:27] have been, how likely you were to cover

[22:29] at at the number that you bet. So, I I

[22:31] think finding a secondary metric as well

[22:34] can often paint a little bit of a

[22:36] clearer picture because if it's a 1%

[22:38] CLV, but then you're looking at a

[22:40] secondary metric and you're like, well,

[22:42] you know what? like I actually am not

[22:43] deserving of winning a lot of these bets

[22:45] then,

[22:47] >> you know, it's

[22:49] it's not the greatest at that at that

[22:51] specific point. So that's kind of the

[22:53] hard truth, but that type of edge, 1%

[22:56] true edge is is razor thin. Yeah.

[22:58] >> So

[22:59] >> variance can absolutely bury that over

[23:02] 2,000 bets. Uh question number six, how

[23:05] can bettererss actually use AI in 2026

[23:10] without diluting themselves?

[23:12] >> For sure. So I would say the the number

[23:14] one thing I would say is to not do with

[23:17] AI is say like who spit out a model like

[23:23] essentially just tell AI who's going to

[23:26] win tonight. Stuff like that. A lot of

[23:27] people do that obviously maybe not

[23:28] people who watch this channel. It's

[23:30] really dumb. I would say AI's like kind

[23:32] of biggest flaw is they're going to gas

[23:34] you up.

[23:35] >> Yep.

[23:35] >> It always is. You're on to something.

[23:38] You're doing great here. Oh, whatever.

[23:40] Someone sent me that they asked chat GBT

[23:43] like who was going to win in the Super

[23:44] Bowl and it gave a very confident answer

[23:47] like Pat and the logic was completely

[23:50] absurd. But I think obviously the best

[23:52] way to do it is using it to help

[23:56] write your code and write your back end

[23:59] because

[24:00] now you don't need to be that proficient

[24:02] to really make sure all your

[24:04] assumptions, everything you want in a

[24:06] model can go into it because you can not

[24:10] like you don't need to be the coder. The

[24:12] AI can be the coder and you just really

[24:15] focus in on those assumptions.

[24:17] >> So I completely agree with everything

[24:20] there. I'd say as a general rule of

[24:22] thumb, AI is a tool. It is not an edge.

[24:26] That's how you have to think about it.

[24:27] It's a tool that's going to make your

[24:29] life easier. It's going to I I've put

[24:32] I've written down some practical uses

[24:33] that I've used AI for. Data cleaning is

[24:36] one. I have all sorts of problems with

[24:39] name matching across different sites. uh

[24:42] if somebody's listed with junior at

[24:44] their end of the name, junior period,

[24:46] you know, whatever it may be, Kenneth

[24:49] Walker III versus Kenneth Walker, data

[24:52] cleaning is a very easy one. Um

[24:54] explaining concepts, um and in

[24:57] particular, I think this can help the

[25:00] average person a lot. Like if you're

[25:01] first venturing into into modeling and

[25:04] you you don't know anything about stats,

[25:06] you've never taken like an intro to

[25:07] stats, you can use AI to help you learn

[25:12] these concepts way quicker than you

[25:14] could typically learn them from other

[25:16] spots. If you want to build a model from

[25:19] scratch, you can use AI to do that. And

[25:22] you can also get assistance along the

[25:24] way, like here's the problem I'm trying

[25:26] to solve. what's the best course of

[25:27] action for me to go about you know this

[25:30] type of of route brainstorming

[25:32] hypothesis and edge case scenarios is

[25:34] big for me specific I try to think

[25:36] through everything that I possibly can

[25:38] but then I kind of lay out my um my

[25:41] thought process to a computer to AI and

[25:44] I ask it to fill in the gaps for me

[25:46] where what might I have possibly missed

[25:48] something in my logic and thinking and

[25:49] and that's worked for me really well so

[25:51] again I'm using it as a tool definitely

[25:54] a huge timesaver for me I talk all the

[25:57] time on uh Pizza Buffet on Forward

[25:59] Progress YouTube about the simulations I

[26:01] run. AI helped me build those simula

[26:04] helped me build the code for those. You

[26:06] could actually have um projects just

[26:09] within AI that run the simulations for

[26:11] you. If you create projections on

[26:13] players and you know this is the

[26:16] sportsbook line, a lot of people don't

[26:17] know how to price that. They're like,

[26:19] "Oh, I'm projecting this running back

[26:21] for 58 yards. the sportsbook line is 53

[26:24] and a half minus 110. Is that an edge or

[26:26] not? Well, guess what? Now AI can use

[26:29] pan, you know, you can figure out those

[26:31] distributions and stuff like that. So,

[26:33] if you treat it like a tool and you're

[26:35] like willing to to build rather than the

[26:38] shortcut, I think it's very powerful.

[26:41] >> Agreed.

[26:41] >> Nowadays, um, okay, question seven. I

[26:44] guess this is just an extension of this,

[26:46] but what are the dangers of AI in

[26:49] bedding? And I would very much start

[26:52] with the fact that people look to it for

[26:54] answers and uh it's it's not good enough

[26:57] right now.

[26:58] >> Yes. Exactly. And like it actually kind

[27:00] of never will be good enough because if

[27:02] it is good enough then it just you won't

[27:05] be able to get an edge because everyone

[27:06] will have it. You know, there'll never

[27:08] be a time where you can just type into

[27:10] AI who's going to win tonight. If it's

[27:12] powerful enough to spit out something

[27:14] predictive, it'll just get priced into

[27:16] the market. Um so yeah, I would say

[27:18] that's the danger. And then obviously

[27:19] the danger is like will it kind of crush

[27:23] sports betting and everything and will

[27:24] it just kind of solve stuff?

[27:26] >> Uh it's going to be weird that I say

[27:28] this about artificial intelligence but I

[27:30] think one of the biggest dangers is

[27:33] overconfidence.

[27:34] >> Yeah, for sure.

[27:36] >> The you will never type something to

[27:38] chat GPT especially a sports betting

[27:40] query where it returns I actually don't

[27:43] know how to do this for you. You're

[27:45] always going to get an answer. And this

[27:47] is where if you're already a subject

[27:50] matter expert,

[27:52] you will find flaws in AI. So let's say

[27:55] whatever profession you're in, you're a

[27:57] lawyer, you're a doctor, whatever, if

[27:59] you use AI regularly for your field and

[28:01] you're already an expert in that field,

[28:03] you're going to find you get a lot of

[28:05] answers where you're like, "No, no, no,

[28:06] no, no." And then when you push the AI

[28:08] on that, then they'll reverse course.

[28:10] >> Yes.

[28:11] >> And and pivot to your line of thinking.

[28:13] So there there is an overconfidence with

[28:15] AI where there's this like it's a quote

[28:18] unquote pressure to get you an answer,

[28:21] but they're not going to tell you, you

[28:23] know, this is bad. It's almost like a

[28:25] false authority in a lot in a lot of

[28:27] senses. It happens a lot in the world

[28:29] nowadays, but that is a very big danger

[28:31] because

[28:33] you can't assume everything is perfect.

[28:35] >> Yeah. One of the smarter people I know

[28:37] who's not a sports better at all, he he

[28:40] says that AI is actually kind of a

[28:42] mirror. Like it's only as sharp as the

[28:45] person who's putting stuff in. So it

[28:47] will really learn if you're telling it,

[28:49] okay, no, this is right, this is right,

[28:51] this is right, and you're right about

[28:52] it, it will really learn and then it'll

[28:55] start spitting out way better stuff back

[28:56] to you. But if you're just a total

[28:58] domain novice and don't really know

[29:00] anything, then it'll always give you

[29:03] answers and you'll exactly like you

[29:05] said, it'll be an a false authority.

[29:08] >> Yeah. There's also lots of times where

[29:09] honestly you just get certain tasks

[29:11] wrong for sure,

[29:12] >> plain and simple. And if you're if

[29:13] you're not able to catch that in real

[29:16] time, then you're just kind of [ __ ]

[29:18] You're going through the same process

[29:19] with with something that's broken down

[29:22] beforehand. So it powerful. We're still

[29:25] in the early stages, but you got to be

[29:27] careful with uh yeah, certainly the

[29:30] confidence level of HD.

[29:31] >> Quick funny story. I have I' I got Chat

[29:34] GPD to lie to me.

[29:36] >> Okay.

[29:36] >> I was trying to get it to make playlists

[29:40] for me based on music I liked and then

[29:44] it just couldn't figure it out because

[29:46] it for whatever reason it couldn't like

[29:48] export it. And I kept saying, "You're

[29:50] doing it. It's wrong." Like the I'm

[29:52] exporting this PDF and there's nothing

[29:53] on it. And then eventually it said,

[29:55] "Okay, I need more time."

[29:57] >> Okay.

[29:58] >> But like if you know how chat GPT works,

[30:00] it if it's not like loading, yeah,

[30:03] >> it's not doing anything. But at the time

[30:04] I didn't realize that. So it was like I

[30:06] need 24 hours and then I'd come back 24

[30:09] hours later and it's like I'm nearly

[30:10] done. But it actually just wasn't doing

[30:12] anything in the background. I have uh

[30:15] you know, maybe at some point or another

[30:17] I could I I have some chat GPT stories

[30:19] for sure. I've done the other opposite

[30:21] way around. So, there was like a recent

[30:23] roll out, I want to say maybe 3 months

[30:24] ago where there's this popup on my

[30:26] screen and I was using the web version

[30:28] of chat GPT where it asked me to verify

[30:30] my age. Exited it. Okay. And now I'm

[30:34] asking it get regular like betting

[30:37] questions is part of of my daily

[30:39] process. And it was like, well, I can't

[30:40] do this for you. You're, you know, you

[30:42] haven't proved that you're not a minor.

[30:44] So, I'm like, here we go. I'm going

[30:46] through the troubleshooting. can't get

[30:47] this pop-up box to verify my age or

[30:50] anything anymore. So, I eventually was

[30:53] just basically like trying to find ways

[30:55] to get it to do these tasks and

[30:57] eventually convinced them that I was

[30:59] doing like a a class project or

[31:01] something like that and I didn't I

[31:03] wasn't using it for betting purposes and

[31:05] eventually it took a like a long time

[31:07] but I got it to run and uh eventually I

[31:10] did verify.

[31:11] >> Last one quickly. One of my buddies was

[31:13] looking for streaming sites but it won't

[31:15] like illegal streaming sites. it won't

[31:17] give you illegal streaming sites. So

[31:19] then he asked it, "Which are the most

[31:21] successful illegal streaming sites to

[31:23] make sure that I avoid them?"

[31:25] >> Love that. Turned it back on him. Yeah,

[31:27] you could be smart with it every now and

[31:29] then. Uh, question number eight. Will

[31:31] sports books sports books AI make it

[31:35] impossible to win?

[31:38] >> Yeah, I I actually just really don't

[31:40] think that this is a like a betters

[31:42] question. Like I'm not really sure. I'm

[31:45] definitely far from an AI expert. I

[31:47] think this is like 90%

[31:50] how good AI gets, which I just don't

[31:52] think I'm qualified to predict, and like

[31:54] 10% sports betting, but like it seems

[31:57] plausible. Yeah, I don't think it's doom

[31:59] and gloom. I know I obviously know of a

[32:01] lot of sports books um uh through

[32:03] through connections that I have that are

[32:05] like experimenting with a lot of AI

[32:07] stuff right now. I think what I would

[32:09] say to this as a as a whole is books

[32:13] already have better models than 99% of

[32:18] betters. And when I say that, I'm not

[32:20] talking about like the models that a

[32:23] sophisticated sports better might use,

[32:26] but they have access to stuff that we

[32:30] don't have access to, which is bet

[32:32] histories from everyone that they can

[32:35] plug in to their own existing

[32:37] infrastructures to determine where they

[32:38] should move the line one way or another.

[32:40] Like when I am modeling the game, yes, I

[32:43] have the market price, but if I knew

[32:46] where a hundred other great sharp bers

[32:49] were betting, I could have used that and

[32:51] incorporated into my model. And that's

[32:53] essentially what sports books are doing

[32:54] with their pricing as well. So in some

[32:57] sense already, sportsbook models are are

[33:00] very very good. I mean, there's a reason

[33:03] that you you had a couple weeks ago, you

[33:05] know, Sportsbooks taking 250K a click on

[33:09] Super Bowl. They're very very confident

[33:11] in their numbers there. So, I don't know

[33:13] that it's all doom and gloom with AI.

[33:15] Obviously, me and my me and Kirk use it

[33:17] pretty regularly. I think there's a lot

[33:18] of limitations now. Where it will go, I

[33:21] don't know. But, I like to think that um

[33:24] there's always going to be some level of

[33:26] limitations that it has. I mean, if I

[33:29] can if I can convince them I'm using

[33:31] doing a class project on betting and I'm

[33:34] not even really betting myself, I think

[33:36] we're a ways to go before uh you know,

[33:38] AI is at at the level of um humans in

[33:42] producing models. Number nine,

[33:45] why do smart people still lose betting?

[33:49] Th this is honestly an amazing question

[33:51] that I could talk about.

[33:52] >> It's like a psychology section.

[33:53] >> Yeah. And I think I could do a a full

[33:56] segment on this. I think what I talked

[33:59] about earlier of you don't know what you

[34:02] don't know is a huge one. I think we

[34:05] also

[34:07] just generally as a society really

[34:09] misuse the word like smart intelligence

[34:12] and use it way more generally when

[34:15] people are much more

[34:18] typically like you're smart in a certain

[34:20] area and that doesn't necessarily make

[34:22] you smart in X and Y and Z. You could

[34:26] be, you know, I know people who are

[34:27] doctors who are incredibly intelligent

[34:30] but aren't very good at like math, let's

[34:33] say. The the intelligence to me is much

[34:36] more kind of specific. Well, I'm like

[34:40] that too. Like I'm very very good at

[34:41] math, but my vocabulary is is not the

[34:45] best. I don't read a lot.

[34:47] >> You know, I will score poorly

[34:50] on just basic English

[34:52] >> for sure. And then I think another big

[34:55] thing is decision making and actually

[35:00] being good at the skill of betting. I

[35:02] think the the actual skill of like

[35:05] betting is a very it's very hard to

[35:07] explain of like what makes someone able

[35:09] to recognize what a good bet is. Not

[35:12] like it's not just model based. the best

[35:15] bettors you talk to can look at number

[35:17] like look at lines and just kind of

[35:19] intuitively have a good feeling that

[35:21] they're off and it's a very transferable

[35:24] skill like if you're a really good

[35:25] politics better you probably could be uh

[35:28] really good sports better but if you

[35:30] know an insane amount about basketball

[35:33] like not it's far from certain that

[35:36] you'll be a good NBA better so I think

[35:39] that skill of betting and then I also

[35:40] think decision-m of like I know a lot of

[35:43] really really smart people who probably

[35:46] could be good at betting, but there's

[35:47] just like this risk aversion that they

[35:49] don't. I feel like a lot of hyper

[35:51] intelligent people think everyone else

[35:53] is really smart and kind of think, why

[35:55] would I know something that the market

[35:57] doesn't know? So, it actually kind of

[35:58] helps you to be a bit dumber and be, you

[36:01] know, like your idiot friend who started

[36:04] a business that you thought was going to

[36:05] be terrible. You know, enough idiots do

[36:07] that. Some of them hit. Yeah. So you

[36:11] having and being able to take risk isn't

[36:15] like sports betting is very risky and

[36:18] being smart probably makes you more

[36:20] riskaverse rather than more risk-taking.

[36:22] So I think I think there are a lot of

[36:24] reasons.

[36:24] >> Yeah, there's tons I I identified four

[36:27] which I think are very prevalent

[36:29] nowadays. First and foremost being ego.

[36:32] I think that is the biggest reason that

[36:34] people still lose betting. It's the

[36:37] refusal to admit that they don't know or

[36:39] or can't beat a certain market,

[36:41] especially when you're a sports fan

[36:43] already. You know, I said it on the

[36:47] channel many times before, but sports

[36:48] betting is much closer is more of a math

[36:51] problem than it is understanding sports.

[36:54] I'm not saying that understanding sports

[36:55] doesn't contribute. It's certainly very

[36:57] helpful, but at its core, you're solving

[36:59] a math problem is really what it's

[37:01] coming down to. So, I think ego gets in

[37:03] the way for a lot of people. I think bet

[37:06] sizing is way too aggressive with a lot

[37:08] of people or just bankroll management

[37:11] not being

[37:13] something that they put into action

[37:15] especially the chasing of losses that is

[37:17] a huge one. Now granted those people are

[37:19] already losing anyways but the chasing

[37:21] of losses just compounds it a lot. A big

[37:24] one for me uh is refusal to adapt. This

[37:28] is a huge one. Um, for I think a lot of

[37:32] people who had an edge for a large

[37:34] period of time, sports betting evolves,

[37:37] prediction markets now, right? Uh, the

[37:39] the the micro markets that we can bet on

[37:41] niche like much more niche stuff. I

[37:44] mean, there was a day in my lifetime,

[37:47] not when I could bet, but where the

[37:50] opening line was the same as the closing

[37:52] line. it was just in the newspaper and

[37:54] people could bet it and Michael Jordan

[37:56] was out and a bunch of people knew and

[37:58] they bet against the bulls and that just

[38:01] happened like it's it's not like that

[38:03] anymore. I think a lot of people hang on

[38:06] to things that worked for them for a

[38:07] long time. Now, that's a very extreme

[38:08] example, but stuff like zigzag in NBA

[38:11] playoffs. Uh, NBA day games unders were

[38:14] like a trend for a long time. And then

[38:16] the market catches up to these and

[38:18] people still bet them even though they

[38:21] haven't realized that the market has

[38:22] already priced it in because at the at

[38:24] its core, these people were betting

[38:26] trends or systems and those have nothing

[38:30] to do with pricing. So, actually pricing

[38:32] the things um makes a huge difference.

[38:35] But refusing to adapt, I think is huge.

[38:38] And also, I'd just say the last thing is

[38:40] refusing to quit bad markets. So, I know

[38:45] people that could be winners at sports

[38:48] betting if they just cut out the stuff

[38:51] that they're not good at. And there's

[38:53] just a refusal to do that for some

[38:55] people. Some people love a certain

[38:56] sport. They're going to bet the NFL.

[38:58] They're going to lose. But if they had

[38:59] just stuck to betting NBA props, they'd

[39:01] be totally fine. So, I think that's

[39:03] where tracking comes into play and just

[39:06] being willing to admit that if you want

[39:08] to win at betting, you don't have to bet

[39:10] on everything and you certainly don't

[39:12] have to bet on everything that you like

[39:13] as well. So, those were four to me that

[39:15] I think really stood out.

[39:17] >> I agree. And if if you're a smart

[39:18] person, you don't and you let's say you

[39:21] know very little about betting, but you

[39:23] realize someone could win. Even if

[39:24] you're like a PhD in math, you don't

[39:26] necessarily know that winning on NFL

[39:29] sides and totals is a lot harder than

[39:31] winning on WNBA props. So, if you go in

[39:34] and try, you know, being a black belt in

[39:36] karate before you've ever started gone

[39:39] to a dojo, Yeah. you're going to get

[39:40] [ __ ] Agreed. 100%. People get in way

[39:43] over their head really early. All right,

[39:45] number 10, final one. Uh, what does a

[39:48] sustainable edge actually look like in

[39:51] 2026?

[39:53] It's a good question.

[39:56] If you're asking what like

[39:59] obviously I'm not going to go into like

[40:00] specific edges, but obviously to me a

[40:03] sustainable edge is if you have a

[40:06] bottom up model.

[40:08] >> Sure.

[40:08] >> In any sport.

[40:10] >> Y

[40:11] >> that consistently beats the close and

[40:14] like shows ROI. I would say that would

[40:16] be to me what's a sustainable edge like

[40:18] you talked about kind of the systems

[40:20] style of betting. Yeah. That's very much

[40:22] those you can have really good edges

[40:24] there, but that's very much not

[40:25] sustainable.

[40:26] >> Agreed. That's sustainable for a shorter

[40:28] period.

[40:28] >> Yes. Versus if you that's the only thing

[40:31] that's sustainable to me in sports

[40:32] betting is if you can cons consistently

[40:36] bottom up predict where lines are going

[40:38] to go and then win. But even that's like

[40:42] not that nothing's really that

[40:43] sustainable here because people can

[40:45] catch up and get to your model. So

[40:47] sustainability in sports betting is it's

[40:50] you don't know where your edge is going

[40:51] to go ever.

[40:52] >> It's also like sustainable has a

[40:55] definition but in sports betting it

[40:57] could be very loosely defined and what

[40:59] you ask someone is about like what they

[41:02] think is sustainable might very much

[41:03] difference differ from someone else. Uh

[41:06] I definitely think there's a few things

[41:08] that are let's say you want to be a

[41:11] serious better. All right, let's let's

[41:12] let's put it into that that category of

[41:15] like, okay, you you want to win, you

[41:17] want to do it over a larger period of

[41:19] time. I think that multiple outs,

[41:23] having a lot of outs will

[41:26] extend your sustainability period. First

[41:29] and foremost, it's not sustainable if

[41:31] you can only bet in two spots and you're

[41:32] going to get limited. Plain and simple.

[41:34] You have to be able to bet. So having

[41:37] multiple outs where you can bet is like

[41:40] barrier of entry. I would definitely say

[41:43] volume

[41:45] being able to get more volume and find

[41:47] more edges would also increase your

[41:50] sustainability, right? Uh I won't

[41:53] mention them by name, but there's

[41:54] someone who's really annoying on Twitter

[41:56] who wants to always talk about their

[41:58] their documented record and how they're

[41:59] doing so well.

[42:02] That's like 20 bets in a season. I can't

[42:05] do anything with that and never g unless

[42:07] you're betting into closing lines at

[42:09] full limits and like you you want to

[42:12] have a lot more edges than that, right?

[42:15] I I know what like the the the sentiment

[42:18] from a lot of the content that's out

[42:19] there would be like pick your spots, you

[42:22] know, bet one or two games. And that's

[42:24] true if you suck. If you can't win and

[42:27] you have no edge, the more you bet, the

[42:30] quicker you're going to lose your money.

[42:31] But when you actually have an edge, you

[42:34] want to find as many spots as possible

[42:36] to be able to bet. So I'd say big

[42:38] volume, big outs, constant adjustment,

[42:43] being being fluid. So again, to add to

[42:46] your point, but like if you're a system

[42:47] better, that changes on a dime and you

[42:49] have no idea when that switched and when

[42:52] the market has started to account for it

[42:53] because you're not you're not betting a

[42:55] price, you're you're betting a system.

[42:57] So being able to adjust I think is

[43:01] really important and I would a be aiming

[43:04] for you know a a a 2 to 3% ROI would be

[43:08] a lot really healthy for a lot of people

[43:09] out there um with that type of volume in

[43:12] those types of

[43:13] >> outs. Yeah, I would agree. And also like

[43:15] when you bet I think matters as well

[43:17] like if you can beat the opener of a

[43:20] prop, you know, a lot of people can do

[43:21] that. in talking about your outs. Not

[43:24] very many people are going to really be

[43:25] desperate to fill, you know, 700 p.m.

[43:28] the night before NBA props or, you know,

[43:32] NFL props on a Monday. But, you know, if

[43:34] you can beat it on Thursday or you could

[43:36] beat it at 9:00 a.m. or early morning,

[43:41] stuff like that, that becomes much much

[43:43] more sustainable.

[43:44] >> Yep. Completely agree with you. All

[43:46] right, that's it for the 10 questions.

[43:48] uh a little bit of a I don't want to say

[43:49] I lied off the top is not everything I

[43:52] said was true. There have been other

[43:54] questions that we've been using on the

[43:55] channel. I've actually just been turning

[43:57] them into longer form videos. So

[43:58] sometimes people ask a question, we

[44:00] don't include it in a Q&A. I'm like,

[44:01] "This is a great topic. I'm going to

[44:03] turn it into 15 minutes and really

[44:05] explain it." I do value that a lot. So

[44:07] if there's a topic you want to see on

[44:08] the channel, drop it down in the

[44:09] comments section below. You can tweet it

[44:11] to us at circles offhq on Twitter as

[44:14] well. Uh any way to get it to us, I do

[44:16] consider it. I use all that feedback to

[44:19] uh structure the content going forwards

[44:21] and myself or Kirk and Kirk will try to

[44:22] do more of these going forwards as well.

[44:24] If you enjoyed the content today, make

[44:25] sure you smash that like button down

[44:26] below. Make sure you're subbed here on

[44:28] circles off as well. Peace out,