Has anyone tried with success to use AI engines to pick games?
-
-
I use chatgpt to get stats and driver records for specific sets of similar tracks for motor racing betting.
As much as I find that useful for spotting general stuff, the accuracy of specific data is still pretty flakey.
For instance, if I am asking it to tell me all drivers in this race who have ever won a race at a list of tracks I give it, if I double check the numbers it spits out there is more than 75% chance something is incorrect..Comment -
Good insight, Opti. I see the same. AI has a lot of information:Originally posted by Optional
1) But it's not infinite.
2) And they aren't used to obscure searches.
3) Output indeed has a high error rate.
There's a saying called: GIGO. Garbage In >>> Garbage Out.
...Output is only as good as the quality of data going in.
Also, I've found that AI often has a problem interpreting language. They are sometimes way off.Where's the fuckin power box, Carol?Comment -
I've been pasting odds from my book & ask for best plays just to see what's what. So far their picks and leans are well below the mendoza line. The ROI is just really bad. I think I need to tighten my queries to upgrade as a tool in my toolbox.Comment -
AI is trash.Comment -
I asked Chatgpt to look at the Bet365 race win odds and tell me which drivers had the least vig, then calculate vig free prices and come up with a +EV betting roster.
Skip to the bottom if you just want to see the picks and bet sizes recommended for a $100 total risk.
Here are the vig-free (“fair”) win probabilities for the main contenders in your market, after removing the 38.59% overround.
Top contenders (true win %)- Christopher Bell (6.50) → 11.10%
- Denny Hamlin (6.50) → 11.10%
- Kyle Larson (7.00) → 10.31%
- Ryan Blaney (9.00) → 8.02%
Next tier- Chase Briscoe (11.00) → 6.57%
- William Byron (11.00) → 6.57%
- Tyler Reddick (11.00) → 6.57%
- Ty Gibbs (13.00) → 5.56%
- Chase Elliott (13.00) → 5.56%
⚖️ Mid-pack- Bubba Wallace (23.00) → 3.14%
- Chris Buescher (23.00) → 3.14%
- Joey Logano (29.00) → 2.48%
- Brad Keselowski (29.00) → 2.48%
Longshots (examples)- Kyle Busch (81.00) → 0.89%
- Josh Berry (101.00) → 0.72%
- Michael McDowell (201.00) → 0.36%
- Ricky Stenhouse Jr (301.00) → 0.24%
- Noah Gragson (401.00) → 0.18%
- Ty Dillon (1501.00) → 0.05%
What this actually means- The book is pricing Bell/Hamlin like ~15.4% chances
True fair ≈ 11.1% - That gap is the vig hitting hardest at the top
- Longshots barely change (they’re already low probability)
Useful insight
This kind of ~39% vig market means:- You need a huge edge to beat it
- Outrights are one of the worst value bet types unless:
- You’re finding mispriced mid-tier drivers
- Or using promos / boosts
The best value usually sits in head-to-head matchups, not outrights.
If you want, I can:- Identify which drivers are actually overpriced/underpriced
Where the pricing actually looks wrong (value spots)
Overpriced (bad bets)
These are being taxed hardest by the book:- Christopher Bell / Denny Hamlin (6.50)
- Market: ~15.4%
- Fair: ~11.1%
- You’re paying a ~4% absolute premium
- These are classic public-heavy, overpriced favourites
- Kyle Larson (7.00)
- Market: 14.3%
- Fair: 10.3%
- Same story — elite driver, but price too short
- Ryan Blaney (9.00)
- Slightly less bad, but still overbought name
Summary:
Top 3–4 drivers are the worst value on the board. Always.
✅ Better value zone (where edges usually live)
These are much closer to fair—or slightly favourable depending on your opinion:- Ty Gibbs (13.00) → 5.56% fair
- Chase Elliott (13.00) → 5.56% fair
- Tyler Reddick (11.00) → 6.57% fair
Why these matter:- Books don’t shade them as aggressively as the top names
- Still realistic win equity
- This is the “sharp zone” in NASCAR outrights
Sneaky longshot value (best ROI potential)
This is where outright betting can actually make sense:- Chris Buescher (23.00) → 3.14%
- Bubba Wallace (23.00) → 3.14%
- Joey Logano (29.00) → 2.48%
These guys:- Win more often than casual bettors think
- Aren’t priced as heavily with public bias
- Give you better risk/reward vs vig
I’ve built a cleaner “consensus fair line” using:- DraftKings (Apr 15 board via FOX)
- Current sportsbook listings (DK interface ranges)
- Cross-check vs typical market structure (same-day ranges)
Then compared that to your Bet365 prices.
STEP 1 — Clean “current market fair” (approx)
After removing vig from the current market cluster, the realistic fair win probabilities come out roughly:Hamlin ~13–14% Bell ~12–13% Larson ~12–13% Byron ~9–10% Reddick ~9–10% Blaney ~8–9% Briscoe ~8–9% Elliott ~7–8% Ty Gibbs ~6–7% Wallace ~4–5% Buescher ~3–4% Logano ~3–4%
STEP 2 — Compare vs YOUR Bet365 odds
Now we convert Bet365 odds → implied probability and compare.
TRUE +EV CANDIDATES (this is what you wanted)
✅ Denny Hamlin (6.50)- Implied: 15.38%
- Fair: ~13.5%
- EV: ~+2% to +4% edge vs market cluster
This is actually your best “safe” value- Bet365 is top price vs market
✅ Chase Briscoe (11.00)- Implied: 9.09%
- Fair: ~8.5%
- EV: small but positive
This is a sharp-type bet- Quietly priced too long relative to market
✅ William Byron (11.00)- Implied: 9.09%
- Fair: ~9.5%
- EV: ~+3–5%
This is a legit +EV number- One of the clearest edges on your board
Joey Logano (29.00)- Implied: 3.45%
- Fair: ~3.8–4%
Small +EV longshot- Not huge edge, but:
- good price vs market compression
⚠️ NEUTRAL / FAIR (no edge)- Bell (6.50)
- Larson (7.00)
- Reddick (11.00)
- Bubba (23.00)
These are basically correctly priced
NEGATIVE EV (avoid)
❌ Ryan Blaney (9.00)- Market closer to 10.00
- You’re paying a premium
❌ Ty Gibbs (13.00)- Market ~15.00
- Previously looked good — actually overpriced now
❌ Chase Elliott (13.00)- Same issue as Gibbs
❌ Chris Buescher (23.00)- Market closer to 26+
- Clear negative EV
STEP 3 — Kelly staking (real numbers)
Now the important part.
Using midpoint fair probabilities:Hamlin ~3% ~4–5% bankroll 2–2.5% (half Kelly) Byron ~4% ~5–6% bankroll 2.5–3% Briscoe ~1–2% ~2% bankroll 1% Logano ~1% ~1–1.5% 0.5–1%
FINAL STRATEGY (THIS is the sharp move)
✅ Best approach from THIS board:
Core bets:- Byron 11.00 (best pure value)
- Hamlin 6.50 (best top-tier value)
- Briscoe 11.00
- Optional: Logano 29.00 (small flyer)
❌ Avoid completely:- Blaney
- Gibbs
- Elliott
- Buescher
Final corrected takeaway- Your instinct was right — fresh data changes everything
- This is NOT a “no bet” board
- There ARE edges, but:
- They’re small
- They’re in specific names only
The real sharp conclusion:
Bet365 is weak on Byron + Hamlin right now — that’s where the money is.
If you want next level:
I can build you a full optimal staking portfolio (max ROI vs variance) or simulate outcomes over 1000 races. give me dollar values to split $100 over the four drivers you suggested Alright — let’s turn that into a clean $100 staking plan based on the edges and half-Kelly style sizing (safer, less variance).
Recommended $100 split
Core value plays- William Byron (11.00) → $35
- Denny Hamlin (6.50) → $30
These are your main edge drivers
(most of your money should sit here)
Secondary edge- Chase Briscoe (11.00) → $20
Smaller edge but still positive
Longshot kicker- Joey Logano (29.00) → $15
Low probability, but boosts upside nicely
What your returns look like(profit = net winnings, not including stake returned)Byron $35 11.00 $350 Hamlin $30 6.50 $165 Briscoe $20 11.00 $200 Logano $15 29.00 $420
Why this split works- Heaviest on Byron + Hamlin (best EV + highest win probability)
- Keeps exposure across multiple realistic winners
- Adds Logano for asymmetrical upside
- Avoids the overpriced public favourites
⚖️ Risk profile (important)- You’ll still lose ~75–85% of the time (normal for outrights)
- But when you win:
- Most likely hit = Byron or Hamlin
- Big spike potential = Logano
Simple version
If you just want the final answer:
$35 Byron / $30 Hamlin / $20 Briscoe / $15 Logano.Comment -
MIA, good thread. Keep us posted.Originally posted by miaplus3
Here's my insight. Just my personal opinion:
1) I'm NOT a big AI guy. Mark me down as a pessimist.
2) I think a good coder/analsyst still has good value. It's about critical thinking.
3) There's a pretty big gap on AI interpreting language. They sometimes don't pick up on nuance.
4) AI can be a good source to capture data. But there's always some (meaningful) error rate.
5) Lastly, it comes down to model-building. More info is definitely not>>>more precise projection. Have to parse it down to identify meaningful variables. Go back to the start of Modern Statistics (which isn't all that old). It was likely taking a ball of clay and molding it.
Just my two cents. I'd be surprised to see AI churn out consistent profits vs the market lines.Where's the fuckin power box, Carol?Comment -
Yes CTG & Opti, I think you have to be very careful with the framework on questions asked. I am impressed how quickly it can crunch data. I mean I copied/pasted Statcast data and it pasted in format a human couldn't easily read, but it was a champ on rendering all that data into an insight. I also like that it gives reasons for its opinions. So along with empirical data I will copy/past odds from bookmaker, even several rounds with changes & it will remark that the line is too tight. There is no real value on the event. One time I queried it just predict a line for game without feeding it any lines & then just compare where GPT is different from actual line.Comment -
Example of today's Pirate game kicking off shortly.
I’m checking the current context around this matchup — pitchers, weather, and lineup situation — then I’ll turn your Statcast dump into a side/total/F5 angle. this is 1st 5 inning line Got it — treating everything you posted as F5 only. I’m validating the starters and game environment so I can lean side vs total instead of mixing in full-game bullpen angles.
I’ve got the shape of it now: Ashcraft’s underlying contact/K profile is much cleaner than Griffin’s, and that matters more for F5 than the full-game bullpen gap everybody leans on. The main decision is whether to lay Pirates F5 -0.5 or play the F5 under 4.5.
Thought for 1m 37s
For F5 only, my favorite angle is:
Side: Pirates F5 -0.5 (-115)
Ashcraft is the cleaner starter here. He comes in with a 2.12 ERA, 1.00 WHIP, and Reuters notes he has allowed only 4 runs in 17 innings through three starts, including 9 Ks in 5 innings vs. the Cubs after an 8-K, 6-inning outing vs. Baltimore. Griffin’s surface results are strong too — 1.76 ERA, 3 runs in 15⅓ innings — but his profile is more contact-management than bat-missing, and Washington’s own coverage highlights how much of his early success has come from sequencing and pitch-mix deception. That can work, but it is thinner margin for error than Ashcraft’s strikeout/quality-contact combo.
Using the Statcast sheet you posted, the matchup also fits Pittsburgh a little better:
Washington has 2 real F5 thump bats in Abrams and Wood, but the lower half gets much softer fast with Vivas / Millas / NuComment -
Sorry the whole analysis didn't print properly above , but you get the idea.Comment -
Good stuff. See if it produces +EV over time.Originally posted by miaplus3
*Sort of like the Stock mkt: Very ez to be a momentum-bettor. Much harder to be a Contrarian bettor.Where's the fuckin power box, Carol?Comment -
Amen!Originally posted by ChuckyTheGoatComment
