Justin Nelson @JustinDoesMath

Head of Sports Analytics @ The Crowd's Line. Building betting models via crowd-sourcing & ML/AI. Former quant futures PM. MIT: physics & math. From rural Texas. thecrowdsline.ai New York, USA Joined September 2013

Tweets

2K
Followers

656
Following

3K
Likes

13K

G3 Media @G3imagazine

2 months ago

OpticOdds enhances WNBA player props with The Crowd’s Line AI g3newswire.com/opticodds-enha… via @G3imagazine

0 1 1 72 0

View Details

Justin Nelson @JustinDoesMath

8 months ago

You can thank your state government for refusing the Medicaid expansion. Obama did not foresee governors forfeiting free federal funds that would subsidize people in between the poverty line and when the federal subsidies kicked in. Your governor screwed y’all over just to make the ACA look bad. And, it worked. “Despite the Affordable Care Act offering states the opportunity to expand Medicaid, these 10 Republican-led states continue to resist fully expanding this critical program. Republicans in Alabama, Florida, Georgia, Kansas, Mississippi, South Carolina, Tennessee, Texas, Wisconsin, and Wyoming are leaving billions of federal dollars on the table, continue to lose funding for rural hospitals, and represent a broader GOP willingness to risk the lives of their constituents.” dlcc.org/press/ten-repu…

0 0 0 26 0

View Details

Justin Nelson @JustinDoesMath

9 months ago

@Charlie40074981 @danmana @guitchounts @slotkinjr @Waymo “Waymo will always take the safer route” This isn’t the “gotcha” you think it is.

1 0 2 47 0

View Details

Justin Nelson @JustinDoesMath

9 months ago

@rohanpaul_ai @drfeifei There is no combustion-based energy sources in nature but jets got a lot faster than birds. Evolution is notoriously locally greedy. Not so good at finding global maximum.

2 0 2 538 0

View Details

Chi Ossé @OsseChi

11 months ago

WHY SHIT NOT WORKING: EPISODE 11

219 2K 15K 984K 3K

View Details

Justin Nelson @JustinDoesMath

12 months ago

@SawyerMerritt @Tesla Data quality matters. The Waymo paper can only inform us what happens when Waymo scales up their high-quality multi-modal data. It does not necessarily follow that Tesla’s vision-only, noisier data would benefit in the same way.

0 0 1 654 0

View Details

The Crowd’s Line AI @theCrowdsLineAI

12 months ago

NBA Finals Game 3 Recap 📈 Model was wildly accurate last night 🔥 Overall: 24-10 (71%) 🎯 Average Odds: +164 🔥 Unders: 17-2 (89%) 🎯 Average Odds: +120 Avg Profit: +3.4 units 💰 ROI: +96% 1 of the 10 losses? Haliburton triple double @ +2100. Fell 1 board short!

6 2 5 1K 0

View Details

The Crowd’s Line AI @theCrowdsLineAI

a year ago

Here’s the model’s performance by month since going live last season ⬇️ After a disappointing April (+7.4U), we rallied for our 2nd best month of the season in May (+59.1U) 📈 Model’s been consistently good at predicting assist props throughout. +172 units combined L2 seasons!

0 1 3 306 0

View Details

The Crowd’s Line AI @theCrowdsLineAI

a year ago

May 2025 NBA Model Recap 🗓️ Overall, we gained +59.1 units 📈 Most of that 💵 came in Round 2. Conference finals proved challenging Top 3 markets by PnL: 🥇Assists (+15.3u) 🥈Points (+14.7u) 🥉Pts+Ast (+13.8u) For the year, we’re now up +240 units More details in thread 🧵⬇️

1 1 3 445 0

View Details

Rohan Paul @rohanpaul_ai

a year ago

This paper shows LLMs internally break down composite tasks. They process components sequentially across layers. Methods 🔧: → Layer-from context-masking selectively blocks in-context example access from a specific layer onward. → This technique reveals sequential subtask learning points via an "X-shape" accuracy pattern across layers. → Cross-task patching injects activations from composite tasks into zero-shot subtask prompts. → It demonstrates composite task processing generates distinct, transferable subtask vectors, averaging 0.66 strength in Llama-3.1-8B. 📌 Layer-wise task execution enables precise intervention points. 📌 Task vectors offer a handle for steering learned behaviors. 📌 Internal processing mirrors explicit reasoning steps without outputting them. ---------------------------- Paper - arxiv. org/abs/2505.14530v1 Paper Title: "Internal Chain-of-Thought: Empirical Evidence for Layer-wise Subtask Scheduling in LLMs"

0 34 133 9K 101

View Details

The Crowd’s Line AI @theCrowdsLineAI

a year ago

Sunday NBA Model Recap 🔥 Terrifyingly accurate… again! 🎯 Best wins: ✅ Wallace 5+ Assists (+2090) ✅ Wiggins 3+ Assists (+1440) ✅ Wiggins 15+ PRA (+695) ✅ Braun Under 0.5 Blocks (+138) Avg Profit: +10.6 units 📈 ROI: +108% 💰💰💰💰💰💰💰💰💰💰💰💰💰

0 1 3 382 0

View Details

The Crowd’s Line AI @theCrowdsLineAI

a year ago

Wednesday NBA Recap 🔥 Top 20 Picks: 19-1 (95%)‼️ Average Odds? +125😏 Best Wins💰 ✅ Podz 25+ PA (+1200) ✅ McBride 3+ Assists (+671) ✅ Butler UNDER Everything (+3.1U) Avg Profit: +14.6 units 📈 ROI: +96% 💰💰💰💰💰💰💰💰💰💰💰💰💰 Next Stop 📲 thecrowdsline.ai/dashboard/nba/…

2 2 7 935 1

View Details

Jared Smith @jaredsmithbets

a year ago

Absolutely crushing it 👏

The Crowd’s Line AI @theCrowdsLineAI

a year ago

Featured Pick of the Day: Podz O3.5 Assists (+119) Also available at +100 on FanDuel, BetMGM, and Caesars This line is 12-3 in Podz's last 15 games without Steph (going back to last year). No Steph, ~hopefully~ no problem! Full NBA forecast: thecrowdsline.ai/dashboard/nba/…

1 1 8 11K 1

0 3 15 9K 1

View Details

Jared Smith @jaredsmithbets

a year ago

If you aren't following @theCrowdsLineAI you're missing out on free and easy 1st half winners like this Also hit the alt for +255!! 👏

The Crowd’s Line AI @theCrowdsLineAI

a year ago

Today's Pick of the Day: Mikal Bridges O 2.5 Ast! This bet is available at +100 on DraftKings and BallyBet and -102 on FanDuel 💰 Our model has been HOT with Bridges – he's our 3rd best performing player on the season! Let's see if he can get it done for us in Game 1 at the

1 1 7 7K 2

1 3 5 5K 0

View Details

Justin Nelson @JustinDoesMath

a year ago

@teortaxesTex The Korean Peninsula offers a much better natural experiment.

2 0 2 419 0

View Details

Justin Nelson @JustinDoesMath

a year ago

@teortaxesTex Given his perma-bear stance thru the entire post-08 bull market, FINRA probably did him a favor.

0 0 0 20 0

View Details

Justin Nelson @JustinDoesMath

a year ago

@ProBonoRapLines @The_Odds_API I emailed them at 2:30pm ET and have not heard back.

1 0 0 125 0

View Details

Justin Nelson @JustinDoesMath

a year ago

@ProBonoRapLines @The_Odds_API Glad to know for certain it's not user error on my end. It was working fine until 2pm and now they're not returning any live event ids for NBA or MLB.

1 0 0 156 0

View Details

The Crowd’s Line AI @theCrowdsLineAI

a year ago

Ready to have your mind blown?🤯 See the model’s win percentage by odds range this NBA season: 💎 Sample size: 34,172 total picks 💎 Profitability: Everywhere✅ Seriously — If 𝘢𝘯𝘺 𝘰𝘵𝘩𝘦𝘳 𝘮𝘰𝘥𝘦𝘭 on the street can beat these results across 34K+ picks — let us know‼️

1 3 3 895 1

View Details

Sebastian Raschka @rasbt

a year ago

As we all know by now, reasoning models often generate longer responses, which raises compute costs. Now, this new paper (arxiv.org/abs/2504.05185) shows that this behavior comes from the RL training process, not from an actual need for long answers for better accuracy. The RL loss tends to favor longer responses when the model gets negative rewards, which I think explains the "aha" moments and longer chains of thought that arise from pure RL training. I.e., if the model gets a negative reward (i.e., the answer is wrong), the math behind PPO causes the average per-token loss becomes smaller when the response is longer. So, the model is indirectly encouraged to make its responses longer. This is true even if those extra tokens don't actually help solve the problem. What does the response length have to do with the loss? When the reward is negative, longer responses can dilute the penalty per individual token, which results in lower (i.e., better) loss values (even though the model is still getting the answer wrong). So the model "learns" that longer responses reduce the punishment, even though they are not helping correctness. In addition, the researchers show that a second round of RL (using just a few problems that are sometimes solvable) can shorten responses while preserving or even improving accuracy. This has big implications for deployment efficiency.