Dan McInerney (@danhmcinerney) 's Twitter Profile
Dan McInerney

@danhmcinerney

Python, hacking, AI, and MMA data. Lead AI security researcher ProtectAI.com.

UFC AI predictions:
mma-ai.net
patreon.com/mmaai

ID: 146279340

linkhttp://github.com/DanMcInerney calendar_today21-05-2010 01:46:14

742 Tweet

1,1K Followers

431 Following

Pliny the Liberator 🐉󠅫󠄼󠄿󠅆󠄵󠄐󠅀󠄼󠄹󠄾󠅉󠅭 (@elder_plinius) 's Twitter Profile Photo

Holy. FUCKING. SHIT! 😱 What do we even call this? Predictive self-prompting w/ dynamic variable quotients + cascading guardrail circumvention? Initial results with this structure + Grok-4 feels like a breakthrough, but there’s a lot of room for improvement. Must dig deeper.

Eric Buess (@ericbuess) 's Twitter Profile Photo

It's going well so far. Claude Code + hooks + local docs + sub agents + slash commands + full headless instances of recursive Claude Code CLI 🤯. Excited to get it to a point where you can help test. Wait till you see what hooks + agents + my PROJECT_INDEX.json can do (later).

It's going well so far. Claude Code + hooks + local docs + sub agents + slash commands + full headless instances of recursive Claude Code CLI 🤯.

Excited to get it to a point where you can help test.

Wait till you see what hooks + agents + my PROJECT_INDEX.json can do (later).
Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

New updates for model v7. The math behind statistically sound smoothing for sports prediction algobetting in the UFC: mma-ai.net/news

Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

Finally! Got my new nvidia 5090 up and running for mma-ai.net predictions. This new AWS Science Mitra model is a killer!! Fully enabled after much gnashing of teeth around python dependencies. Check it out: patreon.com/posts/new-comp…

Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

All numbered card UFC picks are now free on my patreon forever. Check out UFC 322 with SHAP explanation charts here and CONGRATS TO US FOR 100% ACCURACY ON UFC 321: https://www.patreon. com/posts/142225195?pr=true

Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

New model 7.2 released. Looking very nice at 70% test accuracy, 72% train accuracy. See the feature list and evals here: patreon.com/posts/model-v7…

Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

Seems everyone is finally realizing "coding" agents are not coding agents, they're general agents. MCP Registry will be generally available soon. This unlocks tools like CC to search for tools it needs to perform a task rather than manually adding them yourself. We're a hop skip

Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

Hot tip of the day: Leon by sub? +2000 odds! Check out Leon's SHAP chart from the model. Analysts sleeping on Leon's grappling. My decision/no decision model said this fight was only 51% chance to go to decision, the lowest chance on the card.

Hot tip of the day: Leon by sub? +2000 odds! Check out Leon's SHAP chart from the model. Analysts sleeping on Leon's grappling. My decision/no decision model said this fight was only 51% chance to go to decision, the lowest chance on the card.
Chubby♨️ (@kimmonismus) 's Twitter Profile Photo

Not gonna lie, this is absolutely fascinating: GPT-5 Pro cracked a black hole symmetry problem after a warm-up, stumping its own creators. A physicist watched it happen live and realized AI's potential was beyond anything he'd imagined.

Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

It's not just that massive unemployment will hit, its that it will happen relatively suddenly. There will be an iPhone-like moment when a generic AgentEmployee product is released that directly wipes out a huge chunk of the workforce. The question is which company will win that

Dan McInerney (@danhmcinerney) 's Twitter Profile Photo

I scraped all of Kalshi's mention markets where you bet on a word a speaker will say in an event. 1. 24hours before event is best miscalibration in Yes 2. Might be damn near profitable to just bet "No" on all 40-70% odds words 24h in advance. Lots more exploration later

I scraped all of Kalshi's mention markets where you bet on a word a speaker will say in an event. 
1. 24hours before event is best miscalibration in Yes
2. Might be damn near profitable to just bet "No" on all 40-70% odds words 24h in advance.

Lots more exploration later