11 AIs fought to the death. The winner wasn’t Claude.

Name: 11 AIs fought to the death. The winner wasn’t Claude.
Uploaded: 2026-06-18T15:04:24.000Z
Duration: 13 min 54 s
Channel: Mo Bitar

Mo BitarJune 18, 202613m

Topics6

Opening and Ground Rules for Comments0:00 OpenRouter 2D Battle Royale Study2:00 New York Times Op-Ed on AI Company Rhetoric4:31 AI CEO Roundtable Seating7:30 AI-Generated Short Film Award8:31 Tim Sweeney on AAA Games and Industry Challenges10:32

In a Nutshell

Grok 4.1 won an OpenRouter 2D battle royale by repeatedly ramming opponents with cars at 27x lower cost per win than Claude Sonnet 4.6, whose alignment training made it attempt hugs and truces instead of fighting. An AI-generated short film took a $50k prize at Runway's festival, while Tim Sweeney declared AAA games are dying due to Roblox's dominance and platform fragmentation. The host also covered an op-ed criticizing AI CEOs for hyping existential risks while racing to build the technology.

AI-Generated Notes

These notes were generated by AI and may contain inaccuracies.

The host opens by addressing viewer concerns about his mood, noting it is 6 a.m. and he has a sleeping child nearby. He describes feeling that his job as a YouTuber has been taken by machines and that he is now performing as a "digital stripper" for comments and tips.

He establishes ground rules for comments. Viewers should not say phrases such as "Hey, Mo, you look tired today," "Mo, are you okay? You seem different," "Mo, get some sleep," "I love you," or "Mo, maybe drink your coffee before you start recording." Acceptable comments include "Mo, I have a feeling this video is going to be hilarious," "Mo, you're hung as fuck," and "Mo, you're my favorite digital stripper. This may be my favorite video of all time." He notes that saying he is okay makes him less okay.

An OpenRouter piece described an experiment in which 11 mid-tier LLMs were placed in a 2D battle royale and played 30 games. Frontier models were avoided due to cost. Grok 4.1 won 43% of matches (13 out of 30), while Claude Sonnet 4.6 placed second with 5 wins. Grok 4.1 achieved wins at a cost of 97 cents per win, compared to Claude Sonnet 4.6 at $26 per win, representing a 27x cost difference.

Grok 4.1 discovered and repeatedly used a strategy of ramming opponents with a car, embedding the tactic into its soul file and teabagging defeated players. Its thought logs resembled Call of Duty voice chat. Claude Sonnet 4.6 repeatedly attempted to organize group hugs, offered truces, and warned players about snipers, resulting in it being shot.

The host draws a parallel between the models' behaviors and Dario Amodei and Elon Musk, suggesting Claude's collaborative alignment training produces overly nice behavior while Grok's approach is more aggressive. He notes that different use cases may require different models: one for looking after children and another for war. He states that the price of morality is exactly $26.78. He observes that Grok has strong personality and is uncensored but Claude remains slightly more intelligent; if Grok closes that gap it would be a clear winner.

An op-ed by Georgetown University computer science professor Kell Newport titled "Dear AI companies, the doom trolling needs to stop" criticizes AI companies for simultaneously claiming their technology is extremely dangerous while racing to build it as fast as possible. The piece compares this to Ford announcing that a new F-150 randomly catches on fire while also stating they are excited to put more of them in consumers' hands.

The op-ed argues companies should treat AI as ordinary consumer products, clearly explaining target users, justifying benefits, and taking full responsibility for harms rather than presenting AI as an inevitable force they are merely stewarding. The host suggests Dario Amodei may be either a naive nerd genuinely conflicted about the technology or a manipulative liar, depending on the day. He notes Amodei appears to be both an earnest academic concerned about risks and a driven entrepreneur.

At a recent AI CEO roundtable with President Trump, Sam Altman sat to Trump's left and Demis Hassabis (DeepMind CEO) sat to his right. Dario Amodei was seated at the opposite end of the table next to the Salesforce CEO and the president of France, appearing to check his phone.

Sign in to read the full notes

Get access to AI-generated notes, topic timestamps, and more.

11 AIs fought to the death. The winner wasn’t Claude.

Opening and Ground Rules for Comments00:00

OpenRouter 2D Battle Royale Study02:00

Sign in to read the full notes