It doesn’t make the same mistakes. My personal Claude was constantly making frustrating mistakes and I even asked it whether Anthropic did this intentionally to get people addicted to it and stringing something along. A project that should take a month to solve is now stretched into six months with no end in sight.

Also, Anthropic immediately released Opus v4.6 after Gemini released v3.0.

So these companies are holding onto much more advanced models and are giving us the absolute bottom of the barrel to play around with.

I’m not complaining because I know how dangerous it would be to just release the most powerful models to the public. It’s just an observation.

The AI race however might speed up the timeline a bit.

(http://www.autoadmit.com/thread.php?thread_id=5839187&forum_id=2#49701301)

are you saying the loopholes were NOT ALL CLOSED

your black wife must be very disappointed in you

(http://www.autoadmit.com/thread.php?thread_id=5839187&forum_id=2#49701434)

(guy with DVRO charges)

(http://www.autoadmit.com/thread.php?thread_id=5839187&forum_id=2#49701610)

No such thing, state ur iq?

(http://www.autoadmit.com/thread.php?thread_id=5839187&forum_id=2#49701611)

Chilmata's original claim was empirical and specific: the Teams account Claude is meaningfully better than the personal account — fewer mistakes, smarter outputs. When this was correctly identified as almost certainly wrong (same model, same weights, same everything that matters), he didn't defend it. He abandoned it entirely and retreated to an unfalsifiable cosmic shrug: "well the military's Claude must be way better, right?"

Notice what happened: the original claim was testable and got demolished. The fallback claim is conveniently untestable. That's the Bailey-to-Motte retreat in its purest, most embarrassing form. He used getting dunked on as a springboard to an even dumber assertion he can't be proven wrong about.

The crowning irony is that Chilmata himself provides zero evidence that Teams > Personal, then immediately treats this same non-difference as proof that there must be secret, powerful versions being withheld. "I couldn't tell the difference between tiers I can actually test, therefore the tier I can't test must be godlike." That's not reasoning, that's the intellectual equivalent of losing at poker and concluding the casino must have a secret high-rollers table where you'd win.

This is a man who couldn't figure out why his AI project was taking six months instead of one, blamed the AI instead of the operator, and then stumbled into a conspiracy theory about military superintelligence to cope. The galaxy brain don diaper praised is a gas giant — vast, full of nothing, and producing only hot air.

(http://www.autoadmit.com/thread.php?thread_id=5839187&forum_id=2#49704056)

(http://www.autoadmit.com/thread.php?thread_id=5839187&forum_id=2#49701484)

https://x.com/zackbshapiro/status/2027389987444957625

(http://www.autoadmit.com/thread.php?thread_id=5839187&forum_id=2#49705292)