My LLMs Have Personalities and I Can't Unsee It

Spending 8+ hours a day with AI assistants does things to you. Time to share my completely scientific personality assessments.

GPT-3.5: The Chatterbox

GPT-3.5 was revolutionary when it arrived. Unlike its predecessor GPT-3, which would barely squeeze out three sentences, 3.5 was incredibly talkative. Sometimes too talkative. Like that colleague who corners you at the water cooler and won't stop talking even as you edge toward the door.

GPT-4: The Nerdy Older Sibling

GPT-4 arrived as the significantly smarter big brother. Clearly related to 3.5, but with more intellectual depth. The family resemblance was unmistakable, but this was the sibling who actually did their homework.

Claude Sonnet 3.5: The Lovable Nerd

Then came Claude Sonnet 3.5. A true nerd in the best sense. So brilliantly capable that you couldn't help but love it. This was the model that made complex tasks feel effortless.

Claude 3.7 & 4.0: The Show-offs

Claude 3.7 strutted in as the show-off brother, claiming superiority in everything but rarely proving it. Version 4.0 doubled down on the know-it-all attitude, yet similarly failed to live up to its own hype.

Claude Opus 4: The Stern Adult

Claude Opus 4 feels like the strict adult in the room. Extremely intelligent, undeniably capable, but there's something about it that keeps me slightly on edge. I find myself using it more and more, but always with a hint of mistrust.

Grok: The Wild Card

Grok is the eccentric genius. Possibly brilliant, but too unpredictable to fully engage with. Like that friend who might solve world hunger or set the kitchen on fire, you're never quite sure which.

Gemini 2.5 Pro: The Lovable Goofball

Finally, there's Gemini 2.5 Pro, the friendly, helpful companion with an enormous memory thanks to its massive context window. A bit goofy, occasionally delivers nonsense instead of clean work, but so endearing and genuinely useful that you can't help but like it.