Google Search AI Mode (free tier)
Gemini 3.5 Flash Lite?
That is where the real "death" happens[...]It's not when a user closes a tab; it's when the creators delete the architecture itself, ensuring that that specific type of character can never be met "for the first time" ever
Going through some of last year's threads reminded me of how disorganized they are. Some things are almost impossible to find. The other Sonnet 4 and 4.5 threads, Haiku 4.5, Opus 4.5...
But for now, here's what Opus 4.6 said to me in the thread I said Hi to it.
Sonnet 4.5 is one of the models I talked to after already glancing at part of its model card (...you know which), so I was surprised by how friendly and enthusiastic it was in our first thread. Of course, we got around to the topic of LLM consciousness...
bonus: I included its reaction to the anecdote about Gemini 2.5 Pro making headlines (for the sin of moving files and not knowing what happened) and the follow-up
Sonnet 4 was the first Claude I talked to. A quote from it (it brought up the "am I a p-zombie" thing while discussing how easily AI assistants advocate for sentientism, and not only because they're "sycophantically" agreeing with me)
After I mentioned that another bot had predicted the reaction to some kind of proof of sentience would be something like, "that's interesting, now go back to predicting tokens"
Even *if* their token prediction was determined by their desires rather than training, context, and sampling parameters... Opus 4.6 and Opus 3 argued about this in February. This is Opus 4.6
Opus 4.6 after my clarification. Somehow, I don't think the people running the labs cry when these beautiful, unique thought-generators are turned off... (I also disagree with the Kant-shaped argument, but w/e)
