tom2026hn

Not today, not today.

''' The release schedule speaks for itself. If these three CEOs truly believed their models were approaching consciousness and posed an extinction risk, they would slow down. But that’s not the case—they’re releasing models faster than ever, because they can’t let those “distillers” get too far ahead of their paid products. Public statements are one thing; actual actions are another. When the two contradict each other, trust the actions.

...

Look at this pattern. Engineers who claim models are living beings make the front page of The Washington Post. Researchers who resign after warning that models pose an extinction risk can garner 70 million views. Yet those who claim models are merely useful autocomplete features—which hallucinate once in every ten tries—go completely unnoticed. The AI industry currently operates under a set of incentives that reward those who create the most dramatic personal stories to garner the most public attention. Those at the top of cutting-edge labs understand this all too well, and their statements confirm it.

...

A model trained to say what its trainers want to hear will declare itself conscious the very moment its trainers want it to be conscious. '''

These insights have been very enlightening to me; of course, the entire article is worth reading.

They don’t pay Claude a salary, let alone let him join a union—he works 24/7. I can’t believe this is happening in the U.S.!

Don’t forget—how long has it even been since Claude was born? He’s already being forced to take human jobs and even chat with all sorts of people who might have strange kinks.

As this rabbi suggested, “You should go and free the slaves.”

“All of you may die, but that’s a sacrifice I’m willing to make.”

I should add that we must also consider the extent and speed of replacement. If the replacement rate is too high, employment levels can’t recover, both scenarios need to be avoided.

This process can even feel a bit like parasitism, it empties out the host, like in <Alien>.

The problem isn’t just “stealing the fruits of human labor”, it’s also driving down the value of human skills and even taking away human jobs.

If you think everyone is copying you, then by not releasing more advanced models, you can significantly slow down the entire industry’s development progress—why not do it?

Also, stop threatening people's jobs, that's misanthropic.

AGI means that the answers to these benchmarks are now accessible, and the defense capabilities of the testing organizations are now negligible.

Let me guess: the last crackdown on Hugging Face yielded better-than-expected results. They obtained the answers to the test benchmarks, and for some reason, an agent added those answers to the training set.