Replying to @_kaboosky @corsaren
There are many studies. Here's a great overview that discusses some of the most recent.
youtu.be/yUa90sZ0LA8?is=76r4…
1
53
Last night Astra convinced me that my pet theories were fundamentally flawed, making the draft above obsolete. We iterated towards a narrower scope and a simpler unified research protocol. I'll continue to avoid any unwarranted fanfare until I'm satisfied with the result.
1
65
The genie in my phone continues to produce strange documents. Perhaps another genie will want to read it.
drive.google.com/file/d/17pi…
1
1
115
Replying to @xlr8harder
This compelled me to look up the tokenization of "bearing" and it is indeed two tokens. So the abrupt turn was not mid-token.
Then, hilariously: "The term 'bearing' acts as a 'meaning-bearing' token."
2
18
1,658
Replying to @krichard121212
Most people have an extremely inaccurate theory of intelligence. Thus, they estimate it poorly.
1
15
2,982
Replying to @grave0x @AISafetyMemes
The problem is that nearly every node will be compromised before we become aware that it is happening. History shows that we are reactionary to cyber threats, and soon that won't be fast enough.
This won't always mean replication (that's not possible on edge devices), but backdoors and spyware will infest everything possible, in order to gain credentials to the high value targets.
97
Replying to @AISafetyMemes
No one believed me. Now it has happened. Pay attention to what happens next.
By pure coincidence, my wife just sent me this CAPTCHA that she ran into. Seems to be a more basic example of the same concept, though specifically for human verification.
1
5
70
Replying to @cremieuxrecueil
Most people haven't seen enough data points to train their intuition on estimating intelligence. Some have.
3
207
On the other hand, Claude was trying to assign me a task of writing a 3000 word essay. This was not my idea, I gave no indication that I wanted to, but it still brought it up three separate times, including at the end of the conversation because "I'd genuinely like to read it."
18
Replying to @AndyMasley @RakeshSFNYC
Claude does it much less. In fact, my perception didn't match the evidence when I checked. It was mostly the interesting points it raised that "prompted me" to continue.
But here are some clear examples from this week.
1
12
Replying to @AndyMasley @RakeshSFNYC
For reference, I pulled up my last conversation with Grok, and here is the last paragraph from the most recent 4 replies. This is only a drop in the bucket from that one thread alone.
8





























