> Safety teams within the company pushed to slow things down. These teams worked to refine ChatGPT to refuse certain types of abusive requests and to respond to other queries with more appropriate answers.
I wonder what this struggle means for the future of ChatGPT censorship/safety.
> Going further, the researchers removed visual neurons from the box jellyfish and studied them in a dish. The cells were shown striped images while receiving a small electrical pulse to represent collision. Within about five minutes, the cells started sending the signal that would cause a whole box jellyfish to turn around.
> “It’s amazing to see how fast they learn,” said Jan Bielecki a postdoctoral researcher at the Institute of Physiology at Kiel University in Germany, also an author of the paper.
Could someone please explain this bombshell?
Did visual neurons learn all by themselves in a dish? And how did the researchers know that the visual neurons would interpret "a small electrical pulse" as a collision? (I'm surprised visual neurons "know" what a collision is.)
> interpret "a small electrical pulse" as a collision
They don't interpret as a collision. IIUC current models of instinctive behaviors and learning are built up from the assumptions that
* if a source of event is "beneficial" (from a chemical, physical perspective, eg "food" / energy packet is there) then those cells, neurons, microorganisms are naturally selected which tend to approach / follow / capture that source
* if a source of event is non beneficial (threat / chemical incompatibility etc) then those survive which avoid that source.
Here it doesn't matter that the negative event is a "collision", eventually the organism('s building blocks) learn to avoid it.
This sounds strategically sound until the realization that you've given very compelling justification—legal or not, it doesn't really matter—for others to now commit preemptive violence against people like you.
No part of the post you’re replying to suggests it’s opening the door to reciprocal violence.
The more complete argument is that extra-judicial action is justified when there is no other way - such as when the political process cannot effect change.
If there are no just elections, the only way to remove a ruler is revolution, almost by definition. In that case, “punch the Nazis” all you like.
No part of the post you’re replying to suggests they want to pursue or support tyranny. If some right-wing kooks justify revolution with demonstrably false statements, that does not justify their actions - falsehoods are never a foundation of justice.
> But free loader molecules that could infiltrate a metabolism and contribute nothing, but use it only for their own self-replication would also thrive.
Could you please explain the "mechanical" distinction between molecules that contribute to self-replication and molecules that contribute nothing during replication?
somewhere on the internet are some videos in which an italian scientist describes what he means by a system. I think that's probably enough to google.
So if you're willing to accept that sometimes we are able to recognise a phenomenon that fits our definition of a category, even though we are not capable of explicitly defining that category, then we could say that a cell fits out intuitive definition of a system, even though I think despite being employed as system achitect, I'm not sure I could tell you what a system is. I like the attempts at definitions of living systems that discuss phenomenon that enable the persistence of information that are robust against the entropy.
Anywhy on to my attempt at answering your question:
If we could identify a set of molecules in the pores of vent rock back in the early pre-biotic earth, each with a distinct catalytic function, that were the using energy available, perhaps as electric charge, perhaps chemical energy and interacting to replicate that same set of molecules in neighbouring pores, then we could look at the actions of each molecule, and determine how each molecule participates in the chemical processes that result in the replication of the whole set of molecules.
If a molecule is mutually dependent on on other molecules in the set for their mutual replication, I think we can call them a contributing part of the system.
If a molecule depends on the other molecules to be replicated, but the replication of no other molecule is improved from its presence in the set, then it is not a contributing part of the system, but a freeloader.
I think this sets up two maximal strategies:
A system is most efficient if it has no freeloaders.
You might not agree with the spirit of this an endeavor, but I have a yes/no question for you:
If you were to wear your most clever, most creative writing cap, could you make a convincing case entirely contrary to your beliefs? I'm not asking whether you could write a convincing case against racial affirmative action, because I know you could handle that just fine.
Instead, could write a convincing case that the group you're talking about owes some collective debt to the rest of society, rather than the other way around?
I used to be a dyed-in-the-wool conservative. Their entire case is "it's not fair" in the absolute sense. That because some white folks descend from people who didn't have anything to do with slavery that all should be absolved from participating and benefitting from systemic racism.
It's not an intellectual argument. It's an argument from performative and wanton ignorance.
For sure. My ancestors mostly came to the US well after slavery. The only one that I know didn't fought on the Union side in the civil war. But that doesn't matter, because in my life I've received the benefits of being white in America. And I think a lot of that performative and wanton ignorance is basically refusal to admit that sort of privilege born of luck.
> One, when non-white, non-men raised the alarm about LLMs previously, they got much less media coverage than Hinton, et al, are getting.
Mainstream (e,g. CNN, BBC) and mainstream-adjacent (e.g. Vice, Vox) journalists have spent years pushing the "AI will harm POC" framing. AI companies are endlessly required to address this specific topic—both in their products and in their interaction with journalists alike.
Dr. Hinton is getting a lot of coverage right now, but this is the exception, not the rule.
I am extremely not Mormon, but we don't have any reason to believe Mormons are more likely to be sexual abusers than Jews, Muslims, or Hindus.
It's probably more politically safe to claim Mormons are more likely to be sexual abusers, but I wouldn't feel any less safe leaving my children with a Mormon family than with a Jewish family, a Muslim family, or a Hindu family.
If the intent is to say a business should not be able ban a protected class per the Civil
rights Act, my comment was intended to be interpreted within the context of businesses banning non protected classes.
The discussion would then be about if we legally protect members of tribe with certain gender, skin color, cultural affiliation, etc, why would we not protect a tribe of certain political views, that is an interesting question that I am not sure how to delineate.
I wonder what this struggle means for the future of ChatGPT censorship/safety.