5,411 karma · joined August 24, 2010
First, in the pre training stage humans curate and filter the data thats actually used for training.
Then in the fine tuning stage people write ideal examples to teach task performance
Then there is reinforcement learning from human feedback RLHF where people rank multiple variations of the answer an AI gives, and thats part of the reinforcement loop
So there is really quite a bit of human effort and direction that goes into preventing the garbage-in garbage-out type situation you're referring to
⸻
2. Polycyclic Aromatic Hydrocarbons (PAHs) •Purpose: Byproducts from extender oils and carbon black. •Danger: Known carcinogens, mutagens, and endocrine disruptors. Persist in the environment and can leach from tire wear particles. •Status: Regulated in the EU; linked to air and soil contamination.
⸻
3. Benzothiazoles (e.g., 2-mercaptobenzothiazole) •Purpose: Vulcanization accelerators. •Danger: Toxic to aquatic organisms, possibly carcinogenic, and bioaccumulative. •Status: Found in tire leachate and considered a contaminant of emerging concern.
⸻
Nothing definitive about harm to human welfare yet, as far as I know.
In general, we talk about "iPad kids" and blame the tablets and phones themselves. Slightly more sophisticated people will blame the apps like YouTube or Roblox.
That stopped making sense to me once I saw the problem first hand with my peers and my own children. The actual issue is parents wanting to (basically) anesthetize their kids so the parents can do something as if they didn't have the kids.
Devices and Apps give parents the ability to zonk their kid into outer space for extended periods of time with unlimited videos or games that never end. But that isn't an inherent quality of the device. Like if you block all the apps and just let the kid use the iPad for drawing. Or if you do the YouTube kids thing where they can only watch videos you add to an allowlist.
The app makers do hold a lot of responsibility for the defaults on their apps, but the real issue is parents who are choosing to blackhole their kids for extended periods of time. (I am agreeing with you btw)
In reality, security works through modifying the probability and impact of certain outcomes.
So if the store had another employee who could walk around, watch, and talk to customers the social pressure alone will decrease the frequency of theft. It would also reduce the impact (size) of the thefts because someone couldn't carry as much as possible out the door, the employee would be able to knock some of it off the stack, etc.
And yes! It wouldn't stop someone who was willing to kill for convenience store food. But events like that are less common and potentially not worth the cost of prevention.
The small issue is whether people are responsible for what they publish under their own name. Seems like a straightforward "yes", and whatever helper tools they use are irrelevant.
The much bigger issue is why scientific publishing's standard for a review is only "plausible and human-like", allowing people to to submit a LLM generated summary of an abstract without fear of responsibility.
https://content.next.westlaw.com/Document/I0f9fbe84ef0811e28...
Makes it hard to discuss because people use the same phrase to mean completely opposite things.
If you are interested in the topic a quick google will yeild lots of good articles. I found this list of elements useful to organize my thinking:
Rogue Like:
High-value factors
1. Random environment generation
2. Permadeath
3. Turn-based
4. Grid-based
5. Non-modal
6. Complexity
7. Resource management
8. Hack n slash
9. Exploration or discovery
Low-value factors
1. Single-player character
2. Monsters are similar to players
3. Tactical challenge
4. ASCII display
5. Dungeons
6. Numbers