anywhichway
Joined 13 years ago
Joined 13 years ago
I would also recommend explaining the specs and doing a lot of your back and forth with a lower end model and set it to a higher end model only once the conversation history has all the context you feel the higher end model needs.
To borrow the same analogy from the article, image trying to savor a meal where someone else was deciding when you take each bite. Even at a slow pace, the rigidness of the pace and your lack of fine control would still pose a problem with giving each bite it's rightful consideration.
That being said I love audio books and think I would struggle to apply this article's advice in my own life. Slowing down your audiobook is still a step in that direction, though I sometimes find that slowing it down can cause my mind to wander and my comprehension goes down and not up.
It's the necessities that people will continue to buy (or at least replace with close substitutes), regardless of what happens to the price.
Obviously, in this case it worked out much differently, but no, in general you can't say the wealthy people don't respond to price changes due to their wealth.
I agree there is a scarier potential there. And also some do, on occasion, escape their context (mostly credit score). They also have bigger contexts, but not so big that I would jump to the Chinese social credit comparison.
I looked it up, and it was a 3 pence tax per pound. When tea was selling for 2 to 3 pence per pound. So yeah, a 100-150% tax combined with the fact that the East India Company was allowed to sell without paying the tax. That is very unjust and threatens their business a lot more than the tax alone.
While a lot of people have used various methods to try to gauge the strength of various AI models, one of my favorites is this time horizon analysis [1] which took coding tasks of various lengths and looked at how long it takes to humans to complete those tasks and compared that to chance that the AI would successfully complete the task. Then they looked at various threshholds to see how long of tasks an AI could generally complete with a certain percent threshold. They found the length of a task that AI is able to complete with a various threshholds is doubling about every 7 months.
The reason I found this to be an interesting approach is both because AI seems to struggling with coding tasks as the problem grows in complexity and also because being able to give it more complex tasks is an important metric both for coding tasks or more generally just asking AIs to act as independent agents. In my experience increasing the complexity of a problem has a much larger performance falloff for AI than in humans where the task would just take longer, so this approach makes a lot of intuitive sense to me.
I'm not sure this would fit the definition of a product safety defect.
You don't expect Microsoft or Adobe to issue fixes any time someone finds a remote exploit that let's attackers gain control of you system though security issue in their software? I 100% expect this of my software vendors even for this purchase in the past. The expectations for software and hardware are certainly very different, but even for hardware we have laws that force companies to fix their hardware in some situations.
Both "prompt canarying" and "decoy system prompts" give 0 hits on google. Those aren't real things.
The issue here is only that he deleted it after they started an investigation.
The legal eagle video in the post says this explicitly that he would've been fine if he had wiped his phone prior to the flight.