Or just use a lower power setting, which is either an on/off duty cycle of a couple of seconds, or in the fancier models, actual variable power.
Bonus: you don't fry the shit out some of your food while other spots are cold.
I usually do a brief period on high if something came from the fridge, and then drop to a lower power level. On my microwave "programming" this is actually pretty easy; pushing the power button after entering time allows you to set a second power level and time.
Depends on the product you're heating though. I think it's certain fats that make it sometimes start to pop after literally a handful of seconds, and other times you can go at the highest power (I think that's 800W for me) until the food is done after five minutes and all you'll get is some steam. I've tried to use lower power settings for "normal" products that have splatter but not straight away, but that just doesn't get it all the way hot without starting to get splatter after all. Using a cover, like an inverted plate, is the way to go
I do agree about the part where you mention frying part of the product while the rest is still subzero. Power reduction is useful, for me it just doesn't adequately prevent splatter
> You're throwing shit at the wall and hoping it sticks.
A more charitable description might be "You're employing the scientific method to extract value from GPT-like systems." Just like in science, with time you're developing intuition for how the underlying system works, but you still have to run the experiments.
Google erased more than $100B in shareholder value with a single botched demo. I imagine events like this will only further increase their risk aversion.
I have sympathy to both sides of this problem because the truth is, ALL the GPT do this, and what google botched was the message. I don't understand why it wasn't a similar share price plummet in Bing, OpenAI when the lies sold as truth surfaced there.
In part, and I know this is 'blame the victim' I think shareholders truly are dumb, if they thought it was AGI and some god given uplift to the cloud-mind: it was never going to do what they wanted, and rather than 'Google erased' I'd say google communications/marketing erased: they really didn't prepare this package for release well.
Too much infeasible community expectations on something which was incapable of doing what they wanted.
No matter what the $100b is gone. I'd say the buyers at the bottom picked up a bargain btw.
> I don't understand why it wasn't a similar share price plummet in Bing, OpenAI when the lies sold as truth surfaced there.
Because Bing isn't a material profit generator for Microsoft right now, AI search is a lottery ticket for Microsoft while Google is playing defense for their main revenue stream.
> I don't understand why it wasn't a similar share price plummet in Bing, OpenAI when the lies sold as truth surfaced there.
Maybe most consumers ultimately core more about the improved search capacity of Bing than that it might say weird or offensive stuff in some prompt-engineered worse cases. Plus it doesn't need to be 100% accurate, just no less accurate than the average person's SEO-laden Google search results.
He was obligated to write 4.6 songs for his old label, and half-assed them (though, knowing the back story makes me enjoy this one a lot). It seems he could save some effort these days.
I didn't spot any factual errors, but I did notice that NYT uses Intel's titles for processes that aren't used widely by other silicon design teams:
* "Validation" as used by Intel to describe what other groups would refer to "silicon validation" or "product validation". Intel also confusingly uses "validation" as a term for pre-silicon validation, which I've seen more commonly referred to as "digital verification" or "DV" where RTL is executed in a simulator and the results compared against tests. I also don't recall seeing them mention "emulation" (where RTL is loaded onto massive FPGAs and exercised at higher speeds than what the simulator can accomplish).
* "tape-in" is what the factory does after processing the data files sent by the design teams. Most design teams refer to it as "tape-out."
I remember there was another Intel-ism, but it didn't stick with me.
