Replies

My understanding is that LLMs are not deterministic. To oversimplify, they basically rank potential answers based on the probability that they are correct. But instead of always choosing the top one, they will sometimes choose another one near the top. A parameter known as “temperature” adjusts how willing it is deviate from the top answer.

For example, let’s say the top 3 answers are ranked 91%, 90%, and 89% likely to be correct. Maybe they are all correct but just worded differently. The LLM would appear less robotic if didn’t always choose the 91% answer. And say the next answer is a more ambiguous 78%. If the LLM occasionally spits it out, and customer responses are used to train the model, it could upvote or downvote the 78% answer and “learn” whether it’s correct.

1 user thanked author for this post.

Cybertooth

https://www.cnet.com/tech/tech-industry/gates-security-is-top-priority/

“When we face a choice between adding features and resolving security issues, we need to choose security”

– Bill Gates, 2002

2 users thanked author for this post.

Alex5723, Cybertooth

I don’t know why I sometimes have to stop and think about whether 1 = maximum. I read/watched enough DEFCON-filled books/movies as a kid to know better. It’s actually not that unusual – top 10 lists, priority #1, first place finish, etc.

Sorry to beat the dead horse but you got me wondering and I had to find out…

https://en.wikipedia.org/wiki/HLT_(x86_instruction)#Process

Almost every modern processor instruction set includes an instruction or sleep mode which halts the processor until more work needs to be done. In interrupt-driven processors, this instruction halts the CPU until an external interrupt is received. On most architectures, executing such an instruction allows the processor to significantly reduce its power usage and heat output, which is why it is commonly used instead of busy waiting for sleeping and idling.

(Hopefully I’m not duplicating something in earlier replies that are awaiting moderation.)

Thanks for an interesting article. I may be missing something (happens more often than I’d like), but I think there’s a problem with your proposed sleep logic.

When a computer is sleeping, instead of a programmed loop, the CPU is actually stopped until a hardware interrupt wakes it up. A computer would have to have a “low battery” interrupt. Of course, it wouldn’t be 100% reliable since the battery, especially a poorly maintained one, might not have enough juice left at that point to do what needs to be done.

It seems like the solution really depends on the cause. Maybe the persisted state is bad to begin with when entering sleep mode, or glitches out and becomes bad when power is lost. Conceivably it could be manufacturer-dependent and not something Windows can fix generically.

—-

The link below describes several different sleep states that Windows recognizes, a subset of which are supported by a given computer. Some of those states do, in fact, try to account for the possibility of losing power while sleeping.

https://learn.microsoft.com/en-us/windows/win32/power/system-power-states?redirectedfrom=MSDN

The next link says, “With each successive sleep state, from S1 to S4, more of the computer is shut down. All ACPI-compliant computers shut off their processor clocks at S1” and “Details of the intermediate sleep states can vary depending on how the manufacturer has designed the machine.”

https://learn.microsoft.com/en-us/windows-hardware/drivers/kernel/system-sleeping-states?redirectedfrom=MSDN