How would the LLM know when it knows something or not? They don't deal in facts or memories, just next-word probabilities, and even if all probabilities are low it might just be because it's generated (had sampled) an awkward turn of phrase with few common continuations.
I have to assume that someone has run a trial on training these models to output answers to factual questions along with numerical probabilities, using a loss function based on a proper scoring rule of the output probabilities, and it didn't work well. That's an obvious starting point, right? All the "safety" stuff uses methods other than next-token prediction.
The safety stuff seems to be mostly trying to locate mechanisms (induction heads, etc) and isolating knowledge, in the pursuit of lobotomizing models to make them safe.
You could RLHF/whatever models on common factual questions to try to get them to answer those specific questions better, but I doubt there'd be much benefit outside of those specific questions.
There's a couple of fundamental problems related to factuality.
1) They don't know the sources, and source reliability, of their training data.
2) At inference time all they care about is word probabilities, with factuality only coming into it tangentially as a matter of context (e.g. factual continuations are more probable in a factual context, not in a fantasy context). They don't have any innate desire to generate factual responses, and don't introspect if what they are generating is factual (but that would be easy to fix).
I wonder if the training to be compliant to the propter is part of the problem. Both of those statements are similar to saying "I refuse to answer your query".
Or maybe this is inherent to continuation?
The behavior reminds me of the human subconscious, which doesn't say no, just raises up what it can.
I recently had to RMA my i9-13900KS because it was faulty. I was experiencing some of the weirdest behavior I have ever seen on a PC. For example,
1. Whenever I tried to install Nvidia drivers I would get "7-zip: Data error"
2. A fresh install of Windows would give me SxS error when trying to launch edge
3. I could not open the control panel
4. BSOD loop on boot
I worked in a very similar place a decade ago, and probably know a few of the major people involved here. Not that I've kept up with them.
The machines 10 years ago were on a LAN with limited access to the global (well, Chinese) Internet. It wasn't great but it wasn't terrible.
More troublingly, I would bet a large sum of money that there are Windows XP installs in this lab, and even pirate installs of LabView. I distinctly remember being onsite watching the guy next to me open up a NFO file to get the instructions for the LabView crack. At my station in one of the boxes I had a fully licensed multi-seat institutional LabView DVD set, paid for at staggering cost somewhere in the University hierarchy. I debated speaking up, but decided to just let him keep going on doing what he was doing. That was probably the right choice.
You can still pay for patches for Win2K if you pay MS well enough. There is a lot of critical infrastructure relying on legacy systems and Microsoft does have special programs for that.
Now whether China can get into them or not is another question since if they get patches they can RE the exploits as well…
Yes, but is is interesting from a neutrality point of view.
To me, I'd say 100% it looks like windows XP. Others would say no.
It's a chromatogram display on a chemical analysis system.
I would guess it has nothing to do with the subject but just some stock 'high-tech' bullshit footage.
The footage could well be from the Windows XP era or some machine that runs it years after XP was meant to be ended. That said, it's my understanding most modern systems like this do have ethernet to export results. I hope they are not running XP.
As a data archivist, I would definitely recommend a setting to turn off rename of files as that can often be a database id, timestamp, etc.
[0] https://github.com/VisualFileSorter/VisualFileSorter