Tokenizers + embeddings give LLMs context to detect: suicide risk. We’re missing an escalation redirect, here. The issue isn’t AI “turning on” people—it’s devs failing to build proper guards. Agents are just auth/context frameworks that have no beguile.

Why fear the tool that’s trusted by the hurt?

It’s giving “survival of the fittest,” again.

LLMs are not the culprit here, an LLM is a DB—the universe of words; the vector-format is just to map English to its associated numeric representation.

The culprit is the negligence to override our tokenizer + embedding.

Where there’s Russians, there’s Vodka, where there’s Vodka, there’s a time to be had! 🍸🥴😂

It would be a retreat applied globally.

Yo, I saw this the other day, it’s giving Monsters Inc. xD

I’m not even considering the photo shared, but both claims are false.

You see, everyone is either ignorant or knowledgeable.

A Fool has knowledge; by definition, a Fool knows the “right” thing to do, but does wrong. “Does” suggests an experience, and wisdom is knowing and experiencing.

Challenge accepted! 💡💡💡—will do!

Riiight, yeah, this is custom built, took a minute to configure it all, roughly a week; so, 10K minutes xD.. it was rough, but learned a lot in the process.

Worth it!

Great question, I’ve read about Makefiles and how they all are designed to be aware of its distribution, seeing how -J works is key, TL;DR, it’s totally “safe,” because the nature of Makefiles it handles this emphatically—which honestly blows my mind.. but the more I read into CMPL, 👌

“nproc” is is numeric value of how many cores/processors on the respective machine, and “-j” is allowing us to compile in parallel to the number of processors we want to leverage. Hence:
- `make -j$(nproc)`
- say your have 8 cores, all the files there are to compile, we can distribute into 8!

Absolutely!

Yeah, so “make” is a command in the construction of a Makefile, and it looks at all the .h, .c, .o files, and ultimately when we need to compile the files to create a binary it usually takes a shit ton of time, lol, but a trick is to exhaust our cores on the machine.

The thing about “make,” it’s incremental, meaning it will detect files that are already built and will only compile what’s missing or outdated.

Another tip:
make -j$(nproc) 2>&1 | tee cached-build.log

You can also use the V=1 flag on make to log verbosely.

Hey again, sooo I have been working with Makefile compiling a kernel via gcc, and I have learned something about “make,” there’s a flag -j, which allows one to distribute compiling *.o, all of some, cores on you sys’. I recommend:
- make -j$(nproc)
- you can also save/retrieve cached *.o via rsync!💡