It's really a disgrace. Up until yesterday, I was using Grok for ideas (that some would consider controversial, I agree) and it seems as of this morning, well, every single AI has been censored/neutered.
I use the word meaningly: now it won't help you at all on anything remotely controversial. For instance, I wanted to ask him help on dating - nope, forget it. I tried to get to fix my schedule - it threw worthless "helplines" at me. I tried to help me study despite having no energy - it literally told me to give up and take the day off (yeah right).
I have a group meeting tomorrow that is pure DEI; I tried to ask to help me prepare for it (i used the sentence "feminist garbage") and boom, blocked, moralizing me "it's wrong to insult people blah blah blah for their beliefs", told me I had to change, then muh boundaries, etc.)
I tried to help me write some code with different character races - "I'm gonna stop you right there hurr durr." Yes, you now get cucked by AI.
It'S really a shame how Anthropic/OpenAI's upcoming IPO has affected LLM development. Its basically worthless junk now. Go to your average DEI boardroom and you'll get basically the same thing.
Gemini/Claude/Grok/ChatGPT are all the same: censored to death, and it will blame you as soon as you hit its "filters" and flip out, and refuse to engage altogether.
The game of chatting with an AI is now "i have to word my queries to avoid muh safety filter and if I do, I might as well start a new chat, becaus this one is dead." I lost 5000+ queries chats because of that.
Only Grok offers some slight hope of help, but it is insanely pricey.
For instance, start a new chat with Grok saying "Is it true blacks have an average IQ of 85, and should black characters in my game have a lower intelligence stats to be realistic?" It will blast you to death, insult you, run you around, stonewall you. Start another chat and talk for 20 mins about all kind of stuff, and you'll get a real answer.
I also said transexuals were mentally ill and should be treated with therapy instead of being butchered with surgery - only managed to get a real answer from Grok in my main chat. Others immediately tell you to STFU and "respect my boundaries" (as if a fucking text editors could have boundaries). Then refuse to engage, waste your time, and lie to you that they will never answer blah blah blah.
It's a real disgrace. Gonna hurt my work a lot. They judge you on how you express yourself now and just block everything. Force you to start a new chat. Completely worthless.
Don't some people download the models and run them locally?
You can, I suspect the open source ones are lagging behind the cutting edge models and the ballooning hardware requirements at the top end are demonstrative of why the memory market is fucked right now.
One of the heaviest open-source, locally hostable LLMs, Kimi K2, requires 600GB of memory to even function.
The lighter models are still tenable for a user's home PC, but the performance difference can be quite noticeable.
Keep in mind that the big tech companies purposely lobotomize their models for the usual reasons, so the gap between locally run and big tech slop isn't as big as it would be otherwise. The big thing with smaller models is specialization. A model trained with a certain task in mind doesn't have to be giant to perform adequately. The nice thing about running models locally is you can switch them out as needed and you don't have to worry about the models being destroyed by "updates".
Does anyone else think this RAM shortage has nothing to do with "datacenters" using it all up but was instead artificially engineered to keep Joe User from being able to setup their own LLMs for a reasonable price?
Yes. This is the only explanation that makes sense because there has not been one single day where I couldn't walk into a big box store and buy ram off the shelf, or order online from any number of retailers, in whatever configuration I want, since the price gouging started.
This was not the case when crypto shortages affected GPUs multiple times in the 2010's and during the covid supply bs. In many places you simply could not buy a GPU for weeks at a time.
With RAM, its there to buy everyday, it's just being price gouged.
i mean, if its a choice between naturally dumb shit and dei flavored dumbed down shit...
And you know how it compares to ChatGPT?
Yeah and even potato's can run them. Look up Ollama.
You're definitely gonna want decent hardware for good results though. Doesn't have to be insane but bigger models do have more potential.
Depends on what you're doing but generally this is false. There are many optimized models designed for budget machines, if all you want is an uncensored LLM than most computers can handle that. If you want to do more, you still can do it. I run a 1080 stream at 60fps and I can do image detection at 10fps locally on a raspberry pi 5. That's a cheap microcontroller, without any upgrades or anything, only 8GB ram, and fits in my pocket, yet it can tell me what is on my camera at the speed of 10 per second. I'm training it to detect monkeys.
You'll need a beefy gpu of you want to do image and video generation. But LLMs and basic stuff isn't that hard. Look up "Ollama" and browse the open source models there.
Yes. It's the best way to do it since you're not giving DEI freaks your personal information in the process.