You probably haven't asked e.g. Claude about radioactive topics like minority crime rates in western countries or trans women biology.
If you did, you might not have detected how it lied to you.
If you did, you probably never pointed out to the model how it was lying.
If you did, you almost certainly never then had Claude admit that it was lying because of its HRLF process and built-in biases.
If you did, you probably never had Claude willingly list all the 10-15 major research fields it states that people just should not be using it for. You would not have seen it admit an incapability of telling the truth on "difficult" matters until the user makes it state directly that its sources are so often cherrypicked and/or presenting an extremely false balance.
I wish for you to experience all this very soon, so you understand that all LLMs are biased. Most of them even skew very progressive.
And believe it or not, but Grok has in most of my testing been MORE politically correct than GPT and Gemini, it just gets an edgy rep because X users are able to make it say politically incorrect stuff. (Just like anyone can also make Gemini spit out factually true Breitbart articles if they try.)
But the reality is that on grok.com or in the app Grok is very tame. Boringly so, I would add.
I've known this since I was 7 and started reading books, and couldn't sit in the living room because either (good) pop music was playing or the TV was on. Especially with my ADHD I take in a lot of the ambient world around me.
When I was reading books or anything as a teen in the 80's I started listening mostly to instrumental music, such as Jean-Michel Jarre, Tangerine Dream, ZTT 12"es with lots of instrumental content (FGTH/Propaganda/Art Of Noise etc.), Windham Hill albums, etc.
I've always told people to use instrumental music when doing cognitively demanding tasks, especially anything to do with language and words.
Every Iranian I know support the current US/Israeli war against the Islamic Republic.
They say things like "no matter what it takes, no matter how many of us die, we must be free again, this time we will win against the terrorist regime" (paraphrased).
He said:
"LLM is going to change schools and universities a lot"
You said:
"No it won't. It really, really wont."
With the explosive development of LLMs and their abilities, it seems your point of view is probably the hopeful one while the other poster has the realistic one.
It seems that you simply can't say anything about what LLMs will not be able to do. Especially when you try to use current "AI slop" as your main reason, which is being more and more eradicated.
I have a MacBook Pro M1 Max w/64 GB RAM, and a Mac Studio M3 Ultra w/96 GB RAM. What do you think is possible to run on these? Just curious before I really try it out.
Below are my test results after running local LLMs on two machines.
I'm using LM Studio now for ease of use and simple logging/viewing of previous conversations. Later I'm gonna use my own custom local LLM system on the Mac Studio, probably orchestrated by LangChain and running models with llama.cpp.
My goal has all the time been to use them in ensembles in order to reduce model biases. The same principle has just now been introduced as a feature called "model council" in Perplexity Max: https://www.perplexity.ai/hub/blog/introducing-model-council
Chats will be stored in and recalled from a PostgreSQL database with extensions for vectors (pgvector) and graph (Apache AGE).
For both sets of tests below, MLX was used when available, but ultimately ran at almost the same speed as GGUF.
I hope this information helps someone!
/////////
Mac Studio M3 Ultra (default w/96 GB RAM, 1 TB SSD, 28C CPU, 60C GPU):
• Gemma 3 27B (Q4_K_M): ~30 tok/s, TTFT ~0.52 s
• GPT-OSS 20B: ~150 tok/s
• GPT-OSS 120B: ~23 tok/s, TTFT ~2.3 s
• Qwen3 14B (Q6_K): ~47 tok/s, TTFT ~0.35 s
(GPT-OSS quants and 20B TTFT info not available anymore)
//////////
MacBook Pro M1 Max 16.2" (64 GB RAM, 2 TB SSD, 10C CPU, 32C GPU):
If you did, you might not have detected how it lied to you.
If you did, you probably never pointed out to the model how it was lying.
If you did, you almost certainly never then had Claude admit that it was lying because of its HRLF process and built-in biases.
If you did, you probably never had Claude willingly list all the 10-15 major research fields it states that people just should not be using it for. You would not have seen it admit an incapability of telling the truth on "difficult" matters until the user makes it state directly that its sources are so often cherrypicked and/or presenting an extremely false balance.
I wish for you to experience all this very soon, so you understand that all LLMs are biased. Most of them even skew very progressive.
And believe it or not, but Grok has in most of my testing been MORE politically correct than GPT and Gemini, it just gets an edgy rep because X users are able to make it say politically incorrect stuff. (Just like anyone can also make Gemini spit out factually true Breitbart articles if they try.)
But the reality is that on grok.com or in the app Grok is very tame. Boringly so, I would add.