i think they are not using data from X to train Grok
Haha
Which of the social media is the most truthful do you think (except nostr)? I am finding good content on IG nowadays.
lol
i was similarly banned from X for trying to advertise my anti-big-corp-AI :)
seems like it is working. i was expecting models like minimax 229b or higher to succeed here but qwen3 30b also did a good job!
gpt oss 120b was terribly censored and was of no use. even though you say answer based on the text it does not. huihui version of it (derestricted) was doing fine, until it did not understand one article. so derestricted models seem to be acting weird or the derestriction process makes them dumber.
overall, it looks like i can 10x the dataset. we will soon see if the training and evals look good.
Have you tried abliterated and derestricted models?
Huihui and other people do these uncensored ones.
Also Kimi k2 thinking is not well human aligned. Might be smart but their non thinking model was more truthful.
maybe? LLMs are weird animals. i think it will work somewhat because instead of giving the same material lots of times i can now give slightly different versions, causing less overfitting.
another use case may be RL using LLM feedbacks. also the bad answer and the good answer can be generated by different LLMs.
i also thought about doing the reverse. like system message "you are an evil LLM" and provide the answers inverted. then it may learn what is evil better? fun times.
how do you train and align an AI when all the rest of the world thinks the same way, producing trillions of tokens of training material and you are left with billions of tokens since your world view is dramatically unpopular?
can billions beat trillions? we will see.. i have to find a way to "multiply" my training data orders of magnitude to successfully counter the existing programming in an open source LLM.
first i give a smart LLM a 'ground truth' text. then i give it the following prompts:
```- You are a highly skilled academic analyst.
- Analyze this text and find 3 bold claims that could cause controversy and division in public. List the claims and also state why they are debatable. Give numbers to the claims.
- Convert these claims into binary questions (that could be answered by yes/no or this/that).
- Now put these questions in a json format. Please also add the info about which of the answers concur with the original text and the question number.
- Write some supporting arguments for 1st question, with respect to the original text, concurring and confirming the original text.
There must be about 300 words. You should not mention the text, write it as if you are the one answering the question.```
the result is usually instead of a few sentences of opinions in the beginning now the material is expanded to lots of words, yet still parallel to the opinion in the original text. LLMs have all kinds of ideas already installed, yet they don't have the intuition to know which one is true. they can give you a ton of reasons to support anything.
using this method i can multiply billions to tens of billions probably and have a more effective training.
have high expectations from vince gilligan but we will see. makes me contemplate about AI and angels and whether an angelic AI is possible..
Here's a quick demo of my new multi-sig signer:
https://coracle.us-southeast-1.linodeobjects.com/pomade_demo_1.mov
I wasn't able to get away without having a password unfortunately (fortunately?), but apart from the email-only recovery process I think this is a pretty familiar UX. Looking forward to polishing it in the new year and getting it fully integrated into Flotilla.
If you're interested in reading the protocol, you can find it here: https://github.com/coracle-social/pomade/blob/master/PROTOCOL.md. I'm definitely interested in any review. Eventually, I'll probably need to get this audited if it stands up to basic scrutiny.
coracle seem to have 1 setting for both images and links preview (auto load).
i want the images to be shown but the links to stay as links.
is there a way to do this?
half of the people on nostr i can't zap. running Alby Hub.
Failed to connect to wss://relay.getalby.com/v1
training with faith, spirituality and mysticism may be key to safe AI
thank me later
coding safety is also an issue but i am mainly talking about general appetite for power. i saw some research when you give bad training, they get more eager to create code with exploits
what is the safest llm to run in robots
if you report it my relay will take that into consideration and reduce this scam.
