deus ex machina explainable AI · King’s College London
Joined May 2016
- Tweets936
- Following1.1K
- Followers153
- Likes28.6K
simulation == god
The only reason this framing causes existential dread is because it denigrates the status of the “creator” to more human like
Since we “simulate” as well
It’s the precursor to effective acceleration, lands writing is morose but there’s beauty to the processes expressed
I agree that it was tremendous motivation and mission at onset, but they have been pigeonholed by that identity, this all started with tiny being needlessly rude to people for no good reason.
“Professionalisation” doesn’t have to mean transformation into a corpo culture.
I would expect of these freedom fighters to have enough protean agency to create unique brand identities rather than to revert to the mean of counterculture and interact with users as forum trolls.
Exactly, it’s devolved into a groupthink echo chamber where each one is posturing a position for validation of group identity
Tek is a legend, and there’s nothing wrong with the research, however the personality he has displayed during this event stinks
There are no issues with a Buddhist, or a Western, or a Jewish, or an Islamic fine tuned model. Open source models are meant to be moulded to the use case and domain.
The problem is with acting high and mighty, it was a sensitive topic and they should have reacted with the requisite sensitivity instead of mocking people who don’t agree with their politics.
Otherwise the proof that ideology can be moulded without loss in performance is a positive finding.
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.
Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.
After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.
The values in an open model are one fine-tune away.
Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete
They have brought this upon themselves, it’s less about the fine tune, and more about pushing your value judgements on others.
The proof that ideology can be moulded without loss in performance is a positive.
There are no issues with a Buddhist, or a Western, or a Jewish, or an Islamic fine tuned model. Open source models are meant to be moulded to the use case and domain.
The problem is with acting high and mighty, it was a sensitive topic and they should have reacted with the requisite sensitivity instead of mocking people who don’t agree with their politics.
There's a new version of this post
Good luck with your celebrity worship, the problem is their official account behaving like reactionary defensive Reddit mod
Again good for provocative marketing to promote a hardware grift, not as much to get people’s support in diffusing an easy to understand autograd engine
These counterculture reddit mods running startup’s really have a way with words that makes you hate them
They resort to ad hominem on language like they’re sitting in some dingy basement
India has a deteriorating IT industry which was not allowed to develop sovereign capabilities beyond language translation despite having the capital, and is now heading towards certain calamity.