

the use of live wikis rather than ZIM archive snapshots is a clear sign of either malicious incompetence, or malicious intent. There’s zero reason for an llm agent to need the live website.


the use of live wikis rather than ZIM archive snapshots is a clear sign of either malicious incompetence, or malicious intent. There’s zero reason for an llm agent to need the live website.


I was in machine learning for a bit, up until venture capitalists eviscerated research teams chasing ways to make money from it. While there’s definitely a degree of overhyping risks for investment, there are absolutely some real, tangible risks. But the dangers aren’t like skynet/terminator, its more of a slow decline in knowledge combined with an increase of predictive surveillance (https://arxiv.org/abs/1902.10739 as an example of why predictive surveillance should worry everyone)
Every mistake and error in token generation is getting confounded with human text, and being fed back as part of the training loop. Combined with the sheer amount of botting of the modern internet, you get a feedback loop of error being used to train, followed by output of that error dominating the next training cycle as bots drown out voices of the knowledgable. For some basic things it’s easy to see the error, but in any expert domain the ratio of bots to experts is large enough that expert knowledge won’t stay within the training distribution.


deleted by creator


Years of auto update paranoia paid off for me, never trust an executable that doesn’t give you a hash to check.


I’m going to throw my own thoughts in on this. I got into machine learning around 2015, back when relu activations were still bleeding edge innovations, and got out around 2020 for honestly pretty similar reasons.
Emotions can and have been used as optimisation targets. Engagement is an ever present target. And in the framework of capitalism, one optimisation targets rules above all others; alignment with continued use. It’s part of what leads to the bootlicking LLM phenomenon. For the average human, it drives future engagement.
The real danger isn’t the newer language models, or anything really to do with neural net architecture; rather, it’s the fact that we’ve found that a simple function minimisation strategy can be used to approximate otherwise intractable functions. The deeper you research, the more clear it becomes that any arbitrary objective can be optimised, given a suitable function approximator and enough data to fit the approximator accurately.
Human minds are also universal function approximators.


on average, i’m earning 3x as much from bandcamp FLAC sales compared to what i earn off streaming. Hasn’t been ruined so far.
If you want that long tail, bandcamp and soundcloud are better sources. The barrier to entry is low with those, and there’s a plethora of small, niche artists just doing their own thing.
For a representative snapshot of music though, it’s pretty amazing. It shows what a massive percentage of the planet listens to, preserved hopefully across many seeds, and historians will love shit like this in the future.
deleted by creator