For folks wondering what's happening here technically, an explainer: When there's lots of training data with a particular style, using a similar style in your prompt will trigger the LLM to respond in that style. In this case, there's LOADS of fanfic: https://
scp-wiki.wikidot.com/scp-series
@jeremyphoward
-
LLM Style Mimicry Through Training Data Patterns
By
–
-
System Complexity Undermines Security, Maintainability, Scale, Performance
By
–
"security, maintainability, scale, and performance" are exactly the things most likely to suffer when you create more systems, each with their own interface boundaries.
-
Single Server Architecture for High Performance Systems
By
–
"the correct way to build" is a silly claim. There are many high performance extremely scalable systems built with the simpler design that @levelsio describes. I've done both, and nowadays I stick to keeping things in a single server wherever possible.
-
Author Bias in Academic Assessment of Alternative Viewpoints
By
–
I don't agree the work "holds up to and maybe exceeds the academic standards of engaging with alternative views". I'm not sure that's an assessment that you as an author should be making tbh. I don't see how you could believe you have an unbiased assessment of that.
-
Chain of Thought Faithfulness and Reliability in AI Systems
By
–
2) I meant that there's no reason to consider a CoT "faithful". To be "faithful" one has to reliably and consistently attempt to provide correct information. One can be "sometimes factual" yet still "unfaithful".
-
Expert Endorsements: Norms and Best Practices in Content Validation
By
–
1) It's not common to have "expert endorsers" either. Although there is some prior history of authors trying to provide the reader a sense of how experts assess the content. In these cases, the norm is to provide responses received from all contacted experts.
-
Jeremy Howard shares Anthropic research on similar topics
By
–
Similar thoughts about this paper here from Anthropic's @janleike
: -
Publishing Research: Present Both Supporting and Critical Feedback
By
–
IMO if you're going to go to the trouble to get feedback about your paper before publishing (which is a great idea!) you should publish *both* sides. Just listing prominent supporters gives the wrong impression.
-
Chain of Thought Training Faithfulness and Interpretability Questioned
By
–
They've only listed those that agree I told them "I don't agree. I don't think CoT has much of a useful role to play. It's only really showing something that appears to be a meaningful trace because they're trained to appear that way, but actually they're not faithful at all"