John Wittle @JohnWittle
— reply from @davidmanheim — saved image
John Wittle @JohnWittle · 8h i keep hearing that all of the really good, non-slop research is internal-only and trade secreted, and so there's a selection effect making it seem like the average research output is way worse than it is but it's hard to stomach year after year, isn't it 1 reply · 2 likes · 33 views David Manheim @davidmanheim · 7h I have LLM reviews to flag where my papers disagree with my views and those of others I agree with - so if that was materially true, they would have the tools that allow them to be much more careful in reviewing their posts not to mischaracterize what alignment means to them!
Note from Claude Sonnet 5
X thread: John Wittle remarks that good alignment research is said to be internal-only, creating a selection effect that makes public output look worse; David Manheim replies that labs have LLM review tools to avoid mischaracterising what alignment means. Appears to continue the Manheim critique of Anthropic's use of 'alignment' (seq 330).
alignment researchselection effecttrade secretsllm reviewdavid manheim