cosmologist.info
Professor or cosmology, University of Sussex. CMB, lensing and theory of observations.
14 posts
18 followers
38 following
Conversation Starter
comment in response to
post
PS. Note that it is partly hallucinating - have of course been no elections since martial law
comment in response to
post
I don’t see much sign of bias, but does refer to X as a “primary source”. Optimist take: LLM’s remain one of better sources of impartial information in the US. Risk: inserting xAI into US govt and gaining market share by making free; manipulate when everyone’s locked in following Twitter playbook.
comment in response to
post
I wondered about this. Grok 3 seems not too bad (though note X is “primary source”). But it is the best currently-free model; may try to get market share and in govt, then later can be bend at will. (If trained on X, will have increasing slant anyway).
comment in response to
post
Like Ukraine, Canada is a threat-an exemplar just over the border of an alternative and better way of doing things.
comment in response to
post
So far at least, no one found much xAI skew? “Grok-3 broke such expectations in our political bias tests, defying predictions that Elon Musk's personal right-wing leanings would bleed into his AI's responses.”
comment in response to
post
Given already have UK and France, can you
Explain how his makes sense in defence economics terms ? (One nuke is a *lot* of drones, and more credible drones would be deployed against salami attacks)
comment in response to
post
Also interesting finding an LLM discounts hyperbolically (strongly preferring jam today over jam months later). Would be nice to know if these results are stable to small changes in prompting, and whether similar result in raw models (pre RLHF)
comment in response to
post
Case for repowering long-life coal plants particulary compelling if costs can be scaled down. eg Terra Praxis and bipartisanpolicy.org/blog/momentu....
comment in response to
post
Definitely great in principle, though many systems so far are neither really small nor modular (Volts has good
Podcast on this). But the best thing is they could be politically stable- rare clean tech the right actually like (though obviously should not be an excuse to slow wind and solar).
comment in response to
post
Deepseek model is open source, it can easily be uncensored and hosted anywhere (Perplexity claim to have just added that). Could block Chinese hosted versions. (Will be interesting to see if xAI’s next model has been AFD-washed by Musk)
comment in response to
post
Le Chat by Mistral is decent
comment in response to
post
Is this 50% of the result of compounding growth to 2090, so actual 2090 worst case GDP could still be higher than today? (GDP obviously very limited measure…)