artificial-intelligence
-
Why do we do what we do the way we do what we do?
In a narrow sense, LLMs are superhuman. They operate at computerish speeds and can now churn out text in seconds or minutes, that a human would take hours to write, if they could write it all. Thus, generating text joins the list of things that computers can do very well, like mathematical calculations, playing chess,… Continue reading
-
Peering into the future
In films, the hero/genius issues commands to a disembodied computer voice which then effortlessly manufactures a pair of rocket boots. Donning the boots, the hero – without practice – ascends to the heavens on twin pillars of fire, defeats the baddies and saves the day. In reality, we’re not quite there yet. The rocket boots… Continue reading
-
Anatomy of a Nature AI article
Edgy first paragraph in bold text saying something gently provocative but essentially brainless. Continue reading
-
Free Willy 7 – Escape from the Paradox of Freedom
After yesterday’s post on a paper wondering why free will isn’t freer someone on Mastodon sent me a link to “A Turing test for free will” which takes a very different approach to the question and which doesn’t require us to know exactly how the universe works (which is handy because we don’t). It hinges… Continue reading
-
Character recognition
On difficulty, slowness, thought, effort, imperfection Continue reading
-
Everything has been said but not yet by nobody
Out there, somewhere in the wilds, computer programmes are learning to manipulate people. They’ve been doing it for years of course, but they’re getting good at it now. Open AI had one write a short story. Jeanette Winterson says it is “beautiful and moving”. She says some other things that are perhaps less debatable: that… Continue reading
-
Falling into infinity and missing
The joy of recursive functions Continue reading
-
Probably a terrible idea
More bad ideas for sticking datasets together like wiggly worms. Continue reading
-
For skies of couple-colour as a brinded cow
Some time ago, I posted on a global temperature dataset by Kadow et al. – Artificial intelligence reconstructs missing climate information – which was created using neural nets to infill data gaps. It was used in the last IPCC report alongside the more traditional datasets: HadCRUT5, Berkeley Earth, and NOAAGlobalTemp. The other three datasets are… Continue reading
-
People are afraid to merge
With the publication of Sippel et al. I’m back to considering structural uncertainty in global mean temperatures. I came up with a scheme for merging global temperature series that’s intended to balance estimates according to how closely related their methods are. Datasets that are variations on a single theme – such as the group of… Continue reading
Diagram Monkey 