‘My experience of working with the ageing IT infrastructure of the NHS means that I find it hard to believe that the fixes Project Glasswing identifies will quickly find their way onto every computer that needs to be protected.’ @paul3548.bsky.social on AI and coding. www.lrb.co.uk/the-paper/v4...

‘I am curious to meet Mythos. The engineers who tested it say they were struck by how often it referred to the work of the cultural theorist Mark Fisher and to the philosopher Thomas Nagel.’ Paul Taylor (@paul3548.bsky.social) on AI and programming. www.lrb.co.uk/the-paper/v4...

‘Overdiagnosis isn’t only a problem in screening; it is a possibility wherever a diagnosis brings harms as well as benefits.’ Paul Taylor (@paul3548.bsky.social) on overdiagnosis. www.lrb.co.uk/the-paper/v4...

‘One very simple way to address a language model’s deficiencies in problem-solving is to express problems not as exercises in reasoning but in the generation of language.’ Paul Taylor (@paul3548.bsky.social) on DeepSeek and problem-solving AI: www.lrb.co.uk/the-paper/v4...

‘OpenAI hides its models’ reasoning from users, perhaps because the content would reveal the extent to which it has infringed the intellectual property of the writers and publishers whose material its models are trained on.’ Paul Taylor (@paul3548.bsky.social) on AI: www.lrb.co.uk/the-paper/v4...

Post a warning

Discussion of LLMs moving on from generating text to solving problems. Interesting to see how far this will go, given they are machines for generating text. See this at @lrb.co.uk www.lrb.co.uk/the-paper/v4...

‘AI research has paid a huge amount of attention to language models; value networks have received less attention, but these are huge artefacts, similar in scale to the language models themselves, and vastly expensive to develop.’ Paul Taylor (@paul3548.bsky.social): www.lrb.co.uk/the-paper/v4...

‘OpenAI hides its models’ reasoning from users, perhaps because the content would reveal the extent to which it has infringed the intellectual property of the writers and publishers whose material its models are trained on.’ Paul Taylor (@paul3548.bsky.social) on AI: www.lrb.co.uk/the-paper/v4...

‘R1 gets to the right answer eventually, and provides a succinct explanation, but only after what seems an absurd display of self-doubt and overthinking.’ Paul Taylor (@paul3548.bsky.social) on DeepSeek: www.lrb.co.uk/the-paper/v4...

... Thomas Meaney on the German elections Adam Mars-Jones on Dominique Fernandez Alex Harvey on Powell and Pressburger Michael Hofmann on Paul Valéry Francis Gooding on Picasso Paul Taylor (@paul3548.bsky.social) on DeepSeek and poems by Paula Bohince. Read now at www.lrb.co.uk

‘A reduced version of R1 is publicly available. Anyone can download a model which, in theory, is almost as good as the state of the art, and will, in theory, run on a laptop.’ Paul Taylor (@paul3548.bsky.social) assesses China’s DeepSeek AI: www.lrb.co.uk/the-paper/v4...