{"id":1884,"date":"2024-04-10T16:34:12","date_gmt":"2024-04-10T14:34:12","guid":{"rendered":"https:\/\/www.pauljorion.com\/blog_en\/?p=1884"},"modified":"2024-04-10T16:36:58","modified_gmt":"2024-04-10T14:36:58","slug":"the-transformer-is-a-thinking-object-by-claude-roux","status":"publish","type":"post","link":"https:\/\/www.pauljorion.com\/blog_en\/2024\/04\/10\/the-transformer-is-a-thinking-object-by-claude-roux\/","title":{"rendered":"The transformer<\/em> is a thinking object<\/b>, by Claude Roux *"},"content":{"rendered":"

\"\"<\/a><\/p>\n

Spinoza’s portrait by Stable Diffusion<\/p><\/blockquote>\n

A few years back, I had devoured Fr\u00e9d\u00e9ric Lordon’s book Capitalisme, D\u00e9sir et Servitude<\/em>. In particular, I’d been fascinated by the notion of conatus<\/i>, the desire for power. Fr\u00e9d\u00e9ric Lordon explained<\/a> that Spinoza, as a mathematician, had defined conatus<\/i> in the form of mathematical vectors, enabling him to transform his philosophical thinking into mathematical axioms and theorems.<\/p>\n

And it was here that I was struck by the resemblance to transformers<\/em>.<\/p>\n

Indeed, a transformer takes as input a succession of tokens<\/em> or words, which it replaces with their embedding<\/i>. An embedding is a huge vector that has been assigned to each word or token, enabling them to be compared with each other. So we can, for example, perform mathematical operations such as “boy = woman – man + girl”.<\/p>\n

The transformer then combines these vectors via a succession of mathematical transformations learned during training to generate a new vector V<\/i>. This vector is then compared with the set of embeddings<\/i> to identify the closest to generate the next word.<\/p>\n

[E1 E2 E3 … En] –> TRANSFORM –> V –> next token<\/p>\n

It thus takes as input a sequence of words for which it constructs a true interpretation in the form of a new semantic vector V.<\/i> At the next iteration, it injects this word at the end of the context and resumes its generation until it ends up with the complete text. <\/p>\n

I can’t help seeing the succession of V<\/i> as a form of conatus<\/i>. Indeed, a model learned with a reinforcement algorithm is intentioned<\/i>, unlike other models, which are supervised<\/i>. This intention is defined in particular by RLHF (Reinforcement learning from human feedback), which forces the model to stay in line, to align itself with human desires, their conatus<\/i> according to Spinoza.<\/p>\n

This process can be seen as a means of imposing a particular conatus<\/i> on a model.<\/p>\n

According to this interpretation, we are faced with a thinking object.<\/p>\n

* Grands Mod\u00e8les de Langage : Pourquoi les r\u00e9seaux neuronaux ont-ils r\u00e9ussi l\u00e0 o\u00f9 la linguistique \u00e9chouait ?<\/a>, by Claude Roux, March 8th<\/sup> 2024<\/p>\n

PJ TV \u2013 IA : Pourquoi la linguistique a-t-elle \u00e9chou\u00e9 ?<\/a> a conversation with Claude Roux, March 15th<\/sup> 2024<\/p>\n

\"\"<\/a><\/p>\n

Portrait of a thinking object according to DALL\u00b7E<\/p><\/blockquote>\n","protected":false},"excerpt":{"rendered":"

\"\"<\/a><\/p>\n

\n

Spinoza’s portrait by Stable Diffusion<\/p>\n<\/blockquote>\n

A few years back, I had devoured Fr\u00e9d\u00e9ric Lordon’s book Capitalisme, D\u00e9sir et Servitude<\/em>. In particular, I’d been fascinated by the notion of conatus<\/i>, the desire for power. Fr\u00e9d\u00e9ric Lordon explained<\/a> that Spinoza, as a mathematician, had […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_crdt_document":"","footnotes":""},"categories":[3,12],"tags":[401,356,400],"class_list":["post-1884","post","type-post","status-publish","format-standard","hentry","category-artificial-intelligence","category-human-complex-systems","tag-baruch-spinoza","tag-llm","tag-transformer"],"_links":{"self":[{"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/posts\/1884","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/comments?post=1884"}],"version-history":[{"count":3,"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/posts\/1884\/revisions"}],"predecessor-version":[{"id":1887,"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/posts\/1884\/revisions\/1887"}],"wp:attachment":[{"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/media?parent=1884"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/categories?post=1884"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.pauljorion.com\/blog_en\/wp-json\/wp\/v2\/tags?post=1884"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}