Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learning · HackerPapers discussionshttps://hackerpaper.kaipfister.de/p/2203.02053/Collected HN comments and submissions linking this paper.enMon, 02 Nov 2026 11:16:22 GMTMind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation Learning — mentioned by bluejay2387https://news.ycombinator.com/item?id=50108530urn:hackerpapers:mention:2203.02053:50108530Thu, 15 Oct 2026 18:58:00 GMT<p>The claims seem to be an overreach of what is actually proven. We&#x27;ve known about progressive feature complexity and the impacts of early training structures on later training for a really long time (since CNN&#x27;s, The Lottery Ticket Hypothesis, etc...) and neither of those phenomenon are mutually exclusive with parameter overloading. The Multimodal Gap work (https://arxiv.org/abs/2203.02053) seems to imply networks have a tendency towards condensed representations that are clearly not semantic and a bunch of other considerations are already on the table for discussion. I don&#x27;t think these have…</p><p><a href="https://hackerpaper.kaipfister.de/p/2203.02053/">Paper and discussion history</a>. A mention is not an endorsement.</p>cs.CL