Uploaded April 2026 | Updated November 2026, 2 days ago
In this clip, Merve breaks down how AI vision evolved and explains why it matters in practice.
Dense explanation of how vision evolved and why progress feels incremental now.
🤗 Listen to the full podcast episode
👉 Here: youtu.be/SjjCpeTjXIY
Connect with Merve:
- Merve on X — https://x.com/mervenoyann
- Vision Language Models (O'Reilly) — oreilly.com/library/view/vision-language-models/9798341624030
Chapters:
- 00:00 How AI Vision Evolved
- 00:12 Vision Transformers
- 01:06 LLaVA
- 01:38 IDEFICS
- 02:06 CLIP + Projection Layer
- 02:54 Interleaving
- 05:42 Segment Anything
Topics covered:
- Vision Transformers
- LLaVA
- IDEFICS
- CLIP + Projection Layer
- Interleaving
Sources mentioned:
- An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale — arxiv.org/abs/2010.11929
- Visual Instruction Tuning project page — llava-vl.github.io
- IDEFICS: an open reproduction of Flamingo — huggingface.co/blog/idefics
- CLIP: Connecting text and images — arxiv.org/abs/2103.00020
- IDEFICS2 model documentation — huggingface.co/docs/transformers/model_doc/idefics2
- Segment Anything — arxiv.org/abs/2304.02643
In this clip, Merve breaks down how AI vision evolved and explains why it matters in practice.
Dense explanation of how vision evolved and why progress feels incremental now.
🤗 Listen to the full podcast episode
👉 Here: youtu.be/SjjCpeTjXIY
Connect with Merve:
- Merve on X — https://x.com/mervenoyann
- Vision Language Models (O'Reilly) — oreilly.com/library/view/vision-language-models/9798341624030
Chapters:
- 00:00 How AI Vision Evolved
- 00:12 Vision Transformers
- 01:06 LLaVA
- 01:38 IDEFICS
- 02:06 CLIP + Projection Layer
- 02:54 Interleaving
- 05:42 Segment Anything
Topics covered:
- Vision Transformers
- LLaVA
- IDEFICS
- CLIP + Projection Layer
- Interleaving
Sources mentioned:
- An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale — arxiv.org/abs/2010.11929
- Visual Instruction Tuning project page — llava-vl.github.io
- IDEFICS: an open reproduction of Flamingo — huggingface.co/blog/idefics
- CLIP: Connecting text and images — arxiv.org/abs/2103.00020
- IDEFICS2 model documentation — huggingface.co/docs/transformers/model_doc/idefics2
- Segment Anything — arxiv.org/abs/2304.02643





