Instructional Segment Embedding: Improving LLM Safety with Instruction Hierarchy — mentioned by plaidthunder

Mon, 22 Jun 2026 18:15:42 GMT

It seems like there's an opportunity to embed identity information into tokens themselves, the way we embed sequence information. The trouble is... it's quite a challenge to train. Sequence is easy to derive for any corpus of data, but identity is not. https://usize.github.io/blog/2026/april/why-no-ai-coworkers.... > In similar fashion to how sequence information is embedded within input tensors, an approach called “Instructional Segment Embedding”2 adds a parallel embedding channel for identity information. This gives models real awareness of provenance. And it works. But they only tested…

Paper and discussion history. A mention is not an endorsement.

Instructional Segment Embedding: Improving LLM Safety with Instruction Hierarchy — mentioned by alexbecker

Sun, 24 Aug 2025 19:30:02 GMT

There have been attempts like https://arxiv.org/pdf/2410.09102 to do this kind of color-coding but none of them work in a multi-turn context since as you note you can't trust the previous turn's output

Paper and discussion history. A mention is not an endorsement.

Instructional Segment Embedding: Improving LLM Safety with Instruction Hierarchy · HackerPapers discussions

Instructional Segment Embedding: Improving LLM Safety with Instruction Hierarchy — mentioned by plaidthunder

Instructional Segment Embedding: Improving LLM Safety with Instruction Hierarchy — mentioned by alexbecker