Top all-time · #23073
Guide Labs
Company blog · guidelabs.ai/postaiscience
Guide Labs is the company blog at guidelabs.ai, writing about ai and science. Since 2026, 3 different Hacker News readers have linked it 10 times, most often “Prism: When an LLM predicts the next token, which training does it relying on?”. No feed was found, so read it on the site itself.
Too few sampled comments (1) to say how often HN recommends it.?
Why HN links it
Linked by a commenter on a Show HN for Steerling-8B, an interpretable language model. The commenter asks for a published paper on the algorithm and cites a Guide Labs post on scaling interpretable models.
Summarised by Claude from a sample of 1 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
We are building a new class of interpretable AI systems and foundation models that humans can reliably debug, trust, and understand.
Most-linked posts?
- Prism: When an LLM predicts the next token, which training does it relying on?2 people · 4 mentions · first linked 2026-02-24
- /post/scaling-interpretable-models-8b/2 people · 2 mentions · first linked 2026-02-24
What Hacker News says?
Great questions. We weren't quite explicit about the training data attribution process. We'll discuss this in more detail in future work. We can track down which parts of the training data were interpolated to create that sentence. For those training data sentences, we then compare the concepts between generated and training. We can attribute to exact sentences and chunks in the training data. For the first release,…
Comment linking /post/prism/
on “Show HN: Steerling-8B, a language model that can explain any token it generates”Submitted Show HN: Clarity, See what concepts your LLM uses and trace it to training data as a story · 3 points
HN discussionDown to the very exact text chunk in a document! Check this out for an idea of what smaller versions of this style of model can do: https://www.guidelabs.ai/post/prism/ . We'll have more to say soon about it. We can trace any generation to 11B chunks (not documents, but actual chunks in the training data).
Comment linking /post/prism/
on “Show HN: Steerling-8B, a language model that can explain any token it generates”Submitted Aligning LLMs at inference time by suppressing internal concepts as a story · 2 points
HN discussionLooks very interesting. Is there a published paper/article on your algorithm? Would like to take a dab at implementing this on my own. I could find this [0], but not sure if that represents the entire system? (Apologies, I am not that well versed in ML) [0] - https://www.guidelabs.ai/post/scaling-interpretable-models-8...
Comment linking /post/scaling-interpretable-models-8b/
on “Show HN: Steerling-8B, a language model that can explain any token it generates”Submitted Show HN: Steerling-8B, a language model that can explain any token it generates as a story · 2 points
HN discussionExactly! We will have a future post that shows this more granularly over the coming weeks. Here is a post we wrote on how this works at smaller scale: https://www.guidelabs.ai/post/prism/
Comment linking /post/prism/
on “Show HN: Steerling-8B, a language model that can explain any token it generates”Submitted Extracting 100K concepts from an 8B LLM as a story · 1 points
HN discussionIt does :) We constrained the model to do exactly this during training: https://www.guidelabs.ai/post/scaling-interpretable-models-8... .
Comment linking /post/scaling-interpretable-models-8b/
on “Show HN: Steerling-8B, a language model that can explain any token it generates”Submitted Prism: When an LLM predicts the next token, which training does it relying on? as a story · 1 points
HN discussion