Top all-time · #4076
Chroma
Company blog · research.trychroma.comaidata
Chroma is the company blog at research.trychroma.com, writing about ai and data. Since 2024, 19 different Hacker News readers have linked it 21 times, most often “Context Rot: How increasing input tokens impacts LLM performance”. No feed was found, so read it on the site itself.
Too few sampled comments (12) to say how often HN recommends it.?
Why HN links it
HN readers link Chroma's research pages, especially the 'context rot' study, on how LLM performance degrades as the context window grows. It is cited as support in discussions of long-context models and agents.
Summarised by Claude from a sample of 12 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Basic and applied research for search, retrieval, agents, and context engineering.
Most-linked posts?
- Context Rot: How increasing input tokens impacts LLM performance13 people · 13 mentions · first linked 2025-07-14
- Show HN: Generative Benchmarking for RAG4 people · 4 mentions · first linked 2025-04-07
- Evaluating Chunking Strategies for Retrieval3 people · 3 mentions · first linked 2024-07-09
What Hacker News says?
Unfortunately, I think the context rot paper [1] found that the performance degradation when context increased still occurred in models using attention sinks. 1. https://research.trychroma.com/context-rot
Comment linking /context-rot
on “Claude Sonnet 4 now supports 1M tokens of context”Submitted Context Rot: How increasing input tokens impacts LLM performance as a story · 260 points
HN discussionhttps://research.trychroma.com/context-rot
Comment linking /context-rot
on “Claude Sonnet 4 now supports 1M tokens of context”Submitted Evaluating Chunking Strategies for Retrieval as a story · 7 points
HN discussionThis is well established at this point, it’s called “context rot”: https://research.trychroma.com/context-rot
Comment linking /context-rot
on “Claude Memory”Submitted Show HN: Generative Benchmarking for RAG as a story · 4 points
HN discussionArtifacts can reduce token usage by a factor of 2 or more. I had an [agent evolution framework](danieltan.weblog.lol/2025/06/agent-lineage-evolution-a-novel-framework-for-managing-llm-agent-degradation) before that dumped the output analysis into chat. It often timed out before the 10th conversation. After dumping the analysis into an artifacts, and have the LLM only edit it as required, I can go to 15 or more round…
Comment linking /context-rot
on “Unlike ChatGPT, Anthropic has doubled down on Artifacts”Submitted Chroma Technical Report: Generative Benchmarking as a story · 3 points
HN discussionHigh-level, Chroma lets you incorporate your own data into AI. That can be business data, like your help docs, or customer data, such as private documents. I think there's a misconception among many non-technical people that they should "fine tune" models - to make an LLM sound like their brand or know their internal data. But, the best practice today is "context engineering" - to use a commodity LLM, and add propri…
Comment linking /context-rot
on “Show HN: Chroma Cloud – serverless search database for AI”Submitted Evaluating Chunking Strategies for Retrieval as a story · 2 points
HN discussionOkay, here's the tl;dr: Attention based neural network architectures (on which the majority of LLMs are built) has a unit economic cost that scales (roughly) n^2 i.e. quadratic (for both memory and compute). In other words, the longer the context window, the more expensive it is for the upstream provider. That's one cost. The second cost is that you have to resend the entire context every time you send a new message…
Comment linking /context-rot
on “Opus 4.5 is going to change everything”Submitted Query Embedding Adapters as a story · 2 points
HN discussionThe issue is that once you pollute your context window with the “wrong” information even after you have guided the LLM to the right path, it is still more likely to go off the rails. https://research.trychroma.com/context-rot
Comment linking /context-rot
on “Stop squashing your commits. You're squashing your AI too”Submitted Generative Benchmarking as a story · 1 points
HN discussionI've seen a couple of power users already switching to Pi [1], and I'm considering that too. The premise is very appealing: - Minimal, configurable context - including system prompts [2] - Minimal and extensible tools; for example, todo tasks [3] - No built-in MCP support; extensions exist [4]. I'd rather use mcporter [5] Full control over context is a high-leverage capability. If you're aware of the many limitation…
Comment linking /context-rot
on “What I learned building an opinionated and minimal coding agent”Submitted Generative Benchmarking as a story · 1 points
HN discussionsuper-edit: Sorry, this is not a usage related question, I have move it to: https://news.ycombinator.com/item?id=47772971 Here is the question for which I cannot find an answer, and cannot yet afford to answer myself: In Claude Code, I use Opus 4.6 1M, but stay under 250k via careful session management to avoid known NoLiMa [0] / context rot [1] crap. The question I keep wanting answered though: at ~165k tokens used…
Comment linking /context-rot
on “Claude Code Routines”Submitted Evaluating Chunking Strategies for Retrieval as a story · 1 points
HN discussionAs the context grows, the output (usually a structured function call) remains relatively short. This makes the ratio between prefilling and decoding highly skewed in agents compared to chatbots. The problem is that context engineering is still an emerging science, even though for agent systems, it's already essential. This is overall a really interesting post and makes me think of Chroma’s technical report on contex…
Comment linking /context-rot
on “Context Engineering for AI Agents: Lessons from Building Manus”this is a pretty important piece and the research backs you up. Moving that context out of your system prompt dynamically is going to help reduce your lost in the middle effect. Context rots almost immediately. I've got a project that is being built to address this directly as well, but I'm still very early days. Keep it up! you're on the right track. Hong, K., & Chroma Research Team. (2025). Context rot: How increa…
Comment linking /context-rot
on “Ark – AI agents waste ~30% of context on tool schemas.I built runtime that learn”Incidentally, Chroma also produced the single best study on long-context degradation that I've come across: https://research.trychroma.com/context-rot Before that, I cited nolima ( https://www.reddit.com/r/LocalLLaMA/comments/1io3hn2/nolima_... ) constantly to illustrate how difficult tasks involving reasoning or multi-step information gathering degraded much faster than the needle-in-haystack benchmarks cited by th…
Comment linking /context-rot
on “ChromaDB Explorer”
