kundan_s__r

6 ms·

1.
by kundan_s__r 9mo ago
That framing resonates a lot. In production, creativity is often just unbounded variance. Once each step is intentionally boring and constrained, failures become predictable and debuggable — which is what engineering actually optimizes for.
2.
by kundan_s__r 9mo ago
That’s a very real example of the core problem: LLMs don’t reliably honor constraints, even when they’re explicit and simple. Instruction drift shows up fast in learning tasks — and quietly in production systems. That’s why trusting them “a
3.
by kundan_s__r 9mo ago
A helpful way to learn this is to separate models, machines, and practice. For computation models, the circuit model and measurement-based computation cover most real work. Aaronson’s Quantum Computing Since Democritus and Nielsen & Chu
4.
by kundan_s__r 9mo ago
The real risk with LLMs isn’t when they fail loudly — it’s when they fail quietly and confidently, especially for non-experts or downstream systems that assume structured output equals correctness. When you don’t already understand the doma
5.
by kundan_s__r 9mo ago
That’s fair — if you’re already skeptical and paying attention, the failures are obvious and often funny. The risk tends to show up more with non-experts or downstream systems that assume the output is trustworthy because it looks structure
6.
by kundan_s__r 9mo ago
This matches our experience too. The biggest reduction in hallucinations usually comes from shrinking the action space, not improving the prompt. When inputs, tools, and outputs are explicitly constrained, the model stops “being creative” i
7.
by kundan_s__r 9mo ago
Fair enough. A healthy dose of skepticism has served us well for every overhyped wave so far. The difference this time seems to be that AI systems don’t just fail noisily — they fail convincingly, which changes how risk leaks into productio
8.
by kundan_s__r 9mo ago
Really impressive work, especially on mobile. The mmap + zero-copy, read-only approach feels like the right tradeoff for files at this scale. Curious how it behaves with extremely wide objects or deep nesting — do index build time or memory
9.
by kundan_s__r 9mo ago
This matches what I’ve seen as well. A lot of “debt relief” and “settlement” services are essentially rent-seeking intermediaries that leave consumers worse off or stuck in long programs with unclear outcomes. Non-profit credit counseling (
10.
by kundan_s__r 9mo ago
That’s a very sane stance. Treating LLM output as untrusted input is probably the correct default when correctness matters. The worst failures I’ve seen happen when teams half-trust the model — enough to automate, but still needing heavy gu
11.
by kundan_s__r 9mo ago
This is a very pragmatic take. The “90% accuracy is a liability” line resonates — in high-stakes systems, partial automation often costs more than it saves. What I like here is the field-level confidence gating instead of a single document
12.
Ask HN: How are you preventing LLM hallucinations in production systems?
3 points by kundan_s__r 9mo ago|13 comments
13.
by kundan_s__r 9mo ago
This resonates. A lot of AI reading tools optimize for removal of effort (summaries, shortcuts), which often ends up weakening comprehension rather than strengthening it. One thing I’m curious about: how do you decide when the AI should int
14.
Show HN: Verdic Guard – Deterministic guardrails to prevent LLM hallucinations(verdic.dev)
1 points by kundan_s__r 9mo ago|0 comments
15.
Show HN: Verdic Guard – Deterministic guardrails to prevent LLM hallucinations
2 points by kundan_s__r 9mo ago|1 comments
16.
by kundan_s__r 9mo ago
Whether or not Hallucination “happens often” depends heavily on the task domain and how you define correctness. In a simple conversational question about general knowledge, an LLM might be right more often than not — but in complex domains
17.
by kundan_s__r 9mo ago
Interesting reflection — but I’d push back on treating surface similarities between human conversational quirks and LLM failure modes as evidence they’re really the same thing. The article lists things like “not stopping generating,” “small
18.
Show HN: Verdic Guard – deterministic guardrails for production AI
1 points by kundan_s__r 9mo ago|0 comments
19.
Show HN: Verdic Guard – validating LLM outputs against intent, not just prompts
2 points by kundan_s__r 9mo ago|0 comments
20.
by kundan_s__r 9mo ago
This framing resonates a lot. The core issue you’re pointing at isn’t model accuracy, it’s epistemic accountability. In most current deployments, an AI system’s output is treated as transient: generated, consumed, forgotten. When that outpu
21.
by kundan_s__r 9mo ago
please check verdic.dev
22.
Show HN: A policy enforcement layer for LLM outputs (why prompts weren't enough)
1 points by kundan_s__r 9mo ago|0 comments
23.
Show HN: Verdic Guard – Policy Enforcement and Output Validation for LLMs
1 points by kundan_s__r 9mo ago|0 comments
24.
Verdic – Intent governance layer for AI systems https://www.verdic.dev/
1 points by kundan_s__r 9mo ago|0 comments