airylizard

8 ms·

1.
A design brief as one image: can a 27B vision model build the animation?(automationoptimization.github.io)
3 points by airylizard 11d ago|0 comments
2.
Show HN: Patient Glue a more affordable SMS solution for healthcare that I built(patientglue.com)
1 points by airylizard 1y ago|0 comments
3.
Think Before You Speak – Exploratory Forced Hallucination Study [pdf](github.com)
2 points by airylizard 1y ago|0 comments
4.
Evaluating AI Agents with Azure AI Evaluation(techcommunity.microsoft.com)
5 points by airylizard 1y ago|0 comments
5.
by airylizard 1y ago
Exactly what leads to inaccurate output in LLM's. The semantic interpretation of each individual token isn't the same between us and it. "Interpretation", we likely define accuracy and reliability pretty closely the same
6.
by airylizard 1y ago
Are you continuing research? Is there somewhere we can follow along?
7.
by airylizard 1y ago
The fact that embeddings from different models can be translated into a shared latent space (and back) supports the notion that semantic anchors or guides are not just model-specific hacks, but potentially universal tools. Fantastic read, t
8.
by airylizard 1y ago
As more and more people use brute force loops to make their AI agents more reliable, this hidden inference giant will only continue to grow. This is why I put my framework together, using just 2 passes as opposed to n+ can increase accuracy
9.
by airylizard 1y ago
love it. any llm can be made to perform reliably and accurately which is the biggest pre-requisite when it comes to creating an "AI Agent". I think this gives people the opportunity to start somewhere because they can leverage mul
10.
by airylizard 1y ago
The data "supply chain" has already surged ahead of production elsewhere. Companies aren't just passively taking what's out there, they actively harvest highly curated content, benefiting even further when we voluntarily
11.
by airylizard 1y ago
I like the idea, TSCE framework should make the individual agents more reliable and deterministic: https://github.com/AutomationOptimization/tsce_demo
12.
by airylizard 1y ago
Right, the 4.1 training checkpoint hasn’t moved. What has moved is the glue on top: decoder heuristics / safety filters / logit-bias rules that OpenAI can hot-swap without re-training the model. Those “serving-layer” tweaks are wh
13.
by airylizard 1y ago
Hey, thanks for kicking the tires! The run you’re describing was done in mid-April, right after GPT-4.1 went live. Since then OpenAI has refreshed the weights behind the “gpt-4.1” alias a couple of times, and one of those updates fixed the
14.
by airylizard 1y ago
The test isn't for how well an LLM can find or replace a string. It's for how well it can carry out given instructions... Is that not obvious?
15.
by airylizard 1y ago
Why I came up with TSCE(Two-Step Contextual Enrichment). +30pp uplift when using GPT-35-turbo on a mix of 300 tasks. Free open framework, check the repo try it yourself https://github.com/AutomationOptimization/tsce_dem
16.
by airylizard 1y ago
1. What TSCE is in one breath Two deterministic forward-passes. 1. The model is asked to emit a hyperdimensional anchor (HDA) under high temperature. 2. The same model is then asked to answer while that anchor is prepended to the original p
17.
TSCE and HyperDimensional Anchors: Making AI agents/workflows reliable at scale(github.com)
3 points by airylizard 1y ago|1 comments
18.
Show HN: TSCE – Think Before You Speak (Two-Step Contextual Enrichment for LLMs)(github.com)
3 points by airylizard 1y ago|0 comments