Amal Iyer

@jeffreycider Safeway’s Tres Leches is super under-rated

@varunshenoy_ Really solid blog! Loved the predicted v/s actual performance plots. As a follow-up would be nice to see if the fudge factor for a GPU type holds across different decoder only model sizes.

Surprised by how many AI products outsource their capabilities to black box APIs. Optimized open-source models can improve performance and drastically bring down costs (!) You can't do this with just an endpoint. Leverage the hardware. Here's how 👇 baseten.co/blog/llm-trans…

At #AIForward Summit 2023, @aiyerix dove deep into the importance of evaluating #LLMs in pre-production using Fiddler Auditor, the open-source robustness library for red-teaming of LLMs. Watch full video: fiddler.ai/resources/ai-f…

Larger context lengths are an existential challenge to vector DB providers. 128K context length suffices most chatbot applications

Attention: A good f*n bagel is all you need.

Using GPT-4 for evals. Cost is 🥲

@karinanguyen Thanks for sharing the deck! Conjuring a 'good' prompt seems deceptively simple until you start testing the model.

Slides: bit.ly/principles-pro…

Gave a small workshop @aiDotEngineer on how to write optimal prompts for LLMs (in particular Claude) in different contexts: from fashion curation to model-written evaluations! 1/ I explore how I'd now use Claude for the fashion app (InterAlia) that I launched a year ago:

Excited for the upcoming workshop on Responsible and Open Foundation models. Pretty solid line-up!

@goodside & @simonw showed that LLMs are vulnerable to #PromptInjection attacks. Check out how Fiddler Auditor can help minimize prompt injection attacks through #LLM testing: fiddler.ai/blog/evaluate-… (by @aiyerix & Karen He) #GenerativeAI #ResponsibleAI #RobustML #TrustworthyAI

Announcing our 🌟open-source🌟 Fiddler Auditor - the first robustness library that leverages LLMs to evaluate other LLMs! fiddler.ai/blog/introduci… #MLOps teams can now test LLM robustness with adversarial examples, out-of-distribution inputs, and linguistic variations to improve AI performance. #LLMOps #MachineLearning #AI #ArtificialIntelligence #LLM #OpenSource #DataScience #DataEngineering #DataScientist #DataEngineer