What Aishwarya Naresh Reganti thinks about LLMs
AI practitioner and researcher who has helped build more than 50 enterprise AI products at companies including OpenAI, Google, Amazon and Databricks; founder of LevelUp Labs and author of over 40 research papers.
Aishwarya Naresh Reganti did not write this page.
We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.
Filter LLMs · Newest first
Only Aishwarya Naresh Reganti on LLMs. See everyone on LLMs
19 dated positions, 2026, in their own words. Our reading of what Aishwarya Naresh Reganti has said — not written or endorsed by them.
-
Their wordsGiving a model information isn't the same as teaching it how to perform a task.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 1st of 17 in this recording
-
Their wordsSo, the first argument holds when the missing piece is information or when the general model already performs well enough. It doesn't prove that training or fine-tuning is unnecessary when the model consistently struggles with the task itself
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 2nd of 17 in this recording
-
Their wordsif a customized model performs better on your workload today or serves it at a substantially lower cost, waiting for a hypothetical future model isn't automatically the right business decision, right? The real question you should be asking is how long that performance or cost advantage last and whether it's valuable enough to justify fine-tuning now.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 4th of 17 in this recording
-
Their wordsNow, compare all of that work with just calling a foundation model through an API. You don't need a training pipeline, and adapting to a new model might be fairly straightforward if it passes your evals. So, the operational cost with fine-tuning is very real.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 5th of 17 in this recording
-
Their wordsNow, that removes the infrastructure work, not the need for good data or evaluations, and that isn't really free. But remember that it pretty much changes who can realistically experiment with fine-tuning.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 6th of 17 in this recording
-
Their wordsBut it does mean that inference price is not the only consideration. It's about companies wanting to own their data and make sure they control the ecosystem without giving it away to external providers.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 7th of 17 in this recording
-
Their wordsNow, remember that the goal for these companies is not to build something as capable as GPT or Claude or general models. It's almost the opposite, actually.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 8th of 17 in this recording
-
Their wordsAnd if you have a few thousand API calls that are being done, it's really hard to justify fine-tuning a model. But for millions of repetitive requests, that calculation can look way different.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 9th of 17 in this recording
-
Their wordsthe choice is not always an either/or. It's not will I be using just frontier models or just fine-tuning, but a hybrid where difficult and broad tasks can be handed off to frontier models, but smaller tasks that are more repetitive in nature can be handled by fine-tuned specialized models.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 10th of 17 in this recording
-
Our readingThe real question is not whether fine-tuning is dead but whether it is dead for a given workload.
Their wordswhich means the question isn't really whether fine-tuning is dead, it's whether it's dead for your workload.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 11th of 17 in this recording
-
Their wordsWell, in my opinion, I don't think it is, but I also don't think companies should be rushing to fine-tune models because a lot of companies underestimate one barrier, which is the data that's needed for fine-tuning.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 12th of 17 in this recording
-
Their wordsSo, there's no definitive proof that custom models are generally better than frontier models. It just shows that fine-tuning can make sense for specific workloads
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 13th of 17 in this recording
-
Their wordsIn all of the cases that we discussed so far, what mattered more was the judgement, knowing when to fine-tune, which approach fits the problem, and how to prove that it actually worked.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 14th of 17 in this recording
-
Their wordsmost enterprises are still learning how to get value from existing frontier models. So, I expect the adoption to be a little more slow and gradual
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 15th of 17 in this recording
-
Their wordsSo, I don't think anyone is necessarily headed in the wrong direction here. They're just making different bets on where the value is.
↗How RAG, Fine-Tuning, LoRA & RL Fit Together in One AI Systemyoutube.com 17th of 17 in this recording
- 7 days earlier
-
Our readingA memory file loaded on every turn works almost like a system prompt and should be used carefully.
Their wordsbut remember that this is the file that will load up every time the employee is talking. So, it's almost like a system prompt. You want to use it carefully.
↗2,000+ Hours of Learning Harness Engineering in 18 Minutesyoutube.com 5th of 7 in this recording
- 7 days earlier
-
Our readingSubjective qualities such as groundedness, usefulness or brand voice need AI or LLM judges.
Their wordssomething more subjective, like whether an answer is grounded in its context, whether it's useful, whether it's matching the brand voice of the company, those are the cases where you need AI judges or LLM judges.
↗2,000+ Hours of Learning Agentic AI System Design in 20 Minutesyoutube.com 11th of 15 in this recording
-
Their wordsMoving to a stronger model can be one of the options, but that's only if the model is your bottleneck.
↗2,000+ Hours of Learning Agentic AI System Design in 20 Minutesyoutube.com 13th of 15 in this recording
- 8 weeks earlier
-
Our readingPrompt, harness and loop engineering build on one another rather than replacing each other.
Their wordsPrompt Engineering...Harness Engineering…Loop Engineering, every few months when a new “AI Engineering’ practice gets introduced, it sends people into a panic. Everyone is worried that they are going to have to start from scratch yet again and change everything about the way they have been building with AI. Part of this is because most content out there treats these as disconnected and un-related concepts but in reality they all build on top of each other.