Saturday, May 17, 2025
HomeBig DataNot all the things wants an LLM: A framework for evaluating when...

Not all the things wants an LLM: A framework for evaluating when AI is smart


Be part of our every day and weekly newsletters for the newest updates and unique content material on industry-leading AI protection. Be taught Extra


Query: What product ought to use machine studying (ML)?
Mission supervisor reply: Sure.

Jokes apart, the arrival of generative AI has upended our understanding of what use circumstances lend themselves greatest to ML. Traditionally, we’ve got all the time leveraged ML for repeatable, predictive patterns in buyer experiences, however now, it’s attainable to leverage a type of ML even with out a complete coaching dataset.

Nonetheless, the reply to the query “What buyer wants requires an AI answer?” nonetheless isn’t all the time “sure.” Giant language fashions (LLMs) can nonetheless be prohibitively costly for some, and as with all ML fashions, LLMs are usually not all the time correct. There’ll all the time be use circumstances the place leveraging an ML implementation is just not the fitting path ahead. How will we as AI undertaking managers consider our prospects’ wants for AI implementation?

The important thing issues to assist make this determination embody:

  1. The inputs and outputs required to meet your buyer’s wants: An enter is offered by the shopper to your product and the output is offered by your product. So, for a Spotify ML-generated playlist (an output), inputs might embody buyer preferences, and ‘preferred’ songs, artists and music style.
  2. Mixtures of inputs and outputs: Buyer wants can fluctuate based mostly on whether or not they need the identical or completely different output for a similar or completely different enter. The extra permutations and combos we have to replicate for inputs and outputs, at scale, the extra we have to flip to ML versus rule-based programs.
  3. Patterns in inputs and outputs: Patterns within the required combos of inputs or outputs assist you determine what kind of ML mannequin it’s good to use for implementation. If there are patterns to the combos of inputs and outputs (like reviewing buyer anecdotes to derive a sentiment rating), take into account supervised or semi-supervised ML fashions over LLMs as a result of they is perhaps more cost effective.
  4. Value and Precision: LLM calls are usually not all the time low cost at scale and the outputs are usually not all the time exact/actual, regardless of fine-tuning and immediate engineering. Typically, you might be higher off with supervised fashions for neural networks that may classify an enter utilizing a hard and fast set of labels, and even rules-based programs, as a substitute of utilizing an LLM.

I put collectively a fast desk under, summarizing the issues above, to assist undertaking managers consider their buyer wants and decide whether or not an ML implementation looks as if the fitting path ahead.

Kind of buyer wantInstanceML Implementation (Sure/No/Relies upon)Kind of ML Implementation
Repetitive duties the place a buyer wants the identical output for a similar enterAdd my electronic mail throughout varied varieties on-lineNoMaking a rules-based system is greater than enough that can assist you along with your outputs
Repetitive duties the place a buyer wants completely different outputs for a similar enterThe client is in “discovery mode” and expects a brand new expertise after they take the identical motion (resembling signing into an account):

— Generate a brand new paintings per click on

StumbleUpon (do not forget that?) discovering a brand new nook of the web by random search

Sure–Picture technology LLMs

–Suggestion algorithms (collaborative filtering)

Repetitive duties the place a buyer wants the identical/related output for various inputs–Grading essays
–Producing themes from buyer suggestions
Relies uponIf the variety of enter and output combos are easy sufficient, a deterministic, rules-based system can nonetheless give you the results you want. 

Nonetheless, should you start having a number of combos of inputs and outputs as a result of a rules-based system can’t scale successfully, take into account leaning on:

–Classifiers
–Subject modelling

However provided that there are patterns to those inputs. 

If there are not any patterns in any respect, take into account leveraging LLMs, however just for one-off eventualities (as LLMs are usually not as exact as supervised fashions).

Repetitive duties the place a buyer wants completely different outputs for various inputs –Answering buyer assist questions
–Search
SureIt’s uncommon to return throughout examples the place you may present completely different outputs for various inputs at scale with out ML.

There are simply too many permutations for a rules-based implementation to scale successfully. Think about:

–LLMs with retrieval-augmented technology (RAG)
–Resolution timber for merchandise resembling search

Non-repetitive duties with completely different outputsEvaluate of a lodge/restaurantSurePre-LLMs, the sort of state of affairs was tough to perform with out fashions that had been educated for particular duties, resembling:

–Recurrent neural networks (RNNs)
–Lengthy short-term reminiscence networks (LSTMs) for predicting the subsequent phrase

LLMs are an important match for the sort of state of affairs. 

The underside line: Don’t use a lightsaber when a easy pair of scissors might do the trick. Consider your buyer’s want utilizing the matrix above, making an allowance for the prices of implementation and the precision of the output, to construct correct, cost-effective merchandise at scale.

Sharanya Rao is a fintech group product supervisor. The views expressed on this article are these of the creator and never essentially these of their firm or group.


RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Most Popular

Recent Comments