What AI actually does
Mutinex has built what it describes as a “multi-agent system,” where each agent acts as a domain specialist. For example, one agent understands marketing econometrics, another understands competitive pricing theory, another diagnoses model failures.
By combining Tracer, which cleans and makes sense of Hershey’s data infrastructure, with Mutinex’s AI system, Hershey is now able run models in as little as three weeks.
In practice, that means faster iteration on how marketing spend is evaluated and adjusted, rather than waiting for lagging historical reads.
“Most companies don’t have an AI problem. They have a data readiness problem,” said Sarah Martinez, chief commercial officer, Tracer.
</article excerpt>
Instead of the headline, it sounds like they've hired an external company to clean up their ETL pipelines. That seems useful.
I'm going to doubt spooling up <massive LLM> with <appropriate system prompt> is going to be the thing that reduces their analysis time.
I am working with agentic AI on industrial manufacturing data. The speed at which you can get insights and dig into all kinds of rabbit holes is insane. It's just as easy to compare plants so you can make strategic decisions on budget allocation as it is to do root cause analysis on why during a given shift there were so many breakdowns.
And this happens with a natural language interface, instead of Excel (although people of course still want an export to Excel button) or worse: by having to go to the BI analyst, have them change a dashboard and after waiting for a few weeks hope they give you what you want...
Yes, you need to structure your data well. Especially metadata/defintions and accessibility - which is not cleaning up ETL pipelines, although that will help. And obviously have lots of relevant data available already (which was my job before this).
From my experience: fully automated marketing budget allocation... Doubt it. Time to insight reduced >10x? For sure.
I know that articles like this, and the broader dialog, doesn't go too in-depth on how these things work. The phrase, "one agent understand marketing econometrics" from the article makes me wonder what exactly they mean.
This could be anything from "we put a prompt in front of chat-gpt that says 'you are an expert in marketing econometrics'" to "we built a model trained exclusively on marketing econometrics material"
No matter what they actually did, the agent (assuming its an LLM) doesn't understand marketing econometrics, instead it's tuned to produce output tokens that I suppose make more sense when the topic is marketing econometrics.
I'm not an LLM detractor, but I find the kind of thinking that's become prevalent to be so squishy. Humans are great anthropomorphizers and it seems today that no one is attempting to hit the brakes on that instinct. The models don't understand anything, in the way that we commonly use the term understand.
It seems we've confused ourselves because the box that doesn't understand marketing econometrics can produce marketing econometrics analysis and when we ask it why it came to such and such conclusion it can produce convincing explanations.
As an aside, I also feel like I've heard this for 30 years about marketing. Marketing is everywhere, the surveillance economy tracks our every move in more and more invasive ways every day, and still companies go "Aw shucks, we just can't make sense of this data." It reminds me of a time when I was working for Abercrombie and Fitch and there was this massive report our team was partially responsible for generating. 500+ pages, generated everyday, sent to a high speed printer from a COBOL job every morning at 5am so that 10-15 copies could be made for the executives. It had *all the data* and each executive had their own little ritual around which bits they thought were important.
Throughout my career as an engineer I've been asked to get more data, more data, more data. Process the data, analyze the data, create some graphs and tables and help people understand the data.
One thing I've realized is that the people demanding all this data, all this insight, all this analysis, are unlikely to actually need it or use it when available. They are tasked with making decisions, and decisions are scary because you can make the wrong one. They would love to not make the decision and maybe you can find enough data that the choices get cut down to just one. Then if it ends up wrong they aren't to blame, the data made them make that decision.
So all of this surveillance, all of this analysis, all of this data is likely just to make some person feel a bit more comfortable about making a decision.
Right. For the first many decades of computing, recursion was just always the wrong answer for a production software system. (Feel free to provide a counter-example, but please begin with an explanation of how the size of a call stack frame is determined and how exceeding the base allocation is handled on this platform).
So what tree-traversal/quicksort problems tend to measure is how long it's been since you last did CS class homework problems.
Great. Please explain how the size of a call stack frame is determined and how exceeding the base allocation is handled on the particular platform you're proposing to recurse upon.
which city is that? I want to move there with my children.
If you desperately want mediocre chromebook instruction for your children; where your classmates turn in ChatGPT essays as original work, feel free to move to Virginia.
Sublime Text? Sure, doesn't have the long tail of extensions, but surely most people don't need those. The biggest issue with ST being the fact that it costs money...
Even if you don't like the current administration, the rank and file are still out there doing valuable work. The government is more than ICE; it also administers welfare, funds research, collects taxes, and distributes social security payments to the old and infirm.
reply