What occurs while you take a billion {dollars} of pc energy and let it analysis a query for twelve entire minutes? Much more than for those who use that very same energy for just some seconds, that is for positive.
Since ChatGPT Agent is so new, I have been exploring its capabilities and attempting to know what sorts of issues it is best suited to. This time, I requested it to check cloud storage providers. It took twelve minutes to provide you with a solution.
Additionally: 9 programming tasks you shouldn’t hand off to AI – and why
As a result of it is a studying train, I additionally requested ChatGPT 4o to do the identical factor. Our previous acquainted chatbot responded in about two seconds. There was undoubtedly a distinction.
Hold studying to the top. As soon as I add NotebookLM to the combination, the outcomes go from attention-grabbing to astounding…
Analysis cloud storage providers
The immediate I gave each ChatGPT Agent and ChatGPT 4o was this:
I have to retailer 10 terabytes within the cloud. The info can be synced from two customers and two native servers. Discover me costs, and evaluate and distinction the benefits and drawbacks of every choice.
The primary huge distinction is that Agent requested for extra clarification.
This, in flip, resulted in my fascinated about my wants extra particularly, after which telling Agent that I wished cloud storage providers appropriate with my Synology servers.
Have a look at all of them, however they need to have the ability to settle for file syncing. For the servers, they need to be appropriate with Synology servers.
ChatGPT 4o didn’t ask for any clarification, so a number of the suggestions it made weren’t appropriate with my servers.
You can argue that wasn’t the fault of 4o, however actually of the particular person writing the immediate. And you’ll be proper, type of. Positive, had I added that as a criterion, it is attainable that 4o would have restricted its response to Synology-compatible choices. However that pushes the work again on me, not on the AI.
Our entire aim in studying what these programs can do is to learn the way they can assist us, not how they will put extra calls for on us. Agent helped as a result of it requested for some clarification. It elicited a extra full instruction than what I initially gave it.
Additionally: The best AI for coding in 2025 (and what not to use)
That is the identical sort of conduct you’d need from an assistant or a staff member. As a supervisor, I’ve usually inspired my staff to ask questions once they want clarification. The very last thing I would like is to have somebody go off on a challenge for a day or so, solely to return again with ineffective outcomes as a result of they could not be bothered or had been too intimidated to ask questions once they wanted extra info.
As a frontrunner, I have been lucky that the majority of my staff members have been extremely interactive. However over time, I’ve had one or two staff who’ve all the time stubbornly wished to go it on their very own, with out exterior enter, clarification, or steerage, regardless of my encouragement to be extra engaged with the staff. These tenures by no means ended totally positively.
Now, right here, Agent is doing simply the type of clarification interactivity you’d need from an assistant. Immediately, that places a plus in its column vs. the ChatGPT 4o LLM.
Agent gave me an inventory of 17 providers in a desk that contained the plans it was describing, worth for as near 10TB/mo because it might discover, and a few essential notes. For some providers, like AWS and Google, it described totally different variants of their choices.
ChatGPT 4o additionally gave me again a desk, but it surely solely listed six choices.
Each offered summaries of benefits and drawbacks for the providers they spotlighted. Agent did so after the desk, and after its evaluation and proposals part. ChatGPT 4o opened with the benefits and drawbacks part, which made it a little bit more durable to determine the scope of the general response instantly.
Agent’s response began off with a really useful common concerns part that offered a strong and workable framework for the assorted choices.
Agent then adopted up with its model of the benefits and drawbacks dialogue.
Agent then wrapped up with a selected set of suggestions by evaluating the assorted providers.
Lastly, Agent offered a top-line abstract that gave strong steerage for additional exploration.
Total impressions
As is usually the case with primary ChatGPT 4o, I used to be left feeling like I had solely been given a cursory, fast reply. I obtained some insights on what to discover, however felt like there was nonetheless plenty of work to be finished.
Agent gave me again a completely workable briefing. It requested for clarification concerning how I wished to make use of the providers, after which tailor-made its response round these wants.
Additionally: Hacker slips malicious ‘wiping’ command into Amazon’s Q AI coding assistant
I’m acquainted with a lot of the providers Agent mentioned, and whereas I did not double-check all the present pricing, the service descriptions had been correct and so was the pricing, not less than for the providers I am acquainted with.
Google’s storage pricing, for instance, was sourced instantly from a Google support page with pricing.
In my first article on Agent, I confirmed how Agent wasn’t all the time correct or full in its responses, but it surely might generally shine. That is an instance of a challenge the place it does, certainly, present appreciable assist.
I’ve gone down the analysis rabbit gap on enterprise-level storage earlier than (I’ve 60TB within the cloud), and it is all the time been a protracted haul. If Agent continues to develop in reliability, I can undoubtedly see this changing into a really useful adjunct instrument for analysis and evaluation.
Bringing NotebookLM into the combination
Regardless of understanding the dangers of crossing the streams, I made a decision to present the Agent report back to NotebookLM. I first used the share button on the finish of the Agent report, however the obtain button did not do something.
So then I simply chosen the entire report and copied and pasted it as textual content into NotebookLM. That sacrificed the desk formatting, however not less than I used to be capable of get the knowledge into NotebookLM by way of its paste textual content choice. Then I requested NotebookLM to generate certainly one of its amazing conversations.
NotebookLM took about ten minutes to generate its audio overview, however the outcomes had been surreal, in a great way. NotebookLM took the ChatGPT Agent report and ran with it, broke it down into its personal comprehensible model, and even added some worth on high of what Agent offered.
This is likely to be AI’s first magical combo. Use Agent to do some digital desktop pc utilization after which present a deep analysis report. Then increase that report with an audio overview generated by NotebookLM. The ultimate overview ran 14 minutes.
Additionally: My 8 ChatGPT Agent tests produced only 1 near-perfect result – and a lot of alternative facts
So, in lower than 45 minutes (12 for Agent, 10 for producing the podcast, and 14 to hearken to the dialog), it is attainable to get a really robust overview of no matter subject you set Agent free upon.
I do know I come throughout fairly suspicious of AI, the potential risks to society, and its boneheaded strikes. However combining Agent with NotebookLM’s dialog mode is really subsequent stage, not less than on this take a look at case.
- You possibly can view the total ChatGPT Agent session here.
- You possibly can view the ChatGPT 4o transcript here.
- You possibly can hearken to the NotebookLM dialog here.
Have you ever tried utilizing ChatGPT Agent but? What sorts of duties do you assume it is best suited to? Do you favor the fast-but-shallow strategy of 4o, or the extra methodical deep dive from Agent? Have you ever ever paired instruments like ChatGPT and NotebookLM for analysis? How do you see this evolving in your individual workflows? Tell us within the feedback beneath.
You possibly can observe my day-to-day challenge updates on social media. You’ll want to subscribe to my weekly update newsletter, and observe me on Twitter/X at @DavidGewirtz, on Fb at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.