The alpha, twice a week
That was 3 min.
Get the best new moments in your inbox every Wednesday and Saturday. It's free.
Check your inbox
We sent a link to your address. Click it to finish signing up. Not there in a minute? Look in spam.
or
The Logan Bartlett Show · 3 min · Aug 29, 2025
Today’s LLMs are powerful but limited by finite data and algorithms
Marc Benioff
Marc Benioff pushes back on claims that apps are ending or current models amount to AGI.
The episode’s best minutes
Full analysis- 2:19AI agents are filling Salesforce’s neglected sales pipeline3 min
- 12:56Benioff expects customer work to split evenly between agents and peoplePro pick 4 min
- 38:07Today’s LLMs are powerful but limited by finite data and algorithms3 min
- 5 more moments in the full episode
“we forget these large language models these are relatively finite algorithm sets with regular relatively finite data sets and those two things provide you some functionality but not all functionality”
Marc Benioffat 39:08 · The Logan Bartlett Show
Takeaways
- He sees current AI progress as evolutionary, not revolutionary.
- Finite algorithms and datasets provide some, not all, functionality.
- Human creativity and insight remain important advantages.
About this segment
Marc Benioff pushes back on claims that apps are ending or current models amount to AGI. He argues that finite training data and algorithms limit what LLMs can do, leaving room for human creativity and new model capabilities.
Marc Benioff is cofounder and CEO of Salesforce.
Founded Salesforce in 1999, pioneering cloud computing and building one of the world's leading enterprise software companies.
Logan Bartlett, Host · Software Investor, Redpoint Ventures
Backs enterprise software companies like Ramp, Cribl, Cyera and Legora.
From “Marc Benioff Predicts Half of Conversations Will be With AI Agents Next Year”, Aug 29, 2025. Watch the full episode
Answers “When will we get AGI?” See all 26 answers
Answers “Is AI overhyped?” See all 11 answers
Thanks. We'll look at and fix what's off.
Transcript
From the transcript The Logan Bartlett Show publishes with the episode.
- 38:14
well I think that there's you know unfortunately we have this kind of clarinet you know I would say reality distortion field that could create it around the time we've talked about a year ago we're all of a sudden the sea of carna seems like a very new sky I've never met I've had one I'd seem wonderful but his messages changed like 10 times and 10 months and the stuff that he said was truly he's kind of walked back is not true I see on the other side there's people in my industry even some of my peers to talk about the end of apps or the end of crud day you know that enterprise apps are just crud databases it's like whoa I I just think that there's a lot of kind of crazy things I mean I think even some of the talk about AGI may be a two augmented that you know these large language models are really amazing but they're not AGI and I
- 39:09
was like this this weekend I was like figuring out how might communicate this to people because we forget these large language models these are relatively finite algorithm sets with regular relatively finite data sets and those two things provide you some functionality but not all functionality and as evidence of that there's this really cool medical study now that our brains start to create immune cells when we're near sick people and as jokie dot twitter well I kind of missed that feature you know gpt 5 I don't think it does that quite yet so we still have an advantage but I think at some level because you're talking about these finite algorithms and finite data sets which is kind of what all the AI is built on these the same data set these LLMs which is why they're all kind of seems like they're hitting this upper limit where you can see like gtp 3 or 4 might have been
- 40:03
more revolutionary maybe gtp 5 is more evolutionary you know that there's still some cool things happening but it's not these big transformational leaps I think that we could kind of then make a statement of you know where we're going to need another set of models and another set of capabilities before we get that next set and that when you're dealing in the finite like the finite data and the finite algorithms we're infinite you know we're infinite beings we have the ability to have a level creativity and inspiration and insights that's beyond that that's why that example of like hey we're creating an immune we have an immune system you know we're creating immune cells we don't even realize we're creating immune cells we're tapped it into a level of energy and capability that's beyond the finite data set so I think we have to keep that you know front of mind that we
- 41:00
I think we still are an important part of this world that we're in it's an optimistic note in on mark thank you for doing this it's great to see you again Logan I hope you're coming to dream force