Betting on prompt engineering seems like betting on hybrids.
Uncategorized
5
Posts
2
Posters
1
Views
-
Betting on prompt engineering seems like betting on hybrids. Which are selling very well right now. But if you think you have a decade of career left, ask yourself this question:
What is the value of “prompt engineering,” when we get to the following inflection point:
“Siri, make me a copy of Salesforce for my business, and after consulting legal precedent, make the most minimally invasive changes to insulate me from accusations of theft.”
Actually, that’s just the warm-up:
1/2
-
Betting on prompt engineering seems like betting on hybrids. Which are selling very well right now. But if you think you have a decade of career left, ask yourself this question:
What is the value of “prompt engineering,” when we get to the following inflection point:
“Siri, make me a copy of Salesforce for my business, and after consulting legal precedent, make the most minimally invasive changes to insulate me from accusations of theft.”
Actually, that’s just the warm-up:
1/2
What is the value of software companies when we reach that point?
Sure, anyone will be able to prompt planet-scale SaaS into existence. Open AI, Google, and especially MSFT will see to that, while xAI is cornering the market on digital mail-order brides.
But anyone includes customers, no? If the arc of LLMs bends towards being reliable enough to prompt production SaaS into existence, there is a business model problem to solve as well as a whither-labour problem to solve.
2/2
-
What is the value of software companies when we reach that point?
Sure, anyone will be able to prompt planet-scale SaaS into existence. Open AI, Google, and especially MSFT will see to that, while xAI is cornering the market on digital mail-order brides.
But anyone includes customers, no? If the arc of LLMs bends towards being reliable enough to prompt production SaaS into existence, there is a business model problem to solve as well as a whither-labour problem to solve.
2/2
@raganwald I would have been extremely skeptical about LLMs being able to be cajoled into making software at scale a year ago, but I am pretty astounded by how good the best agentic AIs have gotten and how much they can do on their own...they look stupid while iterating and making tons of mistakes, but I also look stupid when I'm iterating and making tons of mistakes. Closing the loop and allowing them to read the errors, search, and try again really does unleash something impressive.
-
@raganwald I would have been extremely skeptical about LLMs being able to be cajoled into making software at scale a year ago, but I am pretty astounded by how good the best agentic AIs have gotten and how much they can do on their own...they look stupid while iterating and making tons of mistakes, but I also look stupid when I'm iterating and making tons of mistakes. Closing the loop and allowing them to read the errors, search, and try again really does unleash something impressive.
@swelljoe And even if they weren’t that impressive, this use case is remarkably compelling:
What if the AI companies become the only software companies? Think of their ability to rent-seek when all automation runs through them!
It’s worth juggling and tuning and adjusting heuristics and hiring tens of thousands of humans to help fine-tune the results.
-
@swelljoe And even if they weren’t that impressive, this use case is remarkably compelling:
What if the AI companies become the only software companies? Think of their ability to rent-seek when all automation runs through them!
It’s worth juggling and tuning and adjusting heuristics and hiring tens of thousands of humans to help fine-tune the results.
@raganwald the AI arms race starts to make sense. Why wouldn't you spend a trillion dollars if you can replace every software company forever?
Gli ultimi otto messaggi ricevuti dalla Federazione
-
A report on Weyhro C2, a new offensive toolkit advertised on underground forums.
The toolkit appears to be the work of the individual behind the (now-failed) Weyhro ransomware from March this year.
https://lumma-labs.com/weyhro-c2-because-ransomware-wasnt-paying-the-bills-anymore-b136fd7ef100
-
@alsivx Però fa come tutti gli altri, ossia che non si sposta qui. Diamine, dici che la roba popolare/closed source è una schifezza e sta rovinando internet, allors perché non vieni qui?!
-
@giacomom In verità Mastodon è uno di una serie di social, il perché non vengano adottati in massa, è per varie ragioni che sarebbe complesso elencare e spiegare qui ma comunque ci sono già delle istituzioni che lo usano. Oltre al parlamento di qualche stato dell'unione, la Commissione Europea ha una sua istanza:
-
Chiudiamo la giornata in bellezza.
-
Brian Eno News (@dark-shark.bsky.social)
https://bsky.app/profile/dark-shark.bsky.social/post/3m7vlk4mzxk2y
> Nico: Live at Reims Cathedral, France, December 13, 1974 - featuring John Cale (viola, keys, vocals), Brian Eno (synths, vocals), Mike Oldfield (guitar, bass), Kevin Ayers (vocals), and Robert Wyatt (drums) #FLAC #ChristaPäffgen tinyurl.com/3mujhbdw
-
Rettifico, l'intero canale è oro colato. Ogni nuovo video è meglio del precedente.
-
@raganwald the AI arms race starts to make sense. Why wouldn't you spend a trillion dollars if you can replace every software company forever?
-
•
verso l'inverno
