Just watched this excellent keynote about how AI is reshaping software by Andrej Karpathy.
 
I’m by no means a software engineer, but I've extrapolated some broader takeaways from it that can be generalised for anyone simply interacting with software, i.e. people like you and me: ⬇️

Andrej Karpathy's (@karpathy) keynote yesterday at AI Startup School in San Francisco. https://t.co/UM1wfFs98S

Y Combinator

Just watched this excellent keynote about how AI is reshaping software by Andrej Karpathy.
 
I’m by no means a software engineer, but I've extrapolated some broader takeaways from it that can be generalised for anyone simply interacting with software, i.e. people like you and me: ⬇️ 

https://x.com/ycombinator/status/1935496106957488566

illiterate enjoyer of crypto and AI art | linktr.ee/buffets.eth

7) Automation in software will not be binary, but will occur along a spectrum between being augmentative and agentic. You will have to think about when AI is best used to directly augment you vs. when AI is better used to simulate you—or a different version of yourself.

8) As more content on the Internet becomes repurposed to be readable by AI agents, a new dimension of media literacy opens up. Questions we’ll need to start asking ourselves: Is this piece of content I’m looking at meant for AI rather than humans? How might it influence the behaviour of an LLM? Whose interests are being served when a particular type of content or information is fed to an LLM?

5) Human work will lean more towards verification than generation. But even the act of verification may be automated. I therefore prefer the concept of taste—the ability to discern what is useful, aesthetic or simply good, cultivated not just from aggregated data but also through wrestling with the subjectivities of our lived experience. When anything and everything can be easily generated by AI, good taste becomes even more fundamental for verification, curation and management. 

6) Since good outputs from AI require well-defined context windows, the ability to recognise, parse and frame different contexts well will be a superpower. Context-switching becomes the new multi-tasking.

1) AI will become analogous to public utilities like electricity—pervasively distributed and accessible on-demand. You are thus handicapping yourself if you don’t use it. 

2) Unlike other tech like computing and the Internet, the diffusion of AI has been largely consumer-first rather than institutions-first. You cannot rely on your employer, your school or the government to prompt you into using AI. By then, it will be too late.

3) How you use AI will be your principal differentiating factor. And how you use AI is in turn a function of how you think, express yourself and engage with others. These are qualities that have long been important before AI, and will continue to be so post AI. 

4) Natural language will become our main interface with software. Your ability to communicate will therefore directly correlate with your ability to make and do things with computers.