Agree with Karpathy's thinking that LLMs aren't code, they're stochastic simulations of human spirits. We're still in the 1960s of this tech, and we need to accelerate on refactoring most if not all of our systems for agents.
Andrej Karpathy talked about what AI engineers actually need to know right no
this isn't a roadmap for getting hired
it's a reality check for everyone already in the industry
here's what he says engineers are getting wrong:
> Most engineers treat LLMs as a single paradigm, but there are actually three
> LLMs have their own "psychology" - encyclopedic memory, uneven intelligence, and anterograde amnesia
> the generation-verification loop is the most important concept nobody talks about
> "keep AI on a leash" is not a metaphor. a 10,000-line diff is useless. small, concrete, incremental changes win every time
> every product needs an autonomy slider. from autocomplete to full agent
> "year of agents" is wrong. agents are a decade, not a year
> building for humans is no longer enough
the engineers who thrive won't be the ones who know the most models
they'll be the ones who understand how to work with fallible, powerful systems
and build products humans can actually trust
full interview and a detailed article on what every AI engineer needs to know are below👇
The internet scaled information. AI scaled cognition. Quantum will scale access to nature’s hidden computational layer.
$IONQ $INFQ
My friend Charles Cao taught me about the importance of magic in quantum spacetime: "Without magic, things are a little too simple." Thanks, @walkingthedot, for the well-written article.
quantamagazine.org/entanglem…
Actuators may become the first true bottleneck of the humanoid robot era. Torque density, precision gearing, thermal control, force sensing, and durability will decide who scales. In Physical AI, motion becomes margin.
I got hooked watching this livestream of @Figure_robot this morning. Outside of the pace, it’s a reminder that endurance, consistency, and uptime will be the real labor arbitrage. youtube.com/watch?v=luU57hMh…
$QS is now extending beyond EVs into AI data centers, which is a massive, under-appreciated opportunity. As compute demand explodes, energy density becomes a constraint. If they deliver, they could soon become a critical part of the overall AI infrastructure stack.
Interesting behind the scenes look into the race to build the next generation of automotive batteries.
$QS $SLDP $LIT #Lithium #ElectricVehicles #Batteries
nytimes.com/2022/07/12/busin…
The last tech cycle scaled bits. But the global economy still runs on atoms. Manufacturing alone is ~16% of global GDP, and demand for energy-transition minerals could 4× by 2040. The next tech innovations will emerge when bits start to reshape atoms.
Apple is heading into its next tech wave with smarter AI, new device categories, and a renewed sense of direction. Even setbacks like Titan or Vision Pro will ultimately seed the breakthroughs of its next supercycle. Excited to see all that is to come here.
Genomics has taken time to really get moving, essentially because biology isn’t software. Reading DNA was the "easy" part. Interpreting it, validating it clinically, and acting on it safely requires cheap sequencing, massive datasets, and AI. Only now are those curves crossing.
Incumbents tend to treat new technology as a feature or an efficiency gain, something to bolt on or to automate. The real disruption often comes though when the technology fully unbundles the core business.
After reading through America's AI Action Plan it's abundantly clear that in the AI era, technological dominance will not just be economic, but rather geopolitical. The new arms race isn’t nukes or oil. It’s compute, models, and influence. Whoever builds fastest, leads.
Agentic AI is starting to redefine how we think about cost and value in AI. Instead of pricing compute, we should consider pricing ouput—like labor. ROI should really be measured in business outcomes, not just token usage.