ignore the rumors, computer science is still cool here's my predictions for 2026

Timestamps: 00:00 Treating RAM like a...Network? 03:55 Lunar Timezones 07:10 Rust vs C++ vs CHERI? 10:58 Slopsquatting 13:40 Non-deterministic ML Compilers? --- 2026 Predictions (simplified): 65% - At least one major cloud provider will release compute offerings with a "tiered" or "burstable" memory class where you explicitly pay less for memory backed by CXL 80% - A published implementation strategy for official Coordinated Lunar Time as an operational standard 40% - A mass produced consumer device ships with a CHERI-enabled processor 70% - A major developer package repository has to temporarily halt registrations for >24hrs 65% - LLVM MLGO lists 3+ ML-guided heuristics

If everyone would “just vibecode” there would be no specialists who actually knows craft and have deep expertise, shout to all, please continue to get better everyday and be obsessed with low level, deep technical things!

Laurie this is utterly off topic, but I have discovered that I think my cat is infatuated with you. Every time I watch one of your videos, he runs up and starts head bomping the screen and rubbing his head all over the monitor. He does that for no other videos. weird little guy

haha aww that's so cute

@lauriewired how do I even get 1% as smart as you. The idea of memory as a service is so interesting, I can def see that happening. Also that walk with this topic was awesome. Need my old CS profs to do that.

'computer science is still cool' shouldnt need to be said but the enrollment numbers say otherwise

feels so true nowadays – "those who don't understand UNIX are doomed to reinvent it, poorly", I'm watching AI teams rediscover process isolation, rate limiting, and retry logic from scratch, damn

I agree that rewriting existing codebases in Rust in order to be memory-safe feels misguided. Watch out for those angry birds. 🪿😆

CS course in 2030: Using AI to troubleshoot non-deterministic compiler bugs.

computer science was never uncool, the people tweeting about it were just briefly louder than the people doing it

Unless you retire, it's never getting bad

who's saying it isn't cool we're literally building the future one neural net at a time

I worked on the Arm Morello project years ago and haven't heard about CHERI since then. I'd be surprised if something hit the market after all these years.

I liked the ML compiler take, prompted me to extrapolate another 5-10 years:

moon time? first we need to fix earth time and use earthtimezone without summer winter time and 29 of feb...

@ThePrimeagen

pin it pls

Great video! Careful of the geese!

@ThePrimeagen Computer science will always be cool!

Cool! Idk about the second one but others seem very much possible. IMO there will be really cool stuff coming in for online PGO with manual or MLGO. Something like they did in OCOLOS. paper -

I just saw your YouTube video and instantly came here to follow.

CS is always cool, however, it will not be cool in a business environment. CEOs from small to big companies, decided that AI can code faster, iteratively. I had experience with a CEO literally ditching project management completely, their idea was to use ai to code, debug and fix

So, DRAM is gonna be the L4 cache. I also think you should do a video at the year end showing how well the predictions hold.

Fantastic 'cast. Thanks for the rational voice on the RAM model. :-)

Always cool yeah

thank god it’s still cool was getting worried I had to pivot to EE

Do you ever wish you were part of Parc?

Check out the ETA10. (on wikipedia). I had just working with the design team (to add Scientific Programming input) since the funding dried up and ETA Systems folded.

Two ideas 1) different types of processors with different architectures and performances on the same supercomputer, and 2) smart central memory where a numerical value (maybe a vector) could be sent to memory and added (maybe multiplied) to a set of memory cells (also a vector

Performance programmers have long had to consider whether memory will reside in the NUMA hierarchy, with very few knobs to tune besides cajoling code ordering or outsource the responsibility to libraries. I know you're discussing CXL-2 , even NUMA isn't handled. Outside of CRAY

Hi Laurie . how are you doing? i hope you are good. could you talk or make series about zero-click exploits ? and make full course about reverse engineer.

Some of these topics get me wondering about the concept of mobile agents: self-contained software entities with embedded code, state, and processing that autonomously migrate between systems without duplication. Think like Star Trek Holograms or Tron 'programs'. Obviously that is super sci-fi, but I've often wondered what form something like that would take in the real world. Perhaps a computing standard where everything runs in a sort of virtual machine, and the virtual machine is deleted with prejudice once the application "leaves" it. (Star Trek Transporters / Suicide Machines?) Curious what you thoughts might be on this.

Where is that place where she's at? It looks like community college.

Laurie’s “Slop Squatting” scenario seemingly/arguably could happen any minute now. How is agentic proliferation going to address this, when agents are so willing to quickly follow gradient descent to complete tasking. I’m sure you verify every single loop with xAI’s massive infrastructure buildout in Memphis, but what’s going to stop the little LLMs from making a mess of things?

Thx a lot Laurie, especially for the C++ v Rust v Cheri part ✌🏻

Whats the alternative...dumb and dumber I guess? CS or CE, never a waste of time, as you still need a big brain and a mind that can speak and understand many languages with ease. That is better than a skull full of jello.