Jared Roesch

I have been verbally sharing my perspective & insights on the AI market having worked in the space for 6 years now. Yet another person told me I should be writing more often. Stay tuned for more! open.substack.com/pub/deepdreams…

I have some to hand out as well! Come find me !

They are really going all in on #NVIDIAGTC in San Jose … even the grocery stores are rebranding.

What three languages are joining cuTile Python? Find out Monday at 4PM at @NVIDIAGTC 2026. nvidia.com/gtc/session-ca…

GPU ♥ LPU: Everything You Wanted to Know I’m joining David Senra (@FoundersPodcast) at @NVIDIAGTC for a conversation about the reality of modern inference. This is your opportunity to learn why Nvidia and Groq partnered together, and what it means for the future of inference.

@Yuchenj_UW Good luck in the next chapter ! The startup journey ending one way (or another) was/is a strange experience I wish you the best navigating it, always happy to chat!

I have decided to step down as CTO at Hyperbolic. Leaving a company you co-founded and poured your heart into is not easy. So many moments still feel vivid: launching our AI inference product for open-source models and seeing tens of thousands of developers sign up in a week; the week we were hit by a massive DDoS attack and the entire engineering team fought around the clock until we won; the day we launched the GPU platform and watched ARR take off. There were also hard moments. That’s the nature of building a startup. I’m grateful for all of it. What I’m most grateful for is the team. Thank you for your trust. Most startups never build something people want. I believe we did. You should be proud of yourselves. I will look forward to seeing your success. What’s next for me? I’m still figuring it out. I believe this is the most extraordinary moment in human history. We’re standing at the edge of the Singularity. AI will reshape everything, and I still feel the same excitement I felt when I first fell in love with AI. Time to start over. Time to climb another mountain. Thank you to everyone who has been part of the journey, — Yuchen

@chamath The main challenge is there are other bottlenecks (ala Amdahl’s law) which block immediate realization of the coding gains, review, product planning, testing, release cycles, just like with cloud it will take ~years for the technology gains to fully permeate the dev lifecycle

Not a single meaningful company has yet to say that they are making at least 2x+ from all of this incremental money they are spending since even last Fall! When ROI?

CUDA 13.2: cuTile Ampere/Ada support and perf gains, CCCL adds modern CUDA RT features, TopK, and Segmented Scan, and Nsight Python. Learn more at GTC next week! developer.nvidia.com/blog/cuda-13-2…

CUDA 13.2 is now live, docs.nvidia.com/cuda/ we dropped a new version of cuTile, and Tile IR with number of new features in the DSL, and a large number of performance and stability improvements in Tile IR. Happy Hacking and looking forward to chatting at GTC this coming week!

@RanjitJhala @jesseproudman True but at least you can go to beach or cry into a burrito

@roeschinc @jesseproudman LOL, CA says "hold my beer"...

Once you’ve taxed all the founders and executives out of Washington state, and then AI displaces all the high income software engineering roles who have been bolstering the regions economy, property market and sales tax base, who’s going to be left holding the bag? Where will the tax cannon be pointed next? This is going to get pretty ugly faster than I think people realize.

I have considered moving home the last couple years but overall like WA in many ways and there are distinct economic benefits post acquisition but this makes it much more palatable to move home without feeling like it’s a big trade monetary trade off, especially when the delta represents money I need/could use to help my parents retire.

@roeschinc @jesseproudman Yuuuup

@frontier_foid In my memory this is comparable to FA3 on H100 in terms of flops, FA4 is 1.6pflops~ iirc so a gap there but not using all the same ticks

this is certainly *an* implementation of flashattention. not a fast one, but an implementation nevertheless.

@frontier_foid This is intended as a tutorial, getting true SoTA is another level of effort / complexity regardless of implementation language / DSL

@ArtiIntelligent You want a backwards implementation or auto-diff?

@roeschinc What about the backwards pass? How do we use this to train models???

There is a great blog from my colleagues on how to use cuTile to write an implementation of FlashAttention, check it out: developer.nvidia.com/blog/tuning-fl…!

to be honest, the best part of making our wedding registry is that it has forced us to really think about what we need to make our place feel like home

@pavpanchekha Yeah I haven't tried 5.3 via the CLI yet, but have it installed and ready, will try compare and contrast in the next few weeks

@roeschinc I feel the same way about Codex 5.3, it's become a very capable numericist.

I know the "inflection point" tweets have been going so strong about agentic coding the last few months, but over the last few weeks I feel like they have finally hit a level of capability where they can productively work with little intervention in code bases like compilers. It has been an insanely fun/interesting experience to watch me and another engineer implement the majority of a large feature while writing almost no code directly, and mostly working on design docs, and PLAN.md files. This has all been on Opus-4.5/4.6 in Claude code. I am starting to get excited about being able to evolve how we do engineering at scale using these tools!

@junrushao 😂

@roeschinc Claude code is now doing slack design debates for me. I feel myself so useless lol

@tiffanyyan_x skiing

In a way, I think Seattle lacks ambition Ambition in Seattle looks like performing well in high school, getting into uw cs, and getting internships and jobs in big tech Where does the ambition go after age 22, when they get the job?