crypto bro pattern recognition actually maps decently well to the current semis market
most of them have negative engineering skill but have seen hype cycles and have a nose for bullshit
and now some know when they lose their edge
A junior partner at a top vc told me he was worried i wasnt coachable. I said, i dont see what you can teach me given youve never founded or run a successful company.
my freshman fall, Matt gave an incredible talk to @thecrimson business board comp that shaped the trajectory of my entire college career
since then, I’ve heard anecdotes from his classmates and mentees that speak volumes
don’t know him personally, but few better opportunities
We have an uzi of AI output coming from @rippling this year. If you wanna work for me (directly) to tell that story, please DM me. Needs a solid technical foundation, coupled with the X savviness of @HunterBiden.
Silicon will run higher for longer. What people think will happen in 2027 won’t happen until at least 2028.
Individual companies will bust in the meantime, but I’m net long semis for foreseeable future.
Today, @AyarLabs announced it has joined the @nvidia NVLink Fusion ecosystem, introducing co-packaged optics as a foundational building block for hyperscalers and system innovators deploying heterogeneous compute in NVIDIA AI factories.
Press Release:
https://t.co/vNhVcHzLIA
I wrote this to our team today:
“I think this is exactly the area where VC investment is worthwhile: The frontier, where the most informed minds vehemently disagree, and there is already $1bn+ of demonstrated customer spend up for grabs in a company's narrow (and expanding) category.”
Asian nerd and jock (40s) catching up next to me at a restaurant
nerd spent a year in his 30s reaching out to *everyone he has ever known*
first “not close friend from high school” he reached out to after 10+ years of no contact is now his wife
you can just do things
.@Voxel51, powered by @Miris_spatial at CVPR this week.
STREAMING at full fidelity on a workstation-class GPU.
Theatre talks at booth 309 on Friday, June 6 and Saturday, June 7 @ 11am and 3pm
& demo on your phone @ https://t.co/SyX7DBVtzm
“Like the tyrant, the tragic hero is an extreme version of a person who needs to overly regulate the complexity of their own mind.”
- Adam Phillips, “On Giving Up”
@_s0han 1) @paulg wrote a great essay about this (linked below).
2) you will see your network compound over time & likely attribute more to physical proximity than you do now.
3) the main assertion of your post is a good one.
@theamberyang There’s a great @paulg essay about this.
The main thesis is that the trade-offs people make to live in a city are inextricably linked to the personality and values/value props of that city.
“Cambridge with good weather, it turns out, is not Cambridge.”
https://t.co/YHRycsQevM
@martin_casado Question 1 is: Will more value capture from the AI OS come from hardware or software.
Question 2 is: Which incremental features will induce marginal spend at scale? We aren’t at scale yet, despite huge #’s.
Latency as key feature suggests consumer hardware as power broker.
@Romy_Holland a lot of that in Seoul is hidden and pretty inaccessible to foreigners without a friendly citizen as a guide.
I thought the same until my third trip there, when I had made several friends who’ve spent many years there.
Mostly due to networking protocols. TLDR it’s graph theory, hardware/software limitations, and Amdahl’s Law.
To send data between connected (racks of) GPUs, some things need to happen in parallel, and some things need to happen sequentially. Some operations take all the memory on a package but don’t use full compute. Some operations (have to) happen standalone but don’t max out the memory or the compute.
Some operations are redundant (being done by multiple chips or racks simultaneously) in case a chip/rack fails. But you need to wait for the last chip or cluster doing the calculation to “arrive” wherever the next operation is.
There’s a whole host of other reasons: packets drop randomly over Ethernet, flow control has not been super optimized historically, variable packet sizes and every-changing best of breed AI algorithms make optimizations hard and short-lived.
Tons more.
@NTmoney Wemby and Maye/Daniels are particularly difficult for this because of the Fanatics/Panini license handoff.
This is the only Wemby I got (/25) despite being a huge early believer.
Didn’t buy any Maye because I only knew the Contenders Rookie Ticket market.