Ishi Singhal

@TanejaPriyal @thecollectiv3e Hey would love to meet up:)

hi i’m priyal! canadian, sf-based looking for more friends in the tech & vc community :) what i’ve been up to lately: - mts @ e3, building ai agents for freight - scaled @thecollectiv3e’s newsletter to 8k readers, writing on the tech behind trending startups - hosting events for founders & engineers - trying every iced vanilla latte in the bay say hi if you want to grab coffee sometime!

@MainzOnX Looking forward to your next article!

@ishi__singhal It’s definitely a tough subject to get into

Finally truly understand kernels

Richard Sutton, the author of the Bitter Lesson, started a new lab @oaklab_ai focused on continual learning. There’s so much happening in this space, we’re going to see agents that evolve with time very soon.

I wish Claude had a skill that could emulate Lovable's design and UI taste

History doesn't repeat itself, but it often rhymes: One of the best historical analogies to open versus closed weight models.

@ruthvik_sl @naval @pratyush_r8 Great content on the Substack! @pratyush_r8 looking forward to following your thoughts and writings:)

@ishi__singhal @naval From my conversations, @pratyush_r8 takes it seriously.

Want to listen to more people like @naval who think deeply yet communicate simply about the fundamental aspects of life. Who should be on the list?

@MollySOShea @adcock_brett at @Figure_robot !

Up in SF this week, who should we interview? Any HQ tours you want to see?

@palak_agarwal6 Would love to join!

now that i can host again i'm starting a bi-weekly writers group at mine! lmk if you'd like to join the first one, i'll have wine and kombucha :)

Why Memory is absolutely critical today - 1. In 11 years, we have only been able to increase memory bandwidth per GPU by ~36x, whereas compute grew by roughly 500x! This is the main reason why memory has become the bottleneck. 2. HBM is most of the cost of a GPU - 58% of BoM for the Blackwell Ultra. 3. Three companies control the production of all the HBM in the world - SK Hynix, Samsung and Micron. 4. HBM revenue is compounding at more than 70% annually, making it an very exciting market to follow.

@Beth_Kindig So insightful! Quick question - why are neoclouds able to deploy these GPUs faster than hyperscalers?

Neoclouds are trying to strike a delicate balance between capturing AI demand and high debt. Below, we examine if $CRWV and $NBIS are a durable extension of the AI trade or a fragile pressure point, $NVDA circular financing and hidden hyperscaler balance-sheet advantages. io-fund.com/ai-stocks/nvid…

While a lot of us only know of Unitree as the backflipping humanoid, the reality is quite different. This article explores the unique advantages Unitree has and how they’re similar to BYD and DJI. Changed my misconception, it’s a must read!