Marcel van der Heijden (@marcel_v_d_h) 's Twitter Profile
Marcel van der Heijden

@marcel_v_d_h

Early stage B2B investor. Happy, never satisfied, always curious.

ID: 844797595

calendar_today25-09-2012 03:35:21

3,3K Tweet

1,1K Followers

1,1K Following

Marcel van der Heijden (@marcel_v_d_h) 's Twitter Profile Photo

I don’t understand why this is not something every deep tech (VC) investor talks about all the time. Awesome piece 👇👇👇. Building the right capital stack for often capital intensive deep tech companies is an innovation in itself venturedesktop.substack.com/p/the-rise-of-…

Marcel van der Heijden (@marcel_v_d_h) 's Twitter Profile Photo

The Trump administration is preparing an historic executive order to reassert American leadership in deep ocean floor mineral extraction, a $100+ billion industry. thehill.com/opinion/526127…

The TLS (@thetls) 's Twitter Profile Photo

'Are LLMs analogous to individual human minds (or, per Sejnowski, those of space aliens) – minds that think, reason and perhaps have their own beliefs, goals and intentions?' Melanie Mitchell: If it quacks like a duck the-tls.co.uk/science-techno…

Marcel van der Heijden (@marcel_v_d_h) 's Twitter Profile Photo

Too funny. Someone's having a bad day working with a programming copilot. "A one-man bug factory. Whispering half-formed solutions into the ears of juniors like a sick, twisted full-stack Rasputin." deplet.ing/the-copilot-de…

Stella Li (@stellalisy) 's Twitter Profile Photo

🤯 We cracked RLVR with... Random Rewards?! Training Qwen2.5-Math-7B with our Spurious Rewards improved MATH-500 by: - Random rewards: +21% - Incorrect rewards: +25% - (FYI) Ground-truth rewards: + 28.8% How could this even work⁉️ Here's why: 🧵 Blogpost: tinyurl.com/spurious-rewar…

🤯 We cracked RLVR with... Random Rewards?!
Training Qwen2.5-Math-7B with our Spurious Rewards improved MATH-500 by:
- Random rewards: +21%
- Incorrect rewards: +25%
- (FYI) Ground-truth rewards: + 28.8%
How could this even work⁉️ Here's why: 🧵
Blogpost: tinyurl.com/spurious-rewar…
Nikkei Asia (@nikkeiasia) 's Twitter Profile Photo

Nikkei has found that research papers from 14 academic institutions in eight countries contained hidden prompts directing AI tools to give them good reviews, concealed from human readers using tricks such as white text or extremely small font sizes. s.nikkei.com/4nuFdn4

Marcel van der Heijden (@marcel_v_d_h) 's Twitter Profile Photo

10 bucks this works nicely for university applications, loan/funding/mortgage applications, jobs, investment memo’s, tax filings, etc.

Marcel van der Heijden (@marcel_v_d_h) 's Twitter Profile Photo

This is not about being unserious or idiotic. It’s either spiteful self-harm or successful price discovery. The simplest explanation …

SemiAnalysis (@semianalysis_) 's Twitter Profile Photo

GPT-5 Set the Stage for Ad Monetization and the SuperApp How ChatGPT will monetize free users Router is the Release AIs will serve Ads Google's moat eroded? The shift of purchasing intent queries semianalysis.com/2025/08/13/gpt…