▶ 10:05 · $SNORKEL · Vincent Chen: Snorkel's thesis - scaling expert supervision is the real data bottleneck

episode briefing

Going In Deep On Data | YC Paper Club

2026-08-20 · 6 company · 13 thematic

sentiment

5 bull0 bear1 neu

speakers

YC partner hosting Paper Club, founder of Focal Systems (computer vision for retail), advocates data-centric AI thesis and highlights $100B+ market cap creation in data category.

volo

Founder of Inception Labs building diffusion language models; Cornell professor; developed Mercury 2 models achieving 1000+ tokens/sec and Tao Forge synthetic RL environment system for real-time voice and coding agents.

vincent chen

Founding team member at Snorkel AI, started Frontier Lab business, leads research on benchmarks and evaluation; PhD from Stanford AI Lab; developed Senior SWEBench and open benchmarks grants program.

quote

“our key thesis is that scaling expertise, actually giving leverage to experts in the field, doctors, clinicians, um journalists, people who actually have the spec in their head for what good looks like is the real bottleneck for building r…”

—vincent chen

now playing · $SNORKEL

$SCALE-AIbullish· highfrancois

Francois: Data businesses created $100B+ market cap despite VC skepticism

VCs believed terminal value of data businesses was zero, but Scale AI and similar companies proved them wrong with over $100 billion in market cap creation over the last decade.

$Y-COMBINATORbullish· mediumfrancois

YC partner calls data companies a top category with $100B+ market cap created in decade

Data infrastructure and labeling companies have generated over $100B in market cap over the last decade despite initial VC skepticism, making it one of YC's best investment catego…

$SNORKELbullish· highvincent chen

Vincent Chen: Snorkel's thesis - scaling expert supervision is the real data bottleneck

The bottleneck for effective datasets is scaling expert supervision — getting judgment from doctors, clinicians, journalists into data programmatically rather than through manual…

$ANTHROPICbullish· mediumvolo

Volo: Anthropic's Opus and Sonnet tie for first on Senior SWEBench coding benchmark

Anthropic's models achieve state-of-the-art on Senior SWEBench, a benchmark for senior-level software engineering tasks requiring architectural decisions and code refactoring.

$INCEPTION-LABSbullish· highvolo

Volo: Diffusion language models achieve 1000+ tokens/sec, enabling real-time voice AI

Diffusion models generate tokens in parallel rather than sequentially, achieving 1000+ tokens/sec on GPUs versus specialized chips, enabling real-time voice applications with bett…

$CBRS···neutral· lowvolo

Volo: Cerebras runs 120B parameter model but diffusion matches speed on GPUs

Cerebras specialized chips enable fast autoregressive models, but diffusion models achieve similar speeds on commodity GPUs without custom hardware.

←