AI Clouds are not all the same. @itunpredictable of @AmplifyPartners shares (in explicit detail) how @modal is just "built different".

RT @Beezer232: I spend a lot of time thinking about what makes a #VC a franchise fund which is why I was so excited to have @dhaliwas & @da….

Join millions who have switched to Grok.

You can read more in depth about cancer’s 95% problem in my first blog post in a while:

Yesterday Tahoe announced a $30M Series A led by us @AmplifyPartners. Their next milestone: 1 billion cellular data points by the end of 2026. We're thrilled to support @nimaalidoust, @iamjohnnyyu, @genophoria, and @kevansf as they chart a new course for precision medicine.

These ML models aren't just finding needles in haystacks - they're revealing we've been looking in the wrong haystacks entirely. Human intuition shaped by limited datasets might be systematically missing patterns visible only at sufficient scale.

The results are already promising. Using models trained on Tahoe-100M, they've identified drug combinations targeting 40% of colorectal cancer patients (vs typical 10-15%). What's fascinating: their discoveries contradicted veteran researchers' intuitions about patient selection.

Then there’s the tracking challenge: how do you measure each cell's reaction in this multi-cell mix? Most labs use artificial barcoding, but Tahoe discovered natural genetic barcodes - unique patterns that let AI map every cell back to its original patient without modification.

"Growth balancing" - mixing cells in proportions that prevent any one type from dominating.

"Mosaic tumors" - pooling cells from diverse cancer patients into composite tumors that represent broad genetic spectrums.

How did they build it? Two breakthrough innovations:.

Tahoe recently released the game-changing Tahoe 100-M dataset: 100M drug-cell interactions and 60,000 drug-patient interactions. It's 50x larger than all public drug-perturbed data combined. LLMs have the entire internet corpus - now scientists have Tahoe-100M.

The concept: Virtual Cells. Digital representations of life's fundamental unit that can predict drug responses without manual lab testing. But building these models requires massive amounts of high-quality data showing drug-cell interactions across diverse patients.

@tahoeai is solving the 95% problem by turning cancer into a data problem. Instead of testing molecules one by one in isolation, they're building comprehensive digital models of how drugs interact with real human biology.

Most AI biotech companies focus on the top of the funnel - generating more drug candidates faster. But if 95% of these "successes" will still fail, the bottleneck isn't creating more candidates; it's predicting which ones actually work in patients.

Basically, we're great at designing molecules that fail spectacularly in humans.

The translation problem is brutal: a drug works on a single protein → fails in a cell → fails in mice → fails in humans. We’re designing more molecules than ever, but only 1 in 10,000 that look promising actually help patients.

Ask any scientist about curing cancer and you'll get mixed signals. Good news: cancer death rates have dropped dramatically over 20 years. Bad news: 95% of cancer drugs still fail in clinical trials.

RT @Forbes: The California-based startup Tahoe Therapeutics, now valued at $120 million, has developed a scalable way to quickly generate c….

Financing news! Excited to share that @AmplifyPartners has led a $30M financing for @tahoe_ai - the latest project in our growing digital biology portfolio. Rigorous science + powerful computing will shape the next decade of drug development, and @nalidoust, @iamjohnnyyu,.

RT @luiscape: We just launched GPU memory snapshotting on @modal_labs in alpha. Speed up cold boots by up to 12x 😇. If you're deploying AI….