Cofounder @vincerebio bridging AI and Biology. On a mission to conquer age related decline.
Boston, MA
- Tweets 5,882
- Following 742
- Followers 1,334
- Likes 9,729
Quit yer pitching and build.
It’s incredible how AI is blowing up the VC model from both sides
Founders can disintermediate VCs by going direct to their LPs, especially exited operators
VCs can disintermediate founders by incubating their own agentic swarm led companies
What does startup creation look like when founders don’t need VCs and VCs don’t need founders?
2
142
Hey @NedNadima need some more mines for your collection?
I listed my island in Tanzania for sale and the offers started rolling in. Gold mines, houses, cows. One gentleman offered $1 with a promise to pay the rest when he could.
I admire the creativity.
The @NewYorkPost wrote the whole thing up.
Lupita started as a wacky idea after a family safari in 2004. Building a resort on an uninhabited island accessible only by boat sounded like a terrible plan. It turned out to be one of the great joys of my life. We just do not use it enough anymore.
$7.9 million or best offer. The cows are still under consideration.
1
5
227
Draper team are awesome partners! If you can, I highly recommend you find a way to work with them.
Our portfolio company @VincereBio just received a $2.7 million SBIR Phase II award from the NIH's NIDDK to advance its USP30 inhibitor through IND-enabling studies for acute kidney injury.
AKI affects over 13 million people worldwide each year, and no approved therapy directly modifies the disease. Vincere's approach clears damaged mitochondria from injured kidney tissue so it can recover.
This is Vincere's second organ from the same platform: their lead program, backed by a $5 million award from The Michael J. Fox Foundation, is advancing toward the clinic for Parkinson's. Over $8 million in non-dilutive funding from the NIH and MJFF now validates the science.
Congratulations to @Dr_Behrouz, @andy_d_lee, and the Vincere team. 🤝
108
Are AI's inherently good? Are humans?
Perhaps both will always require a blend of early & often reinforcement learning + sufficient incentives for aligned behavior.
Maybe the training and incentives aren't even all that different.
1
67
Looking forward to taking a second molecule to the clinic!
Repairing mitochondria has need for a few unique optimized profiles :)
prnewswire.com/news-releases…
1
1
5
132
Agents are a fascinating blend of hardware, harness, context, and model.
Change any piece and they are fundamentally different beings.
I suppose that’s true for all of us.
1
88
Repairing mitochondria may just solve all of these.
Enough of the cure talk. Where are the cures for these diseases?
x Alzheimer's
x Parkinson's
x Heart and vascular diseases
x Diabetes
x Hypertension
x Most cancers
x Multiple sclerosis
x Most autoimmune diseases
X Kidney diseases
x Liver diseases
x Lung diseases
x Depression
x ALS
and a long list of others
1
5
291
I’m glad to see Dario and others drinking the cool aid of AI curing all disease in 5-10 years.
It clearly won’t, because real world validation just takes time.
But every believer pushing resources to find and deliver cures puts us a step closer and a day sooner.
Do good.
1
1
119
When your probabilistic partner changes its story,
do you have raw data and ability to check receipts yourself?
Knowing when and how to enforce deterministic validation will be big business for a while.
2
126
Aging used to be something medicine worked around. Increasingly, it’s something we can work on.
Mitochondrial quality control is one piece of that puzzle, and the subject of a chapter Spring Behrouz and I contributed to Frontiers of Longevity Science: barzilaiconsulting.com/fols
4
214
Glad to see this launch! Spring and I wrote the chapter on mitochondrial interventions covering biomarkers, medicines, replacement, and more.
Looking forward to reading the rest on this amazing evolving field.
IT'S OUT. 🎉
FRONTIERS OF LONGEVITY SCIENCE @SpringerNature
Leading scientists, clinicians, and entrepreneurs on where the field stands, and where it will go.
Science today. Longevity tomorrow. 🚀
Like and repost to help it travel!👇👨⚕️
🔗linkedin.com/posts/agingdoc_…
1
3
451
The best time to plant a tree…
“This is absolutely the single greatest time to start a company, I’m jealous of all of you” @JensenHuang at Startup School
4
184
Where's @HPC_Guru when we need them?
Can someone explain the acronym FLOPS for me?
F - floating-point
L - ?
O - operations
P - per?
S - second
147
Andy D Lee retweeted
I love prediction markets, but I worry they've confused young people about how the stock market and American economy work overall
Prediction markets are zero-sum. In the end: winners pay out losers, no net value is created
They have their place: prediction markets forecast outcomes better than polling & allow contrarians to monetize truth-seeking
But the stock market, capitalism, and the American economy are absolutely not a zero-sum game
Our economy is not a poker table where your win has to necessarily come at someone else’s loss
When someone invents the impossible, unlocks resources previously unavailable, finds a more efficient way…
Or at the grandest scale: opens up an entire new frontier, like space:
That adds so many new chips to the table. And without taking from anyone else’s chipstack
21
18
162
69,757
“Coding” in production has downsides.
Sometimes you ask the agent to change itself and now you’re back to mucking around in config files and cli to get things back online.
2
96
The only metric that matters today:
Are you diluting or improving.
Forward. Always forward.
Big unlock for open-source AI inference: Hugging Face Transformers models can now run in vLLM at native speed, often matching or beating hand-written implementations.
Until now, every new architecture often needed to be built twice:
- Once in Transformers for training and research
- Again in vLLM for fast production inference
That duplication slowed down new models, added maintenance, and created room for implementations to diverge. Now, model authors can implement an architecture once in Transformers and immediately benefit from vLLM’s optimized inference stack.
In our benchmarks, the Transformers backend matched or beat native vLLM throughput across models from 4B to 235B parameters, including tensor parallel and MoE setups. One readable model implementation can now power training, fine-tuning, evaluation, RL rollouts, and production inference.
The conventional wisdom is that abstractions make systems slower. The best abstractions make the whole ecosystem faster.
Write the model once. Deploy it everywhere.
huggingface.co/blog/native-s…
2
99
Why am I not surprised @nvidia is making moves in quantum before most think it’s more than a toy?
Cheers to turning toys to must have deploys.
Quantum error correction just took a major leap forward.
Color codes have always been limited by decoding performance. Today, NVIDIA Ising Decoding has changed that — delivering over 300x improvement in logical error rate and over 7x decoding speedup.
Not only does this make color codes viable, but it gives QPU builders and researchers the tools to train their own models, bringing useful quantum computing one step closer.
Learn more 👉 nvda.ws/3SYbMza
ALT NVIDIA Ising Decoding Cuts Color Code Logical Error Rates by Over 300X
1
200
