Neurons, proteins, genes, math, puns and rhymes. I lick the world to see what it's made of. Molecular neuroscience @MIT
Joined July 2021
- Tweets5.6K
- Following530
- Followers264
- Likes18.5K
Pinned Tweet
New paper! We usually think of neurons as the brain’s computational nodes, but zooming in, each synapse is its own analog computer, integrating activity history, neuromodulators and genetic events to decide whether to strengthen or weaken a connection. 1/9
cell.com/cell-reports/fullte…
This Post is from an account that no longer exists. Learn more
What if the risk from AI is not extinction, but "merely" catastrophic, like a pandemic or war?
In that case you really want to be lucky, and that means seeing the smoke and being light on your feet.
Here is my public prediction and prep for AI catastrophe:
There may be plenty of physical impacts too, as these agents activate both humans and control systems. But I think those will be much rarer, plus not much we can do to prepare for the draught of Osiris
AI alignment being the front-"page" story of YNET, one of Israel's main news websites is quite the vibe shift
Happy 250th to the country where pubs don't last more than a generation because their kids have better prospects than owning a pub
Beny Falkovich retweeted
Baby has gotten good at identifying “dada” but still struggling with surrealism, cubism, even baroque.
I will not free the bugs
I will not tap the pods
I will not read the hugs
I will not toe the gods
Nor taunt the rooms, alas
Or go and touch the signs
I will not eat the grass
Or live within the lines
The best AI Scientist should not have read a single paper.
Papers are a tiny kernel of experiment->outcome reality surrounded by a cloud of fluff by and for humans. I want a foundation model that predicts experiment->outcome and isn't distracted by the fluff.
Who's building this?
I don't even want humans to decide how the papers should be read. The way I envision it working is a wrapper LLM that acts as autoencoder that extracts from each paper two indecipherable vectors of salient "setting" and "measurement" features
This used to be how we do biology, but now there are sophisticated tools to remove windows and phenotype the car. However, this makes things more confusing because some people knockout half the car's windows and report no effect, while others see that the car immediately crashes/
most things that move don't have engines.
Pharma companies decide to invest in the Car-not-moving treatment area and pursue thousands of different drug candidates.
Eventually, a drug called Motoroil (VVLN-0392), which mimics an endogenous engine substance/