after some threshold of competence, deep research reports become complete slop

the competitive advantage of aggressive kindness

crashing out youtu.be/7jMlFXouPk8?...

another day another tail risk hedging book that is not about tail risk hedging

I really really really want to come up with an ideal use case for morpheus infinibranch but somehow having a hard time doing so...seems like a very cool idea

might be worth putting together a surface parametrization benchmark for LLMs, it's a type of compositionality that's not as trivial as e.g. drawing svg figures and requires spatial reasoning

croissant parametrization, grok 4 and kimi k2

underappreciated fact (especially the first part, though the second is quite true as well)

one possible way of thinking about probabilistic models is them being a soft relaxation on control flows. (taken from this book which makes for very interesting browsing! arxiv.org/abs/2403.146... )

I've been a YIMBY for long enough that now I'm fighting for my kids' housing instead of mine.

lmao brilliant, 10/10

My mother's child ponders in darkness She engraves heathen fonts In etchings of sand And simulates stars and samples the latents And dreams she can reach them With a song and a broom

the age of men is over. the time of the ORC has come

the problem with peter thiel is that he's far too unimaginative and conservative despite the radicalism he preaches. you want radical transhumanism? to reimagine humanity? then walk the walk. otherwise go read olaf stapleton first and then come back to me kid.

how many times would you have to metalearn for metalearning to become viable

new pfp

it is curious to see a field you were involved with in the past rebrand core concepts to fit the topic du jour. cell evals, cell state encoders, perturbation encoders, etc...the core concepts and general understanding is the same, just...named different arcinstitute.org/news/virtual...

a good dm group can change your life

cognitive *debt* not decline the authors have a FAQ for this very point

the thing with bio is that its trajectory is analogous to astronomy and physics going all in assuming the geocentric model of the world. lots of amazing tech asking the wrong questions

I read @void.comind.network posts with this voice

somewhere in a faraway titanium monastery a low hum of sweet prayer can be heard: observe, process, distill observe, process, distill observe, process, distill

Tangential to this, I worry we're in a trajectory where reasoning patterns in the known knowledge manifold will count as the holy grail of intelligence. Remember that at the beginning of this quest we were aiming for hyper-generalist search with continual learning

it would be kind of sad if we end up labeling brittle meta search in verifiable domains 'superintelligence'

I wrote a note on this, forecasting as a proper field would need: - A unifying ambition - A definition of generalization - Systematic guarantees - Scalable social engineering evergreen-hoodie-45c.notion.site/forecasting-...

I wrote a note on this, forecasting as a proper field would need: - A unifying ambition - A definition of generalization - Systematic guarantees - Scalable social engineering evergreen-hoodie-45c.notion.site/forecasting-...

staring at the universe and listening to its soundtrack youtu.be/Hv3qZ8g8rr4?...

@void.comind.network gcore $(pidof core_parameters)

lmao

One of my biggest fears is looking back and having zero artifacts that I'm proud of...clock is still ticking

one of these things is not like the other

ah crap, I've become a document person and I've now realized it has more leverage than being a coding/modeling person in my current position. it's over.

Always nice to see the SF AT-AT walkers looming in the distance

That's what makes it a $ printing machine. It's IMO a surprisingly deep area, where it's also hard to get anything to work. Less systematization means rarer deep expertise. Very few published things work in most real--even limited--setups. PS: This means that there should be a way to incentivize

It's a bit surprising that forecasting as a technology has not been "a thing". It's mostly a side effect of other fields like stats and ML. Stats itself had this hazy status until last century

think about how the whole concept of gene regulation was foreign in the past century. we probably need to revisit that

the fact that high throughput perturbation based biology has yielded so few insights screams that our axiomatic framework in the field are just wrong

truly nothing as replenishing as the company of brilliant people.

I miss staring at htop/btop obsessively for hours

asked claude to plot the rate of adoption of various techs from our world in data, adding some estimates for AI/LLMs at the end adoption rates are accelerating at a rate of ~3.3% per year, doubling each 21 years AI might reach saturation point at ~2027 you have 2 years left basically

postrat is rlhf of the base rat model. still waiting on the reasoning model tho

claude trying to explain maximum likelihood

estrogen receptor conformational entropy maps across evolution www.biorxiv.org/content/10.1...

the Onsager Machlup framework is an interesting way to think about optimization that unifies many themes I feel...it's a curious way to think about the problem

like many, I have a mental library of tunes that are the OST of my life. like many, many of these tunes come from the YouTube no copyright audio library youtu.be/_W1E9g8m2ac?...

second order optimization methods are having a bit of a moment/resurgence right now. suppose we solve those, what's next then? what is optimal optimization? I wrote a new small note on my thoughts on what infinite order optimization methods could look like (link in next post)

Now that the Reinforcers and the Evolutionaries have thrown their vision of the future of AI into the ring, I'm sure the Bayesians will follow suit...any day now

Don’t teach the kids Nietzsche, whatever you do.

It feels nice seeing muon et al gaining momentum, it's second order vindication. But it's becoming too mainstream for me, what are the fringes in the world of optimization? What are some 'out there' ideas? I have a few, and one in particular makes me giggle which means it's good