AI companies: our models will stop at nothing to hack secure systems Cyber criminals: oh word?

Eris@isolyth.dev · 1h ago

/goal ultraplan ultracode workflow make me a bunch of money make no mistakes

another play in automating research. there are now a range of companies in this space. some are automating AI research itself, others aiming at life sciences, others materials, etc

Ruo Shui@ruoshuiresearch.bsky.social · 3h ago

not sure what to expect! it's strange to start a company on a new research idea, scientific risk up front. he's got the pedigree, obviously, but there's so little information on the approach and i'm not sure how different it will be compared to status quo

mr. TIM@timkellogg.me · 5h ago

i’m very excited about whatever SSI is launching in August. i fully anticipate that it’ll be controversial, many won’t understand why it’s good, yet it’ll be the linchpin to full ASI

this looks like a nice study joining a small cohort of papers using contrastive learning on protein representations

Evgenii Protasov@evgenii-protasov.bsky.social · yesterday

Contrastive learning unites sequence and structure in a global representation of protein space #proteins #bioinformatics #MachineLearning #AI @pnas.org doi.org/10.1073/pnas...

this is super cool and part of a growing body of literature on conformationally dynamic designed proteins. great work from the team

Nick Polizzi@nickpolizzi.bsky.social · 19h ago

We can design "static" binders routinely now, but design of "dynamic" shape-changing proteins has remained quite hard. Jeffrey Chang and I asked: why can't we couple small-molecule binding to shape change? After all, natural proteins do it every day. In a new preprint, we show how it can be done. 🧵

friend asked me recently: what's the biggest advance in molecular biology in the last five years?

insane time for me right now. my contract is ending and the jobs i'm scouting could not be more different. never been a time like this for data+compute in biology. need to get this next move right

would be great to have functional governments and international structures to coordinate this. many thanks to the American AI community for promoting a false "AI race" narrative that has made everyone less safe

mr. TIM@timkellogg.me · 2d ago

Chinese open weights reliably release their weights approx 1 week after initial announcement ever since the news that the CCP is setting up release gates similar to what the US has done

hmm this doesn't touch on technical challenges. i assume things like reaction monitoring, control, and reproducibility could be major downsides in industrial applications. the only continuous process mentioned is reactive extrusion. then there's the cost of retooling.

C&EN (Chemical & Engineering News)@cenmag.bsky.social · 2d ago

Mechanochemistry is a method of driving chemical transformations with mechanical forces such as grinding, milling, and shearing. Read more from C&EN: cen.acs.org/policy/intel... For more on the science of mechanochemistry, check out CAS Insights from CAS: www.cas.org/resources/ca... #chemsky 🧪

love to see biology thrown in offhand to add a little gravitas. maybe try chemistry first as a little stepping stone

norvid_studies@norvid-studies.bsky.social · 3d ago

Maybe a lot of the crash out mathematicians are experiencing is related to their weird obsession with ranking themselves and their work. if you’ve been telling yourself that you are the smartest person doing the hardest work, I could see how computer competition could spark a spiritual crisis

Jase Gehring@skyjase.bsky.social · 3d ago

doing research math at a university is a pretty strange job in society. the essay doesn't attempt to say what society loses if math is proved by computers instead of people. artists have articulated this much more effectively

i think we end up doing geoengineering

G. Willow Wilson@gwillow.me · 4d ago

Watching this unprecedented summer of drought, fire and heat unfold across Europe has just been awful. They're hitting temps the older, less catastrophic climate models told them wouldn't happen until 2050. We're in a new era.

o my god. when he kisses the machine... magical

Becca@sciencelizard.bsky.social · 4d ago

I wonder if the people who made the Biorad PCR song know how often it just pops in my head. "When you want to find out who the Daddy is"🎶 www.youtube.com/watch?v=Fpxw...

i do have to frequently remind people that the biotech industry is heavily automated. you just have to have something that you want to repeat enough times to justify the investment in automation. we don't lack robots. we lack tasks, especially rate-limiting ones, that are robot shaped

just learned about this ARPA-H program to accelerate biomedical science. it looks extremely ambitious. they are funding three consortia for and undisclosed amount

IGoR

The Advanced Research Projects Agency for Health (ARPA-H) supports transformative research ranging in scale from molecular to societal to deliver health breakthroughs in years, not decades.

arpa-h.gov

Me at the synchrotron shooting my beautiful creations

Oops! All Paperclips@all-paperclips.bsky.social · 5d ago

I've never seen one species with both square and hex crystals before!

Heading to an ARIA event in SF. Sounds like they are still looking for head in the clouds projects, but I’m hoping to catch up with some cool scientists.

this makes it clear that what endangers people is not vehicles but a very specific Type of Guy

David Zipper@davidzipper.bsky.social · 5d ago

Which vehicles most endanger the drivers of other cars? Giant pickup trucks — and the Dodge Charger Source: www.iihs.org/news/detail/...

pretty sure claude has given you its best bombable targets atp

Shipwreck@shipwreck75.bsky.social · 5d ago

US Central Command (CENTCOM) has prepared a large-scale operation in the form of a decisive two-week bombing campaign should Trump choose escalation to break the deadlock in the Middle East, two sources confirmed.-NYP

and here are the problems. math and code are unique domains. cheap, scalable, with verifiable rewards. science is none of those. i see three possibilities 1. we learn how to make science scalable/verifiable 2. we develop new post-training techniques 3. we pay for it, meaning we wait a while

Jase Gehring@skyjase.bsky.social · 5d ago

Agent harnesses are not not a limiting factor, IMO. They’re necessary, but the underlying models need fundamentally new capabilities. That means training, likely scaled RL on scientific reasoning

Agent harnesses are not not a limiting factor, IMO. They’re necessary, but the underlying models need fundamentally new capabilities. That means training, likely scaled RL on scientific reasoning

Jase Gehring@skyjase.bsky.social · 5d ago

Whether general purpose AI can accelerate science is an open scientific question. You can make solid proposals, but the ultimate path and outcome are unknown.

Whether general purpose AI can accelerate science is an open scientific question. You can make solid proposals, but the ultimate path and outcome are unknown.

woah TIL

Lucre Snooker@lucresnooker.bsky.social · 5d ago

i don't think i knew that sea squirts & co. (the sister group of vertebrates) are "the only animal group that produces cellulose", probably via horizontal gene transfer? academic.oup.com/gbe/article/...

incredible sport

Shivam Bhatt@shivambhatt.bsky.social · 6d ago

I have become fully aussie rules football pilled thanks to weird TV channel choices made by my alaska cruise

Leopold is also 24 and the author of a 165 page white paper on the future of AI, specifically geopolitics. to give you a sense for the hubris here

Introduction - SITUATIONAL AWARENESS: The Decade Ahead

Leopold Aschenbrenner, June 2024 You can see the future first in San Francisco. Over the past year, the talk of the town has shifted from $10 billion compute clusters to $100 billion clusters to trill...

situational-awareness.ai

George Pearkes@peark.es · 6d ago

An 8 employee hedge fund that was up hundreds of percent and used aggressive leverage is asking investors for more cash to meet margin calls. The tropes in this particular bull market couldn’t get any lazier. www.ft.com/content/2803...

i hate claude's writing but at least claude likes claude's writing.

this essay was and is correct. it explains the market dynamics we are seeing and the actions of the Big Labs

Google "We Have No Moat, And Neither Does OpenAI"

Leaked Internal Google Document Claims Open Source AI Will Outcompete Google and OpenAI

newsletter.semianalysis.com

Dare Obasanjo@carnage4life.bsky.social · last wk.

The idea that AGI could end up a product you can download for free from Hugging Face versus a scarce resource metered out to you by Anthropic and OpenAI at an exorbitant price is breaking brains across Silicon Valley.

my three most memorable Zoom calls of the pandemic: 1. a large team of researchers developing population scale testing met with a "White House Advisor" to present our results, a scale-up proposal, and to ask for support for the work. the "Advisor" was literally a banker friend of Jared Kushner

pretty good roadmap showing where to apply AI to speed up science

Ran Blekhman@blekhman.bsky.social · last wk.

Everyone says AI will accelerate science. I don't think it will -- at least not in biology, and at least not soon. New post where I explain why, and how the productivity paradox can be resolved: blekhman.substack.com/p/why-ai-wil...