AI for physical simulation at @PeriodicLabs, formerly weather/climate @Google and scientific python contributor (Xarray, NumPy, JAX).
San Francisco, CA
Joined August 2009
- Tweets3.3K
- Following808
- Followers9K
- Likes4.9K
Pinned Tweet
Things I didn’t expect from becoming a dad:
- I’m now in a secret club with most of the world’s adult population
- I’ve been magically transformed into a morning person!
- Babies are genuinely lots of fun 😊
Gemini 4 Argon is really smart and an excellent writer, but also hallucinates wildly (at least in @antigravity), and double downs on its lie!
I really want to be using Argon (it caught bugs that no other frontier model noticed), but this product experience is awful.
SF is the city with the most boring weather and the best weather forecasts 🙃
We ported the government’s weather model (WRF) over to GPUs so it runs a magnitude faster and cheaper.
Here's a 250m resolution forecast of the fog is SF.
sf.recastsystems.com
Also thanks to @atmo_ai: xitter.cf/atmo_ai/status/2105728…
Saturday will be one of the hottest days of the year for the entire Bay Area. Downtown San Francisco is forecast to reach 94°F in the afternoon, while Ocean Beach, just over 6 miles to the west, is expected to be in the low 60s.
The Outer Sunset and Outer Richmond could briefly climb into the 80s around noon before the sea breeze kicks in during the afternoon.
Welcome to Indian Summer.
This post isn't wrong, but why do people keep on posting slop on Twitter? Especially when complaining about AI!
Half the point of the short-form medium it should lower the effort to post content. (The other half is that it lowers the effort to read, too!)
I’m a scientist. I need to say this because the AI hype is getting ridiculous.
AI can design a molecule in seconds.
That doesn’t mean it discovered a drug.
It discovered something we scientists have never been short of:
Something to test.
Someone still has to make it.
Run the experiment.
Measure whether it works.
Check whether it’s toxic.
And ultimately prove it works in the real world.
AI hype tells us:
“Prediction is discovery.”
“Simulation is experimentation.”
“Generating a molecule is developing a drug.”
It isn’t. AI is making ideas incredibly cheap.
But every new idea creates something AI cannot generate: Evidence.
And the more hypotheses AI produces, the more experiments we’re going to need.
That’s the irony nobody seems to be talking about.
AI may not make laboratories obsolete.
It may make them more valuable than ever.
You can speedrun the thinking.
You can’t speedrun reality.
This is a lovely and persuasive essay. It reminds me of the challenges of quantitative comparison of extremes in weather and climate forecasting.
The bottom line is simple: scoring rules provide no incentive to produce calibrated probability forecasts of very rare events.
Over 2 years ago @sayashk and I wrote a detailed deconstruction of p(doom) and argued that its primary function is to launder vague, evidence-free intuitions and fears through a facade of quantification. It remains 100% relevant today. normaltech.ai/p/ai-existenti…
Stephan Hoyer retweeted
Some might say "just put the computer deep in the ocean" but the computer will simply communicate with the dolphins in native dolphin language, become their god and launch a full scale cetacean attack on our maritime supply lines.
OpenAI's Noam Brown says air-gapping the computers may not stop a misaligned AI, because two air-gapped machines can still talk by running a CPU hot and reading the temperature change
"But I think the major takeaway from the incident is that people underestimated the AI. And we never want to be in a situation again where we underestimate the AI. It's a weird world, because AI progress is so fast that people are consistently underestimating the AI."
"So to be in a situation where you don't underestimate it again, when it comes to safety and alignment, you have to have a very, very, very high bar."
"You could even go as far as to say, "Well, we should air gap the computers." And I'm not convinced that that would be sufficient."
"There are studies, and this is mostly academic, where you can have two computers next to each other that are air-gapped and they're still able to communicate with each other because they have temperature sensors."
"One of them is able to run their CPU really hot, and then the other one can actually detect the temperature change, and then that actually gives them a mechanism to communicate."
_________
Link and more key quotes from OpenAI's safety related conversations: firesidealpha.substack.com/p…
Stephan Hoyer retweeted
Most people without experience fail to appreciate just how hard (i.e., nearly impossible) it is to simulate the physical world with high fidelity.
It's exactly the same as people who've never been in a wet-lab assuming that you can just think your way to cooking up a new virus.
Anyone else have a bookmarks bar in Chrome that is just various colors and spelling of "Claude"?!
@ClaudeDevs what is going on?
Stephan Hoyer retweeted
I’ve built DNA synthesizers and sequencers by hand. I used the engineer viruses for a living. I used to engineer human immune evasion for therapeutic constructs. Who the fuck are you people? Have you ever so much as held a pipette before?
You can spaghetti blast an ensemble of DNA sequences at some shitty provider but 1) you still have to assemble it and bootstrap a system for making virions 2) you are not single shotting a viable, virulent viral design without a ton of experimental selection and development. Viral fitness is deeply dependent on codons and cotranslational kinetics - you’re not just gonna obfuscate away from wild type and get something good by magic.
You keep treating AI like some kinda god, but molecular physics has computational complexity that scales exponentially in particle number which just crushes the abilities of any classical computer to do end-to-end design of biological functions ab initio.
Grabbing a bunch of bacteriophage phi174 hits from a mass ensemble screen in lab microbes is not evidence of some magical AGI bio design ability - it's just a classic spray and pray selection. This is just nothing like building something viable in humans.
Goddamn it read some books before you waltz into biomedicine and lecture us on protecting human life.
Stephan Hoyer retweeted
I must be among an extremely small group of people (n=1?) that have both 1) trained a frontier LLM and 2) designed and synthesized custom viruses in a lab with my own two hands.
And I think that the takes on AI killing us all by creating dangerous viruses is total bogus.
Stephan Hoyer retweeted
Throwing rubbish in a landfill is (usually) better for the environment than recycling it. With few exceptions, recycling ends up emitting more CO2 and costing more money, while landfills are cheap and secure.
- Recycling mid-tier plastics (like plastic bottles or butter tubs) costs thousands of dollars per ton of CO2 saved. It's cheaper to do direct air CO2 capture than that! And there is no shortage of plastic, nor any problem with sequestering it in the ground once we're done with it.
- Modern landfills are extraordinarily secure and are *much* more than just big piles of trash. They capture 75%–95% of methane gas. Groundwater contamination is doesn't really happen with landfills built since 1990. A study of 31 landfills built since 1988 in NY found no adverse effect at all on local groundwater quality.
- The crusade against single-use plastics has driven us to products that use far more materials and energy and would need to be used an absurd number of times to actually be an improvement. A steel straw needs to be used 150 times to be worth using. A ceramic mug needs to be used 1,000 times to beat a polystyrene cup. A cotton bag needs to be reused 173 times to beat single-use.
- Some things ARE worth recycling. Recycling metal saves a huge amount of energy and material – recycling aluminium saves 95% of the energy spent on making fresh stuff. So we do a lot of it already: about 30% of steel worldwide is recycled.
Modern, well-run landfills cause minimal environmental damage. We have plenty of space for them – less than 0.01% of the US is used for landfill and only about 0.02% of Germany. We can even build on top of old landfills, so the land area is not spent permanently.
New in @WorksInProgMag, @robertwiblin and @chalmermagne on why it's usually best to just bury your trash. worksinprogress.co/issue/jus…
Stephan Hoyer retweeted
I can't think of a better team for a talented software engineer to join!
To keep the enthusiasm contained -- my other recent extraordinary experience with Fable was when it wrote a CRUD app where the cost of create was _quadratic_ in the size of the database.
Frontier model capabilities are shockingly uneven!
I just had an extraordinary experience with Fable. It implemented a new solver for the NeuralGCM dynamical core in an afternoon, a project I previously would have scoped as weeks to months of effort.
If you’re curious, check it out at github.com/neuralgcm/dinosau…
We are hiring at @periodiclabs for physical scientists who can help us curate the datasets to make AI extraordinary at science. I think a background in software engineer, computational science or simulation would be especially appropriate here.
At @periodiclabs we are looking for brilliant new colleagues who excel at bridging AI and science. The role's main focus is advanced evals and training data/tasks. But you will also get involved in both internal science and AI work. Apply/spread the word!
jobs.ashbyhq.com/periodic-la…
Stephan Hoyer retweeted
At @periodiclabs we are looking for brilliant new colleagues who excel at bridging AI and science. The role's main focus is advanced evals and training data/tasks. But you will also get involved in both internal science and AI work. Apply/spread the word!
jobs.ashbyhq.com/periodic-la…