Xander Balwit

Editorial @AnthropicAI. Formerly editor-in-chief at @AsimovPress. Well-fed vegan.

[Xander Balwit @AlexandraBalwit]

Thrilling to see a little coverage of my future food dinner in the @TheEconomist!

People can invoke a future of GLP-1s and meal replacement shakes all they want, but I am pulling for the Food Abundance Agenda.

[Xander Balwit @AlexandraBalwit]

Thrilled to see Asimov Press return under the stewardship of @NikoMcCarty and Brady Huggett. Pitch them!

[Asimov Press @AsimovPress]

It's official: We're back!

Asimov Press is returning with a sharper focus on intensive reporting in biotechnology. We're also working with @RenPhilanthropy to build our ideas in the real world.

See more: https://t.co/SEpahlZVAN

[Xander Balwit @AlexandraBalwit]

For gorgeous table book and scientific history fanciers everywhere. Feeling grateful I had a hand in the early stages of this, and floored by how far @EllaWD_PhD and @NikoMcCarty has taken this. It's truly beautiful

[Niko McCarty. @NikoMcCarty]

Our @AsimovPress book on the history of the molecular biology lab is nearly finished. It's currently 455 pages, with 100+ images. Some previews below.

We've spent the last 6+ months making this book as beautiful, and detailed, as possible. And we're excited for you to hold it.

[Xander Balwit @AlexandraBalwit]

@ulkar_aghayeva @AsimovPress https://t.co/yXEkxCwnsm

[Xander Balwit @AlexandraBalwit]

I've been thinking about the aesthetics of scientific judgement and experimental work since editing @ulkar_aghayeva's wonderful essay on what makes experiments beautiful @AsimovPress.

Now, I'm curious whether AI will ever be capable of producing "tasteful" work. Tell us below

[Samuel Stanton @samuel_stanton_]

What is scientific taste?

Tell us what you think in this 10-min Claude-led interview on how scientists choose problems, judge evidence, and decide what's worth doing.

https://t.co/imq1F3CPgm

[Xander Balwit @AlexandraBalwit]

@s8mb Thank you so much Sam! Carry the torch 🫡

[Xander Balwit @AlexandraBalwit]

As excited as we both are about what's to come, this is a hard transition for us. Working on @AsimovPress was such a joy. On any given day, I was deep in mid-19th century books like "An account of the fishes found in the river Ganges and its branches Atlas" looking for primary research or having phone calls with our brilliant contributors. I am so proud of what we built, and grateful for all I learned along the way.

[Asimov Press @AsimovPress]

We're pausing @AsimovPress for awhile.

Thanks to everyone who has taken this journey with us so far. We will plan to see you again in a few months :)

Read: https://t.co/Y98PtC1jmm

[Xander Balwit @AlexandraBalwit]

@lfschiavo @AndyMasley I want to see this essay

[Xander Balwit @AlexandraBalwit]

@NikoMcCarty It's in the works! If only someone at my work place knew how to code fields like that.

[Xander Balwit @AlexandraBalwit]

“AI for science” has become the subject du jour; the theme of a steadily growing number of hackathons, fellowships, and research salons.

As this fervor grows, I am excited to see Anthropic participate in, and help facilitate, such discussions.

When I reflect on why this thrills me, I am reminded of what historian Horace Judson writes inThe Eighth Day of Creation: "Scientists like to ride the cutting edge of a subject, close enough to hear the hiss."

At Anthropic, we are flush against the engine, helping to make the very tools that we hope could "10x the rate of major discoveries." At the same time, we have to ensure we are not getting ahead of ourselves. AI’s scientific capabilities are still incipient. While models are already proving capable at certain aspects of scientific research, they can also hallucinate results, be overly sycophantic, and get stuck on problems a domain practitioner would find trivial.

It is our hope that this blog explores the tensions in this current moment of AI for science, acknowledging where AI-driven science has much further to go while also celebrating the ingenuity that is already unfolding.

[Anthropic @AnthropicAI]

Introducing the Anthropic Science Blog.

Increasing the pace of scientific progress is a core part of Anthropic’s mission. The Science Blog will feature new research and stories of how scientists are using AI to accelerate their work.

Read the intro: https://t.co/1P9BDyX3xG

[Xander Balwit @AlexandraBalwit]

Prof. Matthew Schwartz with some sage advice (and an extremely solid through-line for both these pieces):

"Regarding the question of where this leaves human grad students, my advice to students at all levels (and in any field) is to take LLMs seriously. Do not fall into the hallucination trap: I asked the LLM X and it made something up, so I’m just going to wait for it to improve.' Instead, get to know these models. Learn what they are good at and what they fail at."

I have also noticed a tendency toward "I will just wait until they are super good at the thing I want it to do for me." However, even given that the models likely will keep improving, familiarizing yourself with them now will put you so far ahead. I think about my own AI fluency all the time, and wish I was using them even earlier. I would rather have to deprecate my process while learning some managerial skills than watch something cook that I can't even parse.

[Anthropic @AnthropicAI]

We’re launching with two new posts.

Can AI do theoretical physics?

Harvard physicist Matthew Schwartz led Claude Opus 4.5 through a graduate-level calculation. AI can’t yet do original work autonomously, but it can vastly accelerate it.

Read more: https://t.co/UUfFuLqhb7

[Xander Balwit @AlexandraBalwit]

Congrats @adjajadikerta, this came out so well!

[Niko McCarty. @NikoMcCarty]

I think this is one of the most important articles we've published at @AsimovPress. If you read carefully, there are at least 3-4 ideas in here that *should* be large, well-funded research programs.

The article begins by arguing that existing AI models are good at predicting things *within* an existing framework, but are not good at building new frameworks (and, thus, cannot do paradigm-shifting science). As AI models become more widespread in science, they therefore risk "hypernormal science," meaning we will have less actual breakthroughs and more incremental discoveries.

The author (Alvin Djajadikerta) supports this argument with several examples, one of which comes from germ theory:

"In the mid-nineteenth century, doctors thought that illness was caused by noxious air, and kept meticulous records accordingly. The physician William Farr mapped cholera deaths across London and found they correlated strongly with low elevation, which he thought was because noxious vapors accumulated in low-lying areas. He was actually picking up a real signal: low-lying districts were closer to the contaminated Thames River. But because his data was organized around air quality, he could not find the true cause..."

"An AI trained on Farr’s records could have found even subtler correlations, and would have been genuinely useful for predicting which neighborhoods would be hit hardest in the next outbreak. But it would not be able to derive the concept of a waterborne microorganism, as this was not a variable anyone had yet recorded."

After giving other examples of this, Alvin begins mapping out ideas to solve this problem and create AIs that are "visionary" rather than "merely predictive." My favorite idea, of his, is to use AI agents as a model organism for metascience.

The gist is that many paradigm shifts seem to happen under particular conditions. "Bell Labs, Xerox PARC, and the early Laboratory of Molecular Biology at Cambridge all produced extraordinary concentrations of paradigm-shifting work," Alvin writes, "mostly because they were small groups with enough institutional protection to pursue ideas that looked unproductive by conventional measures."

Alvin continues:

"We have never been able to run controlled experiments on scientific institutions; it is impossible to create labs that differ in only one respect and compare the results. But we could run AI agents in parallel populations under different research conditions, and analyze the results...In this sense, AI scientists may give metascience its first model organism."

"For instance, one could test how group structure shapes discovery: do small, isolated teams produce more conceptual reorganization than large, well-connected ones? Do flat hierarchies outperform rigid ones? One could run AI agent populations that vary these factors independently and measure the results — something that is impractical to do with real institutions..."

This essay is excellent throughout and I hope you'll read it.

[Xander Balwit @AlexandraBalwit]

@mold_time Woah. I didn't know that about Robert Caro. That's extremely cool

[Xander Balwit @AlexandraBalwit]

Also, if there is an AI for science angle you'd like us to cover, drop us a line. We are setting up an email at [email protected].

[Xander Balwit @AlexandraBalwit]

The pieces in this blog will be eclectic. While we intend to publish some features and deep dives, we will also share plenty of workflows and firsthand accounts of using AI for scientific research. Some pieces will be more practical, while others will explore history or theory. Some will be geared toward folks more familiar with the technical aspects of scientific computing, and others will be enjoyable to, say, your humanist pippy pops who struggles with his cell phone but is interested in how AI might affect scientific integrity.

[Xander Balwit @AlexandraBalwit]

@surprisetalk Dear god. I love this.

[Xander Balwit @AlexandraBalwit]

@rSanti97 Yay! Excited to work with you Santi!

[Xander Balwit @AlexandraBalwit]

@salonium @StefanFSchubert Congratulations Saloni!

[Xander Balwit @AlexandraBalwit]

@s8mb I think they may have overcorrected on that one. Imagine one elongated banana seed. Based