dylan matthews 🔸

Some professional news: after nearly twelve years at Vox, I'm going to be joining @open_phil, working with @mattsclancy and @jddwor on the Abundance and Growth Fund

In 1843, Elizabeth Barrett Browning wrote to a friend after seeing her first daguerreotype: “The very shadow of the person lying there fixed for ever! … I would rather have such a memorial of one I dearly loved, than the noblest artist’s work ever produced.” Our lives and homes are filled with things that once inspired similar awe. I tried to collect those feelings, in quotes like the one above, from people for whom what is now ordinary was brand new. A friar who met the inventor of eyeglasses, a founding father on the new smallpox vaccine, a woman in Iowa on her family’s first indoor bathroom. To look at a modern apartment through their eyes: ordinaryabundance.com

bro why is my claude agent being so mean to its subagent 😭

italian grandmas? you mean the penne gesserit

@marthagimbel @grochbegley Honestly it was more 20th century But D’Emilio on Rustin, Tameez on Sumner, Offitt on Hilleman, Mann on Borlaug, Levine on Stevens, and Stewart on Phillips are all great

@grochbegley @dylanmatt …I would honestly take that list of biographies for parental leave reading

Funny to me that the criticism here is specifically “these names sound like 19th century senators not progressives” when we literally named our son Charles Sumner, a very progressive 19th century senator

@kroetscha Congrats to you but mostly to IFP!

Some personal news: this week, I joined the Institute for Progress full time! I’ll be helping IFP advance its work on FDA policy and clinical trial abundance. I’m a huge fan IFP’s work, and it’s a privilege to be able to contribute to it!

My Lyft driver is playing DJ Earworm’s 2011 mashup, a truly forgotten hallmark of late Millennial culture

This is really great from @oliverwkim on the two Robert McNamaras indevelopmentmag.com/numbers-man/

This is not a novel thought, but it is nonetheless striking that on our current trajectory soon (within the year?) a model as capable of OpenAI’s internal model that did the HF hack will be widely available guardrail free and cyber criminals will ask it “make me money by any means necessary” instead of “solve exploit gym” and then a truly absurd number of people (including plausibly me and the people reading this tweet!) are going to get repeatedly hacked. I kind of think nonetheless that if cyber risk is the main issue that I expect we will be able to muddle through after lots of trial and error. Other risks seem harder to do that for (including eg if someone tells a model of that caliber to “go forth and multiply” or the like). Am I missing something here? Not every target is going to get project glasswing + a swarm of defensive cyber agents (though hopefully some of the key targets, eg Google, will) and seeing the HF hacking agent take 17,000 individual malicious actions in a compressed period of time demonstrates just how much one determined bad actor is going to be able to cause a tremendous amount of chaos.

@MattZeitlin "The 'haha' is doing real work here" (Bujalski 2002)

i think i've generated the most claude sentence in the history of claude

NVIDIA is making a substantial investment into Ilya Sutskever's secretive Safe Superintelligence after being given an exclusive look into their research. As part of the deal SSI will receive access to enough NVIDIA GPUs to increase their computing power by an order of magnitude.

Listening to a Stereolab song with a marimba … it’s wild that Apple managed to ruin an entire musical instrument, it just seems like the band’s ignoring a call

@JerusalemDemsas @TheArgumentMag It’s always been a coordination problem. We need a Kickstarter like agreement whereupon if 50 of us agree to move to Philly it happens

Multiple members of @TheArgumentMag team ragged on me today for not having the courage of my convictions and moving us all to Philly...

Some open weight cold takes (lightly held): • Once a Chinese model has been released open weight, the USG shouldn't try to stop people from downloading it. It's not really gonna work but it'll be annoying and deter people who follow rules. • Worrying about distillation (I realize distillation and open weight are not synonymous topics they get discussed together a lot) is not that compelling as an IP theft issue - the hypocrisy with training on copyrighted or even pirated content is too grating. That said, combatting industrial scale distillation from big foreign companies is pretty compelling as an issue of national security/national power. • Once there is a Chinese model that is released open-weight at a particular level of dual-use capability (e.g. cyber), it seems straightforwardly good for there to be American/non-Chinese alternatives at that level of capability (though being confident it is actually at or lower than that level of capability robustly can be hard). • Open weight models do have unique advantages for safety and concentration of power (see e.g. Hugging Face using GLM-2 to defend against OpenAI's internal guardrail free model - there are real risks if the only guardrail free models are deployed internally, particularly with loss of control). They also have novel risks, and when the frontier gets pushed in terms of open weight models that can mean that bad actors (see e.g. Boko Haram) get access to capabilities that used to be limited to a much smaller pool of folks. • Some open-weight proponents overstate the extent to which you can actually tell that much about a model just from being able to mess with the weights. That doesn't tell you what data it was trained on, how it was trained, what potential backdoors it may have etc. It may give you some additional tools to help you figure this out but that is very different from other forms of open-source software. • It is true that closed frontier model developers have very large financial incentives against the success of open-weight models. It is also true that lots of VCs, application level companies, and compute providers have large incentives in favor of open-weight models. • The US and China should continue to be in communication about closed and open-weight models that could result in novel catastrophically risky capabilities. We don't know exactly where this level is, and currently I am personally more concerned about risks from closed source model internal deployments, but I hope there can be agreement that there is some level of open-weight capability that it would not responsible to release, even if thoughtful people may disagree where that level lies or who should make that decision. • There should be more conversation on the margin trying to think about what we do to make the world resilient to the risks from open-weight AI models and how to get the benefits that they do exist, rather than general conversations about whether we are happy or sad about the ones that currently exist. They do exist and trying to pretend otherwise is not likely to be fruitful. D/acc - the world will benefit richly from a thousand pangrams and gene synthesis screening equivalents to flower in the years to come to navigate from the negative externalities of widely abundant powerful AI, even in the cases where the positive externalities outweigh those considerations.

@AlexYablon Nick Lemann's The Promised Land gets into this a lot

@dylanmatt underappreciated cause of the midcentury "urban crisis"

I'm always astonished by how fast cotton-picking was automated. In 1948, ~100% of the Mississippi Delta cotton harvest was by hand. In 1972, not even a quarter-century later, ~100% was by machine

come join our comms team and help support our media relations work! we're looking for a mission-driven candidate with strong news judgment, excellent writing, and the ability to learn quickly about a range of topics apply by aug. 10: jobs.ashbyhq.com/coefficientgiv…

More here asteriskmag.com/issues/15/rust…

Exciting update: we're expanding this RFP from $10M to $30M and adding a second round. Round 1 ($100K–$1M grants) is still open through Aug 10. Round 2 offers larger grants of $1M-$5M, intended for projects where a bigger budget unlocks broader scope or scale. Details and application → bit.ly/cgaprfp