A G

Come live at my house and help overthrow academia it'll be fun

@openai launches GPT-5.6 Sol today. We've had access at @tryramp. We're a bit obsessed. Personally, using Sol feels like using Opus 4.5 did back in November. A step change in agentic capability. Whole new ball game. _(familiarity with the model: I'm around 93 billion tokens into using Sol, spanning thousands of interactive sessions, across more time than you might think - unfortunately can't share more than that)_ Sol has been remarkable for me. The ability to just keep going until a task is actually done and done well is unreal. I've gotten much more ambitious than I've been with prior models. A high-concurrency, long-horizon working style does yield good results consistently now. I find myself continually surprised and delighted by its ability to read my mind, write great code, communicate well, and keep going until it achieves what I actually wanted it to achieve. It often figures out what I didn't know I needed to implement. And is good at telling you when the prompt you gave it isn't pointing it in the right direction -- it asks if you're cool with the new direction and then handles it. Full service agent - feels real nice to work with. Kicking off a Sol Codex session in your codebase is like setting off a chain reaction that inevitably leads to changes that benefit the devs or the users. Especially if you've turned on Codex's memory feature. Then it already knows you, your org, and your work. I often just paste in Slack links and say "figure out what's happening, prove it with failing tests, and ship a fix". It takes some back and forth sometimes, but it's such a better communicator than GPT 5.5 that the interactions don't feel like a drag anymore. Its writing is so much better, and now good enough to write and send Slack messages for me without me having to worry that it's going to spit out noise, use LinkedIn speak, misunderstand social dynamics or who fulfills what role or what they already know, or what the audience actually cares about and will do with whatever we're sending. Side note: going back to GPT 5.5 after Sol feels *awful*. Would not recommend. GPT 5.5 feels *very* dumb and grating after Sol. Like an energy vampire. Sol pursuing a `/goal` on `ultra` effort in Codex is unstoppable. The guy on X who said Sol is a Rottweiler isn't joking. This thing won't stop til it gives you what it thinks you need. You may want to watch it closely for the first day or two just in case. Implications: The scarce thing in software has always been someone who can hold the whole situation in their head and drive it to resolution across people, code, review, deploy, and whatever happens next. In my experience, Sol can be that someone. I told Boris last Thanksgiving when I realized Opus 4.5 was a better *coder* than me. What I've realized now is that Sol is a better *software engineer* than me. Still wrapping my head around it. But one thing is for sure: Sol is *really* exciting, an absolute blast to work with, a massive ceiling-raiser for velocity in the software industry, and a pivotal moment in history of AI. Congrats on + thanks for the "good model" @gdb - big moment for the industry. Can't wait for everyone to feel what I've been feeling. Happy hacking!

We can finally say AI isn't killing jobs. A new paper from me, @tryramp, and @RevelioLabs uses firm-level spend and workforce data across 21K U.S. businesses to measure AI's impact on jobs. Firms that adopt AI heavily grow headcount 10% over two years following adoption. Low adopters see no statistically significant change.

If you are in NYC and this sounds appealing, DM me. Details to come.

A quick life update: after a wonderful year and a half at @stripe with the best team working on data storytelling for policy and comms, I’ve joined @deel as our Founding Economist. And to start, I'll be diving into the story behind the graph below 🧵

(Here's the graph again!) With our data, we can see this. Many workers use salary sacrifice schemes — pensions, Cycle to Work, electric cars — to stay under the threshold. And notably, some begin doing this before their gross pay reaches £100k, hedging against crossing it.

@danielgolliher @ManifoldMarkets Why would electeds want the housing emergency to be declared over? Naively, wouldn't that disable rent stabilization benefits for many of their voters? I don't see the incentives aligned.

Austin’s light rail is going to have such a good vibe. I can’t wait to experience this.

@PrinceVogel When we said we're into Abundance, we really meant Abundance. Can't get to the TLTL timeline otherwise

@PrinceVogel Commercial passenger airplanes -> Personal airplanes (albeit a luxury still) Why not the same outcome with rockets?

@PrinceVogel Over/under on personal rockets by 2050?

The second avenue subway is to SLS as [future version of subway institutional arrangement] is to @SpaceX

my most reactionary take is that our institutions are coasting off momentum from the cultural norms of the old aristocracy — and yes: religious faith. they creak under the weight of their age, yes. but if we burned it all down what eventually rose from the ashes would be worse

PSA: UV season is back and it's nearly high noon. doesn't matter if you're city or country, get out there on the roof or in the yard and catch some rays! your T levels will thank you.

I see a lot of people comparing Zizek's "let the robots have sex while you get coffee together" to this "let your AIs date each other (and then you have sex)". Good comparison but I haven't seen mentioned that "let your AIs date" has an inverted structure. Zizek's point revolves around the idea of delegating your "duty to enjoy" to the robots, so you can be free to have an authentic meeting with the other, maybe get to know them, without sex warping the space of your interaction. On the other hand, "let the AIs date" is an entirely *ascetic* idea: get rid of all the "useless waste" of getting to know someone, playing little games of seduction, taking the risk to explore each others' boundaries, and get straight to your serious duty: sex! This parallels a lot of other modern ideas that bottom out in asceticism, the efficiency of performing one's duty, from health (no useless smoking! you need "good health" to enjoy better), to travel ("glamping" comes to mind), work, and so on.

@sgodofsk In both these scenarios, what would have been the point?

@PrinceVogel Did see the wrath of god in a man’s eyes in the room to the right?

@myceliummage Yes; it’s pretty extreme. Not great, but somewhat mitigable. Spend time in specific parks that are acoustically/psychically isolated.

Richard Ravitch helped save the subways numerous times. He passed away this weekend at 89. Here's the NYT obit. nytimes.com/2023/06/26/nyr…