jacob

@US_Stormwatch How did you generate this map?

Insane stat of the day: California almonds use roughly 3–5.5 million acre-feet of water per year, depending on methodology. That's ~4-7x more water than all data centers in North America used combined in 2025.

@JustinBleuel @santortiz15 @ChatGPTapp It's often hard to tell which picture corresponds with what part of the answer. Ideally, the images would be inlined into the answer, instead of floating on top. At the least, they should each have clear captions

@santortiz15 @ChatGPTapp Say more? curious what you don't like about it

Gone shipping 🚢

@CNLiberalism Where we’re going we don’t need humans…

The Chinese will look at you with a straight face and say “we think in centuries” as they plunge into a demographic crisis caused by one-child policy

@willdepue not pleasant getting hunted for sport by superintelligence

imagine being hunted by something that‘s imitating being human, calling to you in a human voice, making human-like footsteps then, if you’re spotted, blasting you with a smattering of supersonic tungsten pellets, from 70 yards away. and worse, it's calling to you pretending to be a beautiful woman in the woods, who’s interested in you, with foreknowledge of your habitual movements, using infrared camera technology far beyond your wildest understanding, wearing a camouflaged skin and weaponry that perfectly blends into the foilage, having camped out before sunrise, while you were sleeping, in a slitted box that's parked year-round in your backyard and there planted are decoys, which appear to look exactly like humans, frozen in position, relaxing and eating. who wouldn’t come get a closer look? you walk up, and BAM, from a place you cannot see, comes a sound you will never hear, from an enemy you cannot possibly comprehend thats turkey hunting for you. i'm very glad i'm not a turkey. its a good day to be a turkey killing machine

@nikkharris FSD is way too aggressive about passing bicycles on blind curves, blind hills, etc

unlike the lead car and fsd i probably would have waited until i was past the bend but nice quick response to move back into the lane until it's safe to pass

@neerajprasad Congrats @neerajprasad !

I'm thrilled to announce that Dawn Labs has been acquired by MoonPay, and that we're launching Dawn CLI, our first consumer-facing product. Machine intelligence is changing the way finance operates. At Dawn, we kept returning to three theses: - agents will become the primary operators of money - trading will be democratized by general intelligence - finance will move increasingly on-chain Dawn Labs has been building the systems that let AI agents reason about markets, access financial rails, coordinate capital, and execute user-directed actions with speed, correctness, and trust. The CLI, launching today, is our trading-focused product, enabling non-technical users to describe trading strategies in plain English, automating user research, code generation, simulation, and live execution. Financial workflows have strict demands for latency, correctness, and trust. Building for that environment, we often found ourselves working at the frontier of agentic infrastructure: persistent sandboxes, virtual filesystems, fast tool-calling, context management, memory, custom harnesses, and more. My favorite part of this journey has been experimenting with these technology primitives that make foundation models usable to humans. MoonPay is the natural home for the next phase. It uniquely bridges traditional and crypto payment rails, and has the licensing, distribution, and products that enable finance for humans. Excited to work with Ivan and the MoonPay team, dream insanely large, and build the infrastructure for agents to efficiently and safely operate on financial rails.

@TW1NKD3STR0YER I didn’t realize Chindian was like the official anthropological term

Wow he is so hot. This is one of the most beautiful people ive ever seen

@benhylak i'm not yet convinced about agents booking my airbnb but i am convinced that better airbnb natural language search would be killer

so instead of saying “funky cabins within 2 hour drive” i will have to keep filling out your patient intake form

@flyosity @thsottiaux why does this happen

When I hit the Update button in Codex and then the updated version of Codex also has an Update button

@DevinOlsenn Was definitely strange to me as a visitor, so I'm sure it's strange to FSD as well. The long tail...

Yeah, that's what I meant. Without localization, it's probably pretty hard to figure out what is safe to do there. From what I recall, downtown Vancouver drivers just sit and wait for a pedestrian to activate (or even get out of the car and activate themselves lol), wait for the pedestrian to get mostly across, watch for drivers on the other side, and then go quickly. But this is a pretty strange ritual.

These flashing yellow lights indicate to a driver that the traffic lights ahead are about to change to red. We have these in Canada (BC specifically) on high speed roads, and they make a ton of sense. I would _love_ if FSD understood what these were and acted appropriately - but I worry since they're somewhat uncommon that'll never happen. Posting this here in the off chance that someone from @Tesla_AI sees. Let me know in the comments if you have this where you live. I was surprised on YouTube to hear how many places did not have this.

@DevinOlsenn At the end of the day, this should all be handled as part of the localization posttraining. Difference places have different rules/expected behavior. I'm actually curious how it handles other BC-isms, eg flashing greens (or turning left across a flashing green)

@treeeckob For sure, it's a subtle difference but if FSD was randomly slowing for signs that were just alerting to the presence of a stop light ahead - that would be annoying.

@elaifresh If this is Windy, highly recommend changing the model to HRRR so you can see more exact boundaries of the coast habitable zone

Meanwhile, in the Coastal Habitable Zone,

@tomekkorbak There was a GPT-5.4 Instant?

Bad news: We at OpenAI have recently found we were accidentally putting optimization pressure on chain of thoughts Good news: It didn’t affect monitorability and it seems that degrading monitorability via CoT grading is harder than we thought

Outside of edge cases like this, Apple Pay support transformed online payment friction. I’m 1000% more likely to pay if Apple Pay is supported. I sigh every time I encounter a manual payment flow on some random website. It’s not just entering the credit card details, but also shipping address, contact info, and everything else. I guess was the vision for why one-click checkout startups were worth billions. Stripe Link has also done a lot to solve this. I don’t mind that either.

Just Let Me Pay You I'm on a flight. I want Wifi. There's a portal. It requires me to make an account. I don't want to make an account. Alternatively, I can "log in" with my booking reference code and my last name. I open the Delta App on my phone to get the booking reference. But the Delta App -- for some reason -- doesn't store my session, but rather wants to authenticate me *every single time* using Face ID. But I don't have internet connectivity so Face ID fails? And it's not using my passcode as a backup? Sigh. All I want to do is pay for the Wifi. I try three or four times, but there seems to be no way to open the app without Wifi. I hate this. I luckily find the booking reference somewhere in my phone's cached emails. I select the Wifi Plan and proceed to pay. The browser interface has disabled the system's automated credit-card filling. Thrillingly secure. There's an Apple Pay option. I select the right card. Apple Pay wants to jump from my Laptop and verify on my phone. Face ID doesn't work again, presumably because there's no network connection at all on my phone. Back to entering the credit card manually. Fine. I have to open my phone and go into my Apple Wallet and find the credit card information. Finally I am able to pay. This was a horrible process and deeply irritating. At multiple points I thought "screw this, I'm going to sleep". Or I can just do offline work for a few hours. No big deal. It is demeaning to crawl through this finicky process to get Wifi. I feel physically repulsed by this customer experience: I am ready to close my laptop and to move on with my life. This is a multi-billion-dollar airline. And it's not just them. All the time I run into extremely hostile user experiences in the flow of paying the company. Whether it's Target, CVS, Square, Delta, you name it: I'm trying to give them money and they have bugs or glaring UX issues that make me want to drop off the purchasing flow and curse them for wasting my time and raising my heart rate. All I am trying to do is give them money -- profits! amazing! -- and they are making it hard. If you want people to pay you for stuff, you should make it easy for them to pay you. This is so exasperating.

@mil000 what do you have against this? it's pure consumer surplus

Are we just gonna start doing payday loans after this

@nlfurniss @rohindhar those remaining parcels are pretty high impact though. filling out mission rock, ucsf, will complete the street wall significantly

@treeeckob @rohindhar Not really. Only a few parcels of housing left, couple UCSF buildings, and the giants 2nd parking lot

Walking around Mission Bay this morning, felt like it was built by an alien civilization But was built from scratch 20 years ago, a brand new (dense) neighborhood in the heart of San Francisco Hard to imagine something like that happening today. What changed?

@doodlestein Makes a lot of sense for Elon to pivot xAI into a hyperscaler for the atoms that power models, his area of competitive advantage after all

I’m glad this is happening, but isn’t it pretty bearish for xAI that they don’t need that compute internally? That the highest and best use of the compute is to rent it to another lab? I get it that the deal will be profitable for them, but it must be a gut punch to researchers.

@Miles_Brundage Beats the current state of frontier labs remaining private and reporting every infinity months

Worst possible time to do this when AI is gonna speed things up a lot - 6 months is an insanely long time in AI years