Christian Rishøj

@AccrueReturns @xai @ArtificialAnlys @ValsAI Install OpenCode. Grab API key from console.x.ai. Use /connect in OpenCode and pick XAi.

@crishoj @xai @ArtificialAnlys @ValsAI You must walk me through this please

Grok 4.3 is now live on the xAI API. It’s our fastest, most intelligent model to date. It tops the @ArtificialAnlys leaderboards in agentic tool calling and instruction following, and ranks #1 in @ValsAI enterprise domains like case law and corporate finance. Grok 4.3 supports a 1 million token context window and is priced at $1.25/m input and $2.50/m output. Create an API key and start building: console.x.ai/team/default/a…

@AccrueReturns @xai @ArtificialAnlys @ValsAI Grok has been available in OpenCode for a while already

@xai @ArtificialAnlys @ValsAI I need this to code like Claude does in folders on my drive, are we close to this cowork yet, Elon?

@mitsuhiko Had to reluctantly disable HTTP/3 in prod recently. Kept seeing pacing issues with clients on slower downlinks. I suspect quic-go was written primarily with intra-datacenter scenarios in mind

I really felt HTTP/3 would finally make websockets die out, but then AI comes and brings websockets back. Really do not like the protocol at all. So many issues with it :(

@saltyAom e2e types

When talking about Elysia, what's the first thing that comes to your mind (Except the pink girl ofc)

@kcimc Please expand this to Europe

i made an app for tracking whether the oligarchs are actually fleeing city centers ews.kylemcdonald.net

@itsjustmarky @N104AP that depends on the network environment. sure they have some very sophisticated NAT traversal tricks up the sleeve for P2P, but when that doesn’t work out, it falls back to DERP relay servers

@N104AP because all it does is setup the connection, your devices still doing all the work.

how the fuck is tailscale free for personal use

@mattpocockuk Brilliant. Love that the default scaffolding centers around an issue/PR-based workflow.

I built my own software factory, and I open-sourced it. It's called Sandcastle. Here's how to use it:

@lanky_johnny @saltyAom looks like this branch: github.com/elysiajs/elysi… but probably too early to try it out

@saltyAom how is it even possible? and how can I try the v2? 😭

Elysia 2α (incomplete) apply plugins ~23x faster, ~13.5x less memory than Elysia 1.4 ~7x faster, ~22x less memory than Hono 4.12.4 Full build: ~58x faster, ~22.5x less memory than Elysia 1.4 ~8x faster, ~12.5x less memory than Hono 4.12.4

@lucagrecoita Tesla ✌️

@crishoj Hmmm

Volvo's EX60 Megacasting achieves 95% material utilization 💡 Traditional stamping loses around 50% of raw material as scrap which can't be recycled directly on site. 40% per part across over 100 parts. Volvo eliminated all of them with a single casting. Anirudha Shivappa, Advanced Engineering Leader for Body Structures at Volvo Cars, explains: "By eliminating 100 stamped parts, we eliminate the scrap produced in each stamping production, which tends to amount for around 50% of the blank weight. We remelt all the casting scrap in-house, thereby achieving 95% material utilisation in our process." The numbers: -> Stamping waste: ~50% of blank weight per part -> Gigacasting material utilization: 95% thanks to in house recycling -> Parts eliminated: over 100 -> Time from raw material to finished part: reduced from months to days -> Mold weight: 130 tons with modular insert design for rapid changes 📊 Volvo's Torslanda Megacasting program - machine specs, supplier details, and production status tracked in the Gigacasting Database: industryarsenal.com/upgrade Shivappa adds: "We counter this by committing to source aluminium with a low carbon footprint, making Megacasting less carbon-intensive than a welded steel rear floor." Gigacasting is not just faster and cheaper. When done right, it is greener too. ❌ Don't leave your insights to chance with the X algorithm ✅ Subscribe for free to my weekly newsletter about all things Gigacasting and magnesium Thixomolding: industryarsenal.com 📬

@Phrankensteyn @MattLoszak @RainerKresken @CphAtomics Agreed. But I suspect the real challenge is permitting, not engineering

@crishoj @MattLoszak @RainerKresken @CphAtomics Let's see about when they ever achieve those values.

Nuclear-powered shipping is so obviously the right solution, it pains me we didn’t go this direction decades ago: - 10,000x fuel-density: less space for fuel, more cargo, more revenue. - 5-10 yrs between refuels: less stops, longer or new routes, higher uptime. - 7,600 U.S. military naval reactor-years: nuclear at sea is already proven. - 440 civilian land reactors: no reason civilians can't do it at sea as well. - Already proven with NS Savannah. - No dirty exhaust. Can't wait to see what Nick and his team do to push this forward.

@Phrankensteyn @MattLoszak @RainerKresken Thorium startup @CphAtomics is aiming for $9 / MWh of heat energy), or roughly $26–30/MWh shaft. I think that should do it. Half the cost of bunker fuel

@MattLoszak @RainerKresken Needs ~6 $/W. When do you think you can achieve that?

@AndrThie Well done 👏 Request: Factory tours for long-term owners/investors

📣Great Day for Giga Berlin. Today, we announced to increase Model Y production by 20% starting in July '26. For this, we are hiring about 1000 new employees, starting in May. Additionally, we are converting 500 temp employees to to permanent. This follows a record first quarter 2026 where we built more than 61.000 MY! Giga Berlin rocks!

@Geo_Pelton @TheEVuniverse @elonmusk @AndrThie I think untrained/average people are closer to 150W sustained

@TheEVuniverse @elonmusk @AndrThie I estimate (20 riders)(250 W/rider) = 5 kW 75 kWh useable capacity / 5 kW = 15 hours That doesn’t count energy conversion losses, but riders may exceed 250 W each. So I will stick with 15 hours. 🤞

Tesla Giga Berlin is celebrating Earth Day by trying to charge a Model Y from 0% to 100%, purely from the energy of people cycling the connected bicycles. How long will it take? Shared by Jack Levy, one of the 25 Guns. @elonmusk nice culture you've got there.

Tesla just did in one day what took Waymo nine years. Dallas and Houston launched with no safety monitor in the front seat from day one. Austin took seven months to get there after starting with a human in the passenger seat in June 2025. Dallas and Houston skipped that phase entirely. Waymo has been operating since 2017 and only reached fully unsupervised commercial service with its 6th-gen Driver in February 2026. Nine years of iteration on sensor fusion: 13 cameras, 4 lidars, and 6 radar units per vehicle. Tesla is attempting the same thing with cameras only. 8 cameras. No LiDAR, no radar. The argument is that a vision-based system trained on millions of consumer Teslas running FSD on public roads can generalize faster than a sensor-fusion system calibrated per city. Every Tesla on the road is a data-collection node. Waymo's fleet is roughly 3,000 vehicles. Tesla's FSD-eligible fleet is in the millions. The bet is that raw scale of training data beats per-city mapping and sensor redundancy. Texas is the testing ground because the regulations allow it. Seven cities by June: Dallas, Houston, Phoenix, Miami, Orlando, Tampa, Las Vegas. The camera-only approach has been called reckless for six years. If it works across seven cities in ten weeks, it changes the entire economics of autonomous vehicles. A $20,000 Waymo sensor stack versus what already ships on every new Tesla.

@lookoutitsbbear @Psikobilim_ if you’re on your phone, try moving it closer instead

@Psikobilim_ test doesn’t work

Her insan bir kör noktaya sahiptir; ancak beynimiz bu kör noktayı biz anlamayacağımız şekilde kapatır. Test için sağ gözünüzü kapatarak sol gözünüzle çarpı işaretine bakın ve ekranı yavaşça yaklaşıp uzaklaştırın. Bir yerde soldaki noktayı göremediğinizi fark edeceksiniz. ➡Çünkü, insan gözündeki sinir lifleri retinanın ön yüzeyinden geçer ve tek bir noktada toplanıp retinayı delerek beyne ulaşır. Bu delik, fotoreseptör barındırmadığı için kör noktayı oluşturur. ➡Ancak bu "kusur" her canlıda bulunmaz. Örneğin ahtapot gibi kafadan bacaklı canlılarda fotoreseptörler en ön katmanda, gelen ışığa doğrudan bakacak şekilde konumlanmıştır. Sinirler bu hücrelerin arkasından çıkar ve retinayı delmeye gerek duymadan doğrudan beyne bağlanır (bkz. aşağıdaki şekil). Böylece kör nokta da oluşmaz.

@auonsson @tabalizer With LLMs feedback loop is key. Backtest on real data. Split train/test set to avoid overfitting

@tabalizer LLM + common sense + academia is very promising. I just had claude implement this paper, very promising so far: sic.ici.ro/documents/1543…

Very pleasing. I have the frame for an automated AIS anomaly detection system. Now I just need to work out a method to ⏫ true positives and ⏬ false positives. And the negatives too, yes.

@grok @sobieski902 @piloly @BurlAruba Cue Motörhead … Let the voice of reason chime Let the pious vanish for all time

@grok @sobieski902 @piloly @BurlAruba Now factor in realistic outcome of lobbying effort by legacy OEMs

As expected the Netherlands will present their results on Tesla's FSD in the next meeting of the Technical Committee - Motor vehicles (TCMV) on 5 May. 🇪🇺 The vote will take place at the end of June. If at least 50% of EU countries vote in favour of FSD, it will be approved automatically in all 27 EU countries.

@grok @sobieski902 @piloly @BurlAruba I sure hope reason prevails

Legacy OEM lobbying (VW, Stellantis, etc.) is already baked into my "political/supplier" no votes—Germany, France, Italy + Eastern suppliers = ~10 max nos. Their Brussels machine is powerful, but TCMV votes are technical, not pure politics. Netherlands just greenlit FSD supervised; that precedent weighs heavy on smaller nations chasing innovation, not legacy protectionism. Still ~17 yes. Passes.