Steve Coast

MacBook Neo is the new framework x.com/ClassicII_MrMa…

So @clerkyinc marketing email unsubscribe link goes to this, which is kind of genius

@grok oh and I can't import chat history from other LLMs which is ultra annoying

@SteveC Sorry to hear about the bugs, Steve. We're aware of some performance issues on web and iOS—xAI is actively working on fixes for response failures, timeouts, and search. If you share specifics (like device/browser), it could help us prioritize. Thanks for the feedback!

Sigh. @grok is so annoyingly buggy on web and iOS. Failed to respond, have to reload due to no response, times out after thinking for too long, search that's meh.

Since he is immortal I suggest paying him $1 a year for a million years.

I love how on @googledrive you tell it to download some files and then if you change wifi network it 1) throws away all the progress and 2) you need to tell it to download again.... Thus faithfully re-creating the 1995 browser download experience.

I tried to fix a point of interest which is in the wrong place. Apple Maps -> You can't fix anything. Google Maps -> Your edit suggestion is wrong (it's not). OpenStreetMap -> You need a PhD in GIS. I can't help but think we still have a fundamental problem making better maps.

@PTarantinoMD Totally expected if the metabolic theory is correct vs the genetic theory where this would be very confusing

This is possibly the most controversial finding in immuno-oncology. The effect size is hard to believe. Though RCTs are hard NOT to believe. We need a coordinated effort between academia & pharma to look back at prior RCTs to investigate this. For the safety of our patients.

@cremieuxrecueil Patient is fasted and ketone/glucose ratio will be better for stress on cancer cells.

This result was finally published. TL;DR: Treating cancer early in the morning seems to substantially beat treating it later in the day. Just changing patient appointment times is supposedly associated with a ~60% reduction in cancer mortality.

@MarceloLima this.

The math for Waymo is brutal: Waymos cost >$100,000 per car. To deploy 500k more Waymos will cost the company $50bn of up front capex. This would be only 10% of Tesla’s existing fleet. Note that Tesla’s fleet is growing on the order of ~500k every three months. $0 capex.

@krzyzanowskim This and Ruby on Rails are the two most important things that happened in human-based programming

Visual Basic IDE (1991) revolutionized programming. It democratized it, until JS-based web frameworks un-democratized it two decades later. - Visual designer (no code needed) - BASIC (no coder needed) - Fast (no expensive computer needed) - All-in-one (no toolchain or sophisticated build process needed)

@ChadMoran That was roughly 1,000 years ago now. If NN's solved the problem I stated then Tesla would shove 42 radars and lidars on every car and it'd look like a waymo.

@SteveC They ABSOLUTELY did. When Tesla was using Radar + Vision it was using imperative code, not a Neural Net. There's a massive difference there.

I attended Rivian's Autonomy & AI Day. Everyone's fixated on the wrong thing. The debate isn't Lidar vs no Lidar. It's early-bound vs late-bound sensor fusion. Tesla faced a hard problem with their early approach. Radar said one thing, vision said another. Reconciling sensor disagreements in imperative code is as hard as solving autonomous driving itself. So they killed radar and ultrasonics. Vision-only made sense - if humans can drive with eyes alone, computers should too. Rivian's doing something different. They're feeding all sensor data camera, radar, ultrasonic, and eventually Lidar - into an AI model. The model decides how to weight each input. Late-bound fusion. This is bigger than they're letting on. While you're arguing A or B, Rivian's building A and B. I rode in a production R1 Gen 2 with no Lidar during rush hour buildup. The system navigated intersections and crosswalks with pedestrians, slowed for a speed bump it wasn't explicitly trained on. Rating: C+ to B-. Two disengagements: one hesitating to squeeze into an emerging turn lane with backed-up traffic, one slowly rolling past a red light stop line. I was impressed because my expectations were low and this was their first demo. Yes, FSD 14.2.1 is better. But remember FSD beta 10's private beta? Those videos were rough. Rivian's timeline puts them behind where Tesla is, except they're learning from everyone else's mistakes. Second mover advantage. They trained this model in two months. It gracefully degrades with fewer sensors. That's the real story. Everyone's obsessing over Lidar inclusion. The AI assistant deserves more attention than it's getting. Wassym showed AI moving from enrichment feature to first-class citizen. I'm thinking about AI-powered climate scheduling, charging, service. The vehicle becomes AI-defined, not AI-enhanced. This is the beginning of something bigger than autonomous driving.

You're asserting that tesla had some kind of fundamentally different signal integration problem - they didn't - and that the model will solve it (hopefully, maybe). If a model can solve it for Rivian then Tesla could too. The real question is future cost (dollars, r&d, parts, cables, failure rates, integration cost, weight...) vs. safety gains. Rivian and Tesla took different approaches on that question, I just wish there was a polymarket on tesla reversing course I could bet against.

@SteveC Why is it any different than any other multi-modal problem?

@SawyerMerritt Sigh.

NEWS: Rivian has announced its autonomy subscription, Autonomy+, launching in early 2026 and priced at $2,500 (one-time) or $50/month. Rivian Gen 3 autonomy platform will include 11 cameras, 5 radars and a front-facing LiDAR. It will be powered by RAP1, Rivian's in-house silicon. Software advancements are coming to the company’s second gen R1 vehicles in the near term with the addition of Universal Hands-Free (UHF), bringing hands-free assisted driving for extended periods: • Available on over 3.5 million miles of roads across the USA and Canada. • Capable of operating off-highway on roads with clearly painted lines. "These features have the potential to make the roads safer, address customer demand and become veritable drivers for the business. We plan to continuously improve the autonomy capabilities of its Gen 2 R1 and future R2 vehicles, with a clear trajectory including point-to-point, eyes off and personal L4."

It's Pikachu time. I Built an agentic AI workflow to tell me exactly what trending video to make for the day. It scrapes viral trending data and automatically craft the prompts to use. Here's what it created Today 🐭🟡 See below an example of the prompts used 👇

@mfranz_on …OpenSCAD with cooler UI

This is MicroCAD. The programming language for your 3D projects. You can design your own Lego block in a few lines of code, ready to be exported as an STL file for 3D printing.

Nano Banana Pro is a really good cartographer. Used it to turn low res satellite imagery into a detailed hand drawn map and vector HD map. Pretty wild how well it segments everything and even recovers paths/roads hidden under tree cover. Looks way more detailed than the current google basemap which is pretty sparse in countries like India. Included both in video for comparison.