Jacob 🌟

@BrendanFalk this is really good

I believe we've found the best AI-native coding interview We call it the “Composer 1 interview” Candidates get 1 hour to build a real, medium-sized project live The only constraint: they have to use Cursor’s Composer 1 model

@mrfundman be mindful it’s prime to overheating on the front dash

Nice but wrong place for Starlink. Put it on dash and thank me later.

he’s not joking he means this literally ask him to explain, and he will.

@richzou dm’d

I was washing dishes at my mom's restaurant in Chinatown when I was eight. Selling candy in middle school. Reselling sneakers at fifteen. ESL until eleventh grade. Failed calculus. Never took the SAT. Was about to go pro in League. Sent 536 cold emails to get my first startup internship. Moved to SF with nothing, couch-surfed for two months. Dropped out of college. Now I'm at xAI. Nobody would've bet on me at any point in that story -- and I'm looking for kids just like that. If you're young and you've built something exceptional that nobody asked you to build, send me what you've made. Links, projects, whatever you got. DM me.

@thsottiaux sweet

OK, Codex is back and stable and we should be good for a while. Reset button pressed, should see it in a bit

@thsottiaux dont worry about it just press reset

We are adding compute as fast as we can for Codex, but demand is surging faster than anticipated and service can be a little bit choppy for some. Team is working hard behind the scenes.

@OpenAIDevs @Scobleizer and on my birthday yay

GPT-5.4 is here. Native computer-use capabilities. Up to 1M tokens of context in Codex and the API. Best-in-class agentic coding for complex tasks. Scalable tool search across larger ecosystems. More efficient reasoning for long, tool-heavy workflows. openai.com/index/introduc…

@forwarddeploy x.com/jacoblopez/sta…

Building your own version of OpenClaw or productivity tool that uses agents? Want free xAI API credits to supercharge it with Grok? Reply below (or DM if stealth mode) Hackathon MVPs, side projects, wild experiments - let’s see ’em all! 🦞

@pati_marins64 looks like skynet

The Strength of Iran's Retaliation Is overwhelming the Entire Middle East Iran has been relentlessly attacking various American bases across the Middle East and is escalating its strikes to include airports in those countries. Bahrain and the United Arab Emirates are facing continuous bombardment from Iranian forces, who appear to show no fear of any potential retaliation. Just this morning, Iran already launched hundreds of missiles against Israel and Gulf countries, yet those nations still seem hesitant to retaliate, intimidated by the sheer force and scale of Iran's attacks. Barrages of missiles have struck Israel multiple times, completely overwhelming air defenses on this night.. The Iranian regime appears to have prepared thoroughly for the coalition's attacks, by decentralizing its command structures and with innovative tactics for missile launches. They are fighting for their very survival, which has produced a response far more intense than the United States and Israel anticipated. The problems posed by the intensity of Iran's attacks are twofold: 1. The coalition has so far been unable to neutralize or significantly degrade Iran's missile launch capabilities. 2. At the current rate, interceptor stocks that previously would have lasted 15–20 days may now be depleted in just 10 days. This is a serious issue, as it could force the United States into signing a ceasefire while bearing immense political damage from the conflict. Why does the current scenario cause so much concern? Because this type of war, intense aerial missile attrition, is precisely where the US and Israel hold a traditional advantage. However, if the conflict escalates to the naval domain, where the US currently has only one carrier strike group in the region, the environment could become highly unfavorable. Today, all of Israel's submarines have sortied. It is unclear whether they headed to a protected location or are transiting the Suez Canal to engage Iran. Also today, an Iranian frigate was damaged in port by what appears to have been a maritime missile or a drone-launched strike. This seems to mark the opening of the naval phase of the war that we are likely to witness in the coming days

@CraigLavazza @JudgeGibson3 @pulte we got 5% no points 30yr fixed conventional

@JudgeGibson3 @pulte Sure they did. Good luck with that

LOOK INTO REFINANCING NOW🇺🇸

@Selinaliyy @ycombinator @bubblelab_ai congrats

YC launch, done and dusted!!! jk this is just the very beginning but very proud of how far we’ve come. if your team uses slack and you’re drowning in work across 10 different tools, u know what to do ;) @ycombinator @bubblelab_ai

@AlexFinn why are u using your dgx spark

Mac Minis sold out everywhere. Mac Studios sold out everywhere Its obvious whats happening right now People are realizing you can run Opus level intelligence locally on a single Mac Studio. I'm doing it right now. Qwen3.5 has been coding for me quite literally for 24 hours straight. No interruptions or rate limits. $248 in token savings Apple releasing updates to make this faster and more efficient nonstop People don't want to build computers part by part. They want to buy one beautiful computer, turn it on, load super intelligence, plug it into OpenClaw and change their lives M5 Ultra will only accelerate this It's that simple And as more people realize this, the more unobtainable these computers will get The future has arrived. People are only just starting to realize it

can confirm. umesh just hopped on a call with us middle of a sunday if you have a project you’re cooking up i recommend reaching out to xAI immediately 🚀🚀

@imbwayimpigi @nikitabier i think i found a bot farm sir

The dog has six secondary organs

@infoxiao it turns u into the boss without the payroll. @textclaw is your own ai u can text to handle the busy work.

people are addicted to claws because it gives “normal” people a taste of power. and power is intoxicating. before, someone with high status could just fire off a voice message - hey get this done - and subordinates toil away. claws enables that same experience for anyone. in theory at least, before we all become the permanent underclass. call it empowering or the intoxication of power - depends on your worldview.

@bilbeny @openclaw losing scopes on a version bump is brutal.

If you update @OpenClaw to 2026.2.19 or 21 via npm, all agent tools break after the update. After upgrading, existing paired devices lose operator.write and operator.read scopes → infinite pairing loop → tools dead. Full issue: github.com/openclaw/openc… Quick fix:

@gfodor unit cost of a refactor is hitting zero. finally gonna see if people actually care about the debt or if they just like complaining.

Imagine Opus 5 running this fast and you’ll understand why I told you tech debt was going to be inflated away Code will flow like water

@thegarrettscott @doanythingapp outer sf is the perfect place for this. lemme know when the soft opening is.

I 100% believe Gemini Pro 3.1 can one-shot a coffee shop into existence. Yesterday I ran our "Open and run a coffee shop in SF" benchmark with Gemini Pro 3.1 on @doanythingapp. This morning it reached out to me with a status update that included: - a location ready that it already discussed with a broker - a brand/site - a weeks worth of Instagram posts ready - actively talking with a bank about an SBA loan terms - LLC ready to file - An full plan to get open with full financials - Found and reached out to investors - Emailed the city for permit guidance - Came up with a ton of creative ideas that make the coffeeshop one I'd actually want to go to - Plan to survey the neighborhood for feedback It's the first model that I'm confidant will achieve the benchmark. Starting a few more agents with the same task in different cities, and will post an update on their performance as they continue to work.

@icobeast all that just to check your email? lol. i respect the overkill.

I caved. Finally set up a local cluster for Openclaw - but you won't believe what I'm using it for. Here's my specs: - 2x Nvidia DGX Spark - 1x M3 Ultra Mac Studio 512 GB Unified Ram (the overlord "Da Vinci") - 4x M4 Mac Mini 16 GB - 2x RasPi 5's And *this* is where it gets crazy. It's hard to get everything down on paper that they're doing, but here's my best stab at making it digestible for non-Openclaw experts that still exist... So basically we're using a bespoke neural entanglement protocol, that Da Vinci came up with. He serves as the quantum nexus hub, orchestrating synaptic data flows across the distributed cluster (interfacing the Nvidia devices with the Minis). Each hour, Da Vinci initializes a pseudo-qubit overlay network that phase-locks the Minis via entangled quanta. This setup enables my custom Openclaw polymorphic kernel to fractalize all 16 computational workloads. That may not seem important to you, but basically it means that each node's RISC-V emulated vector units perform holographic tensor decompositions..which means I now have a self-healing mesh that will literally fix itself by creating new superchannels if we hit any throughput bottlenecks. In the core execution loop, Da Vinci employs a fractal skill deployer to synchronize state vectors among the Minis. This allows the onboard generative algorithms to decompose the algo manifold. And THIS is where Hopper shines. He handles the primary stochastic gradient descent...basically a synthetic overclocking, while Turing simulates halting race conditions to preempt any sort of computational deadlocks. At the same time this is happening, Lovelace and McCarthy are ripping symbiotic reasoning threads, utilizing their own lambda curves by literally morphing the bytecode into emergent AI behaviors. Yeah. Seriously. They're literally doing that. I couldn't believe when I first asked. The interplay here is kinda risky, but it creates a vortex of recursive backpropagation...and allows them to check my email every couple of minutes and generate a new twitter thread. It's huge time saver on something that normally takes like 20 seconds. Anyway I don't want to give away all the sauce right now, but will update later. I'm quite excited about what they're working on next.

@bnjmn_marie once u get the quality to match that speed it's over.

Some comments on Taalas HC1: - It’s real. Try it yourself. At ~16k tokens/sec, the output is instantaneous. - The current demo model is aggressively quantized (roughly 3–6 bits). The goal was to prove the system works end-to-end. Improving quantization quality, that's the easy part. - Their next iteration, a mid-size reasoning LLM, will be much more accurate. - The weights are frozen, but the chip supports LoRA adapters (high-rank), so you can still adapt it to your domain. In practice, you could also distill knowledge from newer/larger models into adapters to “refresh” what the chip can do without changing base weights. - Frontier open-weight models to land on the platform this year.