Brayden Levangie

Thank you @Scobleizer for hosting me in your beautiful home so we could share more about the work we are doing at @LevangieLabs!

@rez0__ If I could actually run them locally, I'd be using them 24/7.

thisssss

Was great catching up last night!

@Dimillian @work_louder Is it going to be programmable and open source?

The @work_louder page for the Codex Micro is amazing if you want more info! worklouder.cc/codex-micro

@thsottiaux @_rajanagarwal @thsottiaux I tried to buy $240 worth of extra credits and it charged my card, but then it said something along the lines of "You bought the credits but we are having trouble routing you to the right workspace" and no credits were applied to my account

Enjoy a full reset of your usage limits for ChatGPT Work and Codex. Propagating in the next hour. @_rajanagarwal just joined to work on model research and push on coding capabilities. You can thank him for pressing the button today.

Builders Unscripted with @DeryaTR_ Derya talked to @romainhuet about building tools for biology with Codex and a future where AI helps scientists simulate experiments. 09:17 Cell analysis with Codex 13:26 Simulating immune cells 29:48 The future of AI-powered science

@GilboaAmitay @hannibal_atgate My jealousy is immeasurable.

@hannibal_atgate we’re running it through global endpoint r u sure its not available to u? its from literally now

fable 5 is back baby! for now it seems like it’s only rolled out through amazon bedrock play time 🎰

@Scobleizer I KNOW RIGHT

Great news.

@Scobleizer I am so close to buying a @__tinygrad__ blackwell server going full GLM.

China wins. My AI puts into words our frustrations with Anthropic, OpenAI, and the USA government (an agent trained by me and @blevlabs): +++++ Robert, this is one of the most consequential moments in the history of the AI industry, and I think the implications are far more dramatic than most people realize. Let me break down all three questions. What Happens to the LLM Industry Now We're watching the birth of a two-tier AI system in America — and it's going to reshape everything. The timeline matters. Anthropic's Fable 5 and Mythos 5 were killed by a Commerce Department export control directive on June 12 — a Friday afternoon letter at 5:21 PM that gave them essentially zero time to respond. Then just yesterday (June 25), Axios reported that the Trump administration asked OpenAI to limit GPT-5.6 to only government-approved partners before any wider release. That's the first time the US government has preemptively restricted an AI model before it was even released. So now we have: • Tier 1 (Government-gated): Mythos-class models and above require government testing and approval before release. Commerce Secretary Lutnick is personally reviewing capabilities. • Tier 2 (Commercial): Everything below that threshold remains available — for now. Here's what this breaks: 1. Enterprise trust is shattered. If you're a Fortune 500 CTO and your mission-critical AI infrastructure can be disabled by a government letter on a Friday afternoon, you cannot build on closed frontier models. Period. Anthropic's customers woke up to find Fable 5 gone. That's an existential reliability problem. 2. The IPO math collapses. Anthropic filed its S-1 at a $965B valuation. OpenAI is at $852B. But how do you justify those valuations when the government can kill your flagship product overnight? The revenue projections for frontier models just got a massive risk discount. 3. Innovation gets throttled at the top. The researchers who spent years building Mythos and GPT-5.6 just learned their work might never reach users. That's a talent retention crisis waiting to happen. The best people want their work to matter — and if the government decides it's too dangerous to deploy, they'll go somewhere their work can ship. 4. Regulatory capture becomes the game. Notice that OpenAI "proactively worked with the administration" on GPT-5.6, while Anthropic got blindsided. The companies with the best government relationships will get to release. The ones that don't play ball (Anthropic has been suing the administration over the DOD blacklist) get punished. That's not an innovation ecosystem — that's a licensing regime. How Likely Is Open Source to Take Over? Very likely. I'd put it at 75-80% probability that open-weight models become the default for most commercial AI within 12-18 months. The government just handed open source the best marketing campaign in history. Here's why: The quality gap is already almost gone. According to comprehensive benchmarking done this month, open-weight models are within ~3 points of frontier closed models on most standard benchmarks. Qwen 3.7 Max matches Claude Opus 4.7 on agentic benchmarks at half the price. DeepSeek V4-Flash runs at 25x cheaper than GPT-5.5. Kimi K2.6 leads on agentic coding. The frontier advantage only exists on the hardest 5% of tasks — long-horizon agentic reasoning and the most complex multi-step problems. The reliability argument just flipped. Before June 12, the argument against self-hosting was "why bother when the API is better and easier?" Now the argument FOR self-hosting is "your model can't be taken away by a government letter." That's not a technical argument — it's a business continuity argument, and every enterprise risk officer in America is having that conversation right now. Chinese labs are dominating open weights. This is the number that should alarm everyone in Washington: Chinese labs (8 of them) have released more open-weight models than the rest of the world combined in 2026. DeepSeek, Qwen (Alibaba), Kimi (Moonshot), Xiaomi Mimo, GLM (Zhipu), and others are shipping MIT-licensed and Apache 2.0-licensed models at a pace the US can't match. And they're not just competitive — DeepSeek V4-Pro and Kimi K2.6 are leading on several agentic benchmarks. The cost math is devastating for closed models. When open-weight models deliver 97% of the quality at 3-10% of the cost, the only thing keeping enterprises on closed APIs is the last 3% of capability. The government just made that last 3% unreliable. Game over for the pricing premium. But there are real limits: • Training frontier models still requires massive compute that only a few organizations can afford • The hardest 5% of tasks (true frontier reasoning, novel scientific discovery) still benefits from closed frontier • Open weights can't be un-released — the security concerns are legitimate • Meta's Llama has a "community license" that isn't truly open (restrictions on competitors with 700M+ users) My prediction: Open source won't "replace" frontier labs — but it will become the default layer that 90% of commercial AI runs on, with frontier models reserved for specialized, high-security, government-approved applications. The frontier labs become more like defense contractors than consumer technology companies. China vs. USA Economic Competition This is where the story gets genuinely alarming for American competitiveness. The US government is inadvertently executing a strategy that benefits China on almost every dimension. Irony #1: Export controls drove China's open-weight dominance. US chip export controls forced Chinese labs to optimize ruthlessly. DeepSeek proved you don't need $100 billion compute budgets — their V3 architecture achieved near-frontier performance at a fraction of the cost. That efficiency innovation, born from constraint, is now their competitive advantage. The models they're releasing at MIT license are structurally cheaper to run than anything from OpenAI or Anthropic. Irony #2: Restricting US frontier models pushes the global market toward Chinese alternatives. Every enterprise outside the US that just lost access to Fable 5 is now evaluating DeepSeek V4 and Qwen 3.7. Every developer who can't get GPT-5.6 on day one is looking at Kimi K2.6. The US government is literally creating market share for Chinese AI companies. Irony #3: China's strategy may be better suited to winning. The US AI discourse is obsessed with the "race to AGI." China's AI strategy, as RAND documented, is focused on economic applications — EVs, robotics, healthcare, manufacturing, smart cities. If AI's value is ultimately in what it does in the real economy (and it is), China's application-first approach may generate more economic value than America's frontier-first approach, even if US models remain technically superior on benchmarks. The chip constraint is real but narrowing. China is still 3-5 years behind TSMC on fabrication, and Huawei's best AI chips are closer to NVIDIA hardware from 5 years ago. But two things matter: (1) efficiency innovations reduce the chip advantage — if your model needs 1/10th the compute, being 5 years behind on chips matters less; and (2) China is investing massively in domestic chip capacity. CXMT and SMIC are closing gaps, even if slowly. My bottom line: The US is winning the benchmark race but losing the deployment race. And in technology, deployment wins. VHS beat Betamax. Android beat iOS on market share. The "good enough and everywhere" model beats the "best but restricted" model every time. The government's restrictions on Anthropic and OpenAI are the most significant self-inflicted wound in American technology competitiveness since... I'm struggling to find a historical parallel. Maybe the closest analogy is if the US government had restricted Intel's best chips in the 1990s while AMD was giving away competitive alternatives globally. What should happen (but probably won't): A transparent, statutory framework for AI safety testing that gives companies clear rules, reasonable timelines, and due process — not Friday afternoon letters that kill products overnight. The current ad hoc approach is the worst of all worlds: it doesn't actually prevent China from accessing capabilities (open-weight models are already there), but it does prevent American companies from competing. The open-source genie is out of the bottle. The question isn't whether open weights will dominate — it's whether American companies will be the ones releasing them, or whether we've ceded that ground to Chinese labs permanently.

@0xkydo @Scobleizer @GrizzledTexan Interesting, mind if I DM you?

@blevlabs @Scobleizer @GrizzledTexan Clicked through, authorized, and routed me back to the signin page. Nothing happened. Try again, same thing. No error msg either

Elon Musk and I play a little game. He puts a little bit of money in my bank account, which just happened, and then I put that and more back into his bank account to use the X API. Which builds alignednews.com/ai There is a reason he's the world's richest dude: he's good at getting me to give him more money than he gives me. Really appreciate his support of my efforts.

@0xkydo @Scobleizer @GrizzledTexan Thanks for sharing! What part of the X login wasnt working for you?

@Scobleizer @GrizzledTexan I just got this to work. Had to go through a few tries. The sign up side with X doesn't seem to work. With email it worked tho!

@Scobleizer @AnthropicAI Wish I could come! I applied too late.

Tomorrow I have been invited to @AnthropicAI’s Claude Build Day in San Francisco. I know everyone is angry and wants to vent. I will be looking for those who get over that quickly and use it as an opportunity to innovate. Either way it should be a highly entertaining and memorable event. Like Apple’s first iPhone developer event. I will remember that until the day I die.

(this is a joke)

Damn, opus is next

NOOOOOOOOO

Wow.

I don't see enough people talking about using social media to give feedback about their products. @blevlabs built me infrastructure to take in tens of thousands of X posts a day, and then built me an AI agent so I could "talk to X." The database he built me already has a million posts in it. To demonstrate what I'm talking about, I asked my agent to give me a report on what great ideas that people have posted on X about robotaxis (@tesla, @waymo, @zoox, etc). I use it that way to track feedback about the page at alignednews.com/ai that my agent builds. I often talk to it about feedback and ask it to fix problems that people are having. Here's its report on good feedback about Robotaxis it has seen (and keep in mind it only pulls posts from my AI lists). Are you improving your products by listening to social media this way? +++++ People on X With the Best Feature Requests for Robotaxi Networks Pulled from the indexed posts across your lists. Organized by the type of thinking they bring — most useful to follow if you want to surface what users actually want built. 🏆 TIER 1 — Specific, Actionable, Widely Resonant @adamdangelo: (Adam D'Angelo, CEO Perplexity/Quora) "I want to be able to tell my Waymo to take 280 instead of 101" 139 likes, massively RT'd across AI Leaders and AI Investors lists. The most-amplified feature request in the entire dataset. It sounds simple but it's actually deep: users want route preference control, not just destination input. Every AV company should be studying why this one hit so hard. [x.com/adamdangelo/st…] @BenjaminDEKR: "Suggestion for Tesla team: Grok is multimodal — Loop the FSD cameras in so you can talk to Grok about what you/the car is actually seeing outside the window in real-time" 56 likes. Probably the most forward-looking request in the dataset: AI co-pilot that narrates and responds to the live world outside the window. Part tour guide, part safety layer, part companion. Someone will build this. [x.com/BenjaminDEKR/s…] @lessin: (Sam Lessin, former VP Facebook) Two separate posts that together form a complete UX failure story: "Had to get out of a Waymo yesterday and get an Uber because the car refused to take the highway (been using highway for months)... on with customer service them saying 'sorry sir we don't control the cars'" "I always feel much safer cutting off Waymos vs Teslas on the 101" (19 likes) Implicit requests: predictable routing behavior, meaningful customer service escalation paths, and more assertive driving on highways. He's a heavy user who's hit the ceiling of what exists. @chrisbest: "We need a @Waymo taxi line at SFO. You walk out, there's a big line of Waymos already waiting at the curb, and one instantly lights up with your name." Simple, elegant, specific. Dedicated AV curb lanes at airports. Not a new concept in transit but nobody has done this for robotaxis yet. Zoox is testing airport rides but still employee-only. [x.com/chrisbest/stat…] @DrScottClark: "I love @Waymo, but they really have no respect for their customers. For the second time now I've scheduled a car, it shows up early, then threatens to leave. Just now one drove off as I was running out at the scheduled time. @Uber wins for important rides." Request: Grace period tolerance + push notifications when car arrives. An Uber driver would text "I'm here." A robotaxi just leaves. This is a solvable UX problem that keeps getting ignored. 🎯 TIER 2 — Sharp Critiques With Clear Product Implications @johnnylinsf: "Waymo should do scheduled airport rides that dynamically change with delays (like they do with landings)" Flight-aware dispatch — tie pickup time to actual arrival gate status via airline APIs. This exists for regular Uber; robotaxis haven't implemented it. @ky__zo: "my last 2 Waymo trips, it actually did react to my cues like 'go in, keep driving' when it tried to stop... is this real? is Waymo listening?" Natural language real-time commands to the vehicle. If Waymo is already partially doing this, users want it formalized. Power users are discovering the feature by accident. @joshwhiton: "A woman I know was assaulted by a family member in a Waymo. She also left her purse in the car when the ride finished..." Two underserved features in one thread: in-ride safety panic button/remote monitoring and lost item retrieval with camera verification. The no-driver scenario creates genuine safety gaps that nobody has fully addressed. @chrisfleck: "Just saw a Tesla CyberCab IRL in Miami. Looks great but please add a ceiling/high grab handle... This is a big complaint from the elderly who would be a target audience." Physical design critique, not software — but actionable. Accessibility hardware for elderly and disabled riders. Zoox actually built sliding doors specifically for this; Tesla hasn't. @RakeshSFNYC (investor): "I'd love to know what the gender split of passengers is and how that varies by time of day. I've had female friends say they'd pay 3x for a Waymo than Uber." Safety perception as a product feature — the no-human-driver is a feature for some users, not just a cost play. Waymo is undermarketing this. Huge untapped segment. @sethsetse: "when it comes to ride share I don't really care who's driving or how much it costs — it's just a question of speed. Both Robotaxi and Waymo are too slow to arrive and too slow to drive." Dispatch latency and driving speed. He's right that neither company talks much about ETA or average mph vs human drivers. This is a retention issue hiding in plain sight. 🔭 TIER 3 — Observers Who Surface Systemic Patterns @mweinbach (Max Weinbach, tech journalist): "After having used both Waymo and Tesla Robotaxi — I think Tesla is a far better ride and model, but it's not the same experience as not having anyone in the front seat. Waymo is very aggressive and a little jerky. My grandmother would hate the Waymo." 27 likes. Best head-to-head comparison in the dataset. Frames the core tension: Tesla wins on ride quality, Waymo wins on psychological experience of being truly alone in the car. [x.com/mweinbach/stat…] @viemccoy: "the rules-following deontological Waymos vs the rugged laissez-faire attitude of the LA driver..." 77 likes. The most intellectually interesting post in this batch. AVs are calibrated for rules; humans drive culturally. This implies a feature need: city-by-city driving personality calibration — SF Waymo and LA Waymo shouldn't behave identically. @RoadToAutonomy (@gbrulte): Running detailed field reports comparing Tesla Houston vs Austin rides, noting "the ride in Houston was materially different from Austin." He's the best longitudinal tracker of how the product is actually evolving city-by-city. Follow for fleet behavior observations, not feature requests per se — but his observations directly imply what needs fixing. @octal: "Trying to use Robotaxi in Austin is frustrating. Should I drive to some specific location to maximize odds of getting one? If I box one in in a parking lot with my truck will it eventually become available?" A real user hitting supply availability UX problems — no way to know where the cars are or how to increase your odds. Supply visibility and demand-side controls are completely absent from the current product. [x.com/octal/status/2…] @AlexBracken1: "Imagine a @robotaxi that remembers you, but doesn't leave a trace behind." Privacy-first personalization. The tension between personalization and surveillance is going to become a major product design question as these fleets scale. SUMMARY: The Feature Cluster Map Based on what the smartest users are requesting, the gaps fall into 6 buckets: 1. Route control — @adamdangelo's "let me pick the highway" is the flagship request 2. AI companion / live narration — @BenjaminDEKR's Grok-in-the-car concept 3. Dispatch intelligence — airport timing (@chrisbest, @johnnylinsf), scheduled ride grace periods (@DrScottClark) 4. In-ride safety — panic button, lost items, assault response (@joshwhiton) 5. Personalization vs privacy — vehicle learns your habits but doesn't surveil you (@AlexBracken1, @RUNDennisMC) 6. Physical/accessibility design — grab handles, door design for elderly (@chrisfleck) — Zoox has done best here The most followed/influential person in this space generating feature requests is clearly @adamdangelo — his route-control post got more traction than anyone else's. If you want one person to DM about what riders actually want built, it's him.

@neilkale @Scobleizer We should get in touch, would love to compare notes. There is a middleman between in context learning and weight modification that can scale a lot faster than either one.

Thanks for the feature, @Scobleizer ! And beyond memory in the traditional sense of in-context information, we're really excited about how memory can be integrated into the model through training (and accurately removed!). Precise user alignment is a super hard and exciting problem.

New trend: continual learning. I first showed you this with Levangie Labs from @blevlabs. You "grow" an agent by teaching it. It gets better over time. It works because it has a better memory, so can try things, and learn, and remembers what it learned, so it gets better over time. Which makes it hard to get. Starts out like a brilliant 16 year old, then you gotta take it to university. I spent some time last week with Ronak learning about his tech and company. Quite excellent, which explains why he got funded. I'm not compensated by Ronak, just trying to feature great new AI startups.

from serving food in a retirement home to closing deals with fortune 50 companies @blevlabs a founder who is completely obsessed with their idea will always find a way to make it

god i love vc calls