𝗧𝗶𝗺𝗶𝗻𝗴 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗖𝘂𝗿𝘀𝗼𝗿 𝗮𝗰𝗾𝘂𝗶𝘀𝗶𝘁𝗶𝗼𝗻 The announcement of Cursor's $60B acquisition by SpaceX is timely, for two reasons: 1️⃣ It is bolstering the share price right now, as confirmed by the 10% bump today. As my colleague @Raph_dOrnano wrote, it adds tremendous credibility and potential to xAI, the weak link in the SpaceX assemblage. What she calls the "coding wedge" is a formidable competitive weapon and SpaceX gets it instantly... 2️⃣ ... as soon as the deal closes. The all-stock deal is expected to close in the 3rd quarter of 2026. The exact conversion ratio will be determined by the volume-weighted average closing price of SpaceX stock over the seven trading days leading up to the transaction's close. That could make a huge difference to how much of SpaceX Cursor shareholders end up with: at todays share price, it's about 2%. My forecast for Q3 share price is $60-$80, which would amount to 6%-7% of SpaceX. linkedin.com/posts/raphaelle…

My prediction that SpaceX would rise 70% in the first two weeks is on track! Up 58%. Now, the forecast for what happens after that is a little gloomier, regression to discounted "fair value." My bet is $60-$80. spacexipo.bonabeauapps.com

I find @arthurmensch's take extremely refreshing in the wake of the Chaton Fat - Gros Chaton memefest. It is both candid and visionary, I wish more AI leaders would follow his example. --- Arthur Mensch Cofounder & CEO @ Mistral AI We somehow got put in the spotlight the last few days! Good time to say a few things. We exist to make sure that everyone gets access to the best AI systems, outside of centralized control exercised by states or corporations that feel the need to control in-fine deployment of AI. We’re here to grow the surface of superintelligence, and to diffuse it. This means deploying as much capital and talent as possible in research, through an acceleration arc that we’re 3 years in. We started later than others and have spent a few percent of the deployment of some of our competitors, in a domain that is fundamentally governed by flops budget. Today, we do not yet own the best language models, but we’ve constantly reduced that gap. We have a very exciting model to come this summer – it will be open-weight, and we’re opening early access to it in July. In domains that are less compute bound, e.g. voice, vision and document processing, we have state-of-the-art solutions. What do we do for a living? We’ve been deploying our models and agent platform (through Studio and Vibe) on the infrastructure of our Enterprise customers, ensuring full control over their usage. We help our customers train models (through Forge). We help them with pro-services (FDE is the fancy name), because this is critical to make our customers successful. We’ve developed a strong global motion in manufacturing, financial services, public sector and defense, growing strongly in the US, Europe, and Asia. Following our investment in compute infrastructure, we’re also proposing hosted AI cloud services (through Studio, that provides high QoS tokens managed agents). We offer a service tier that is completely decoupled from US service providers, because it matters. We have capacity online, it’s growing fast, and we can help secure it. We’ve been focused on large enterprise customers - that’s what you do as a tech business if Europe is your anchor market. We haven’t been great at helping developers and smaller structures: we’re working on that. We’re building under the premise that AI technology is a commodity technology that every organization needs a secured and affordable supply of. States and organizations need to have sovereignty over that technology: they should own and control the systems that are specifically embedding their IP and tacit knowledge, and will end up running their most critical processes; they should benefit from efficient pricing, through AI systems that evolve to reduce costs and improve accuracy through human interaction. AI, just like oil in the 20th century, is about to become the major source of leverage and power in the world. Depending on how the coming years unfold, it will either lead to a world of wealth and abundance for all, or to the worst extractive economies that the world has ever seen. We’re there to fight for the first scenario, as we progress AI research and accelerate its diffusion across the world – we’re hiring if you like the quest. linkedin.com/posts/arthur-me…

Anthropic's Own Goal Excellent analysis by @ReedAlbergotti at @semafor . The US government's stunning decision to limit access to Fable to any foreign citizen can be debated 𝘢𝘥 𝘪𝘯𝘧𝘪𝘯𝘪𝘵𝘶𝘮, it is true that Anthropic has kind been asking for it. After so much "the sky is falling" warning, it should not be entirely surprising that others start seeing a piece of the sky crashing too. Whether a unilateral blanket ban is the right answer to a threat that is mysterious and opaque to the public is another question. It is definitely going to boost the status of open weight/open source models, the best of which come mostly from... wait, what? semafor.com/article/06/12/20…

Exactly @altcap! 𝗦𝗽𝗮𝗰𝗲𝗫 𝗺𝗮𝘆 𝗯𝗲 𝗮 $𝟭.𝟳 𝘁𝗿𝗶𝗹𝗹𝗶𝗼𝗻 (𝗼𝗿 𝗲𝘃𝗲𝗻 $𝟭𝟬 𝘁𝗿𝗶𝗹𝗹𝗶𝗼𝗻) 𝗰𝗼𝗺𝗽𝗮𝗻𝘆. 𝗪𝗵𝘆 𝗱𝗼𝗲𝘀 𝗶𝘁 𝗵𝗮𝘃𝗲 𝘁𝗼 𝗯𝗲 𝗻𝗼𝘄? I thought about the Elon factor and came to this conclusion: SpaceX is an amazing company, a feat of human engineering and ingenuity, a testament to grit, hard work and resilience (and Gwynne) and it would be foolish to "bet" against Elon and his vision. I actually believe that it will be the first $10 trillion company. But it is not there yet. It will earn its valuation, heck, maybe it will reach $100B in revenue by 2027 and then it will be worth $2 trillion. But not today. Even with the Elon premium, the risk is enormous. So why does it have to claim this value today? They have been working on this for 24 years, what's another 2 years? If I were not such an idealist, I would suspect that a lot of rich people who invested in various Elon ventures (X, xAI) need their money back. For sure, Larry Ellison needs his money back.

Thank you @mjmauboussin!

𝗦𝗽𝗮𝗰𝗲𝗫 𝗺𝗮𝘆 𝗯𝗲 𝗮 $𝟭.𝟳 𝘁𝗿𝗶𝗹𝗹𝗶𝗼𝗻 (𝗼𝗿 𝗲𝘃𝗲𝗻 $𝟭𝟬 𝘁𝗿𝗶𝗹𝗹𝗶𝗼𝗻) 𝗰𝗼𝗺𝗽𝗮𝗻𝘆. 𝗪𝗵𝘆 𝗱𝗼𝗲𝘀 𝗶𝘁 𝗵𝗮𝘃𝗲 𝘁𝗼 𝗯𝗲 𝗻𝗼𝘄? Fully agree with my colleague here (not surprising 🚀 ). I thought about the Elon factor and came to this conclusion: SpaceX is an amazing company, a feat of human engineering and ingenuity, a testament to grit, hard work and resilience (and Gwynne) and it would be foolish to "bet" against Elon and his vision. I actually believe that it will be the first $10 trillion company. But it is not there yet. It will earn its valuation, heck, maybe it will reach $100B in revenue by 2027 and then it will be worth $2 trillion. But not today. Even with the Elon premium, the risk is enormous. So why does it have to claim this value today? They have been working on this for 24 years, what's another 2 years? If I were not such an idealist, I would suspect that a lot of rich people who invested in various Elon ventures (X, xAI) need their money back. For sure, Larry Ellison needs his money back.

𝗛𝗮𝘀 𝗢𝗽𝗲𝗻𝗔𝗜 𝗮𝗰𝗾𝘂𝗶𝗿𝗲𝗱 𝗞𝗣𝗠𝗚 𝗮𝗳𝘁𝗲𝗿 𝗧𝗕𝗣𝗡? theregister.com/ai-and-ml/20… Thanks Craig Reynolds for surfacing this story!

🤖 𝗖𝗼𝗹𝗹𝗲𝗴𝗲 𝘀𝘁𝘂𝗱𝗲𝗻𝘁 "𝗲𝘅𝗲𝗿𝗰𝗶𝘀𝗶𝗻𝗴" 𝘁𝗵𝗲𝗶𝗿 𝗰𝗿𝗶𝘁𝗶𝗰𝗮𝗹 𝘁𝗵𝗶𝗻𝗸𝗶𝗻𝗴 𝗺𝘂𝘀𝗰𝗹𝗲 𝘄𝗶𝘁𝗵 𝗔𝗜 I find that "going to the gym with an exoskeleton" is of the best metaphors for (lack of) skill acquisition with AI: it might be good to have an exoskeleton at some point in the real world, but you need to grow your skills first without it. Or with it but as a sparring partner not an assistant. If anything else, so that you are ready when the AI fails and/or the wifi is down. Imagine an airline pilot who only knows how to fly a plane on autopilot: would you feel good sitting in row 17D? You will survive in typical average conditions ± 20%. That's why it is crucial to think through the use of AI in schools carefully. Again, not necessarily to ban it (except in specific situations), but to design the interactions with AI so that it provides personalized nudging and real growth, not full packaged answers the student will barely remember (if they understand at all). That is the opportunity, let's not squander it.

Me showing up at the app gym with my Claude Code Fable exoskeleton.

I think I am going to work 24/7 until 6/22.

Not having an opinion on SpaceX is going to cost you. How much must you pay not to play? Excellent piece by Alphaville on the semi-forced buying of SpaceX stock by benchmark-constrained fund managers who NEED to buy "in the stock memes"... ft.com/content/d4069188-30ca…

𝗖𝗹𝗮𝘂𝗱𝗲 𝗦𝗵𝗮𝗻𝗻𝗼𝗻'𝘀 𝘄𝗶𝗳𝗲 𝗕𝗲𝘁𝘁𝘆 𝘄𝗮𝘀 𝗮 𝗰𝗼𝗺𝗽𝘂𝘁𝗲𝗿 @brianchristian 's 2012 videotaped talk at the @sfiscience contains many little treasures. One of them begins at 15:40 in the video: Claude Shannon's wife was a computer... Until the middle of the 20th Century, computer was a job title, people who perform calculations. In fact, the word computer as it is used today was intended as a metaphor to help explain the new machinery: "𝘵𝘩𝘦𝘴𝘦 𝘮𝘢𝘤𝘩𝘪𝘯𝘦𝘴 𝘢𝘳𝘦 𝘪𝘯𝘵𝘦𝘯𝘥𝘦𝘥 𝘵𝘰 𝘤𝘢𝘳𝘳𝘺 𝘰𝘶𝘵 𝘢𝘯𝘺 𝘰𝘱𝘦𝘳𝘢𝘵𝘪𝘰𝘯𝘴 𝘸𝘩𝘪𝘤𝘩 𝘤𝘰𝘶𝘭𝘥 𝘣𝘦 𝘥𝘰𝘯𝘦 𝘣𝘺 𝘢 𝘩𝘶𝘮𝘢𝘯 𝘤𝘰𝘮𝘱𝘶𝘵𝘦𝘳." Talk about a reversal! Video: youtube.com/watch?v=8Zs-GQ-E…

𝗦𝘁𝘂𝗱𝘆 𝗦𝗵𝗼𝘄𝘀 𝗔𝗜 𝗖𝗮𝗻 𝗣𝗮𝘀𝘀 𝗧𝗵𝗲 𝗧𝘂𝗿𝗶𝗻𝗴 𝗧𝗲𝘀𝘁 𝗠𝗼𝗿𝗲 𝗥𝗲𝗹𝗶𝗮𝗯𝗹𝘆 𝗧𝗵𝗮𝗻 𝗛𝘂𝗺𝗮𝗻𝘀 The title of this nice BGR article did not surprise me (first because the Turing Test, as the article rightfully reminds us, is not very scientific and has highly limiting quirks; second because LLMs have become pretty good at mimicking human text, a super-evolved version of Eliza; and lastly because the preprint it refers to is more than 12 months old, an eternity in AI years even though the PNAS paper just came out). It is a nicely done, pre-registered study. As the study authors state, "𝘛𝘩𝘦 𝘳𝘦𝘴𝘶𝘭𝘵𝘴 𝘪𝘮𝘱𝘭𝘺 𝘤𝘶𝘳𝘳𝘦𝘯𝘵 𝘈𝘐 𝘴𝘺𝘴𝘵𝘦𝘮𝘴 𝘤𝘢𝘯 𝘦𝘧𝘧𝘦𝘤𝘵𝘪𝘷𝘦𝘭𝘺 𝘪𝘮𝘪𝘵𝘢𝘵𝘦 𝘱𝘦𝘰𝘱𝘭𝘦 𝘪𝘯 𝘴𝘩𝘰𝘳𝘵 𝘪𝘯𝘵𝘦𝘳𝘢𝘤𝘵𝘪𝘰𝘯𝘴, 𝘸𝘩𝘪𝘭𝘦 𝘢𝘭𝘴𝘰 𝘳𝘢𝘪𝘴𝘪𝘯𝘨 𝘲𝘶𝘦𝘴𝘵𝘪𝘰𝘯𝘴 𝘢𝘣𝘰𝘶𝘵 𝘩𝘰𝘸 𝘦𝘧𝘧𝘦𝘤𝘵𝘪𝘷𝘦 𝘵𝘩𝘦 𝘵𝘦𝘴𝘵 𝘪𝘴 𝘢𝘴 𝘢 𝘮𝘦𝘢𝘴𝘶𝘳𝘦 𝘰𝘧 𝘪𝘯𝘵𝘦𝘭𝘭𝘪𝘨𝘦𝘯𝘤𝘦." In fact I think it answers the question(s) about the effectiveness of "the" test: it is parlor game of historical significance that has outlived its alleged usefulness. But it did remind me of a delightful and deeply insightful talk I attended in 2012 at the Santa Fe Institute by Brian Christian, the author of 𝙏𝙝𝙚 𝙈𝙤𝙨𝙩 𝙃𝙪𝙢𝙖𝙣 𝙃𝙪𝙢𝙖𝙣 (2011) [and also Algorithms to Live By (2016, co-authored with Tom Griffiths), and The Alignment Problem (2020)]. In the book, he describes the prize given to computer programs best at passing for humans (the most human computer), but it is mostly about the other, weirder prize for the human person who does the best at convincing judges of their humanness: the most human human. The talk is available on youtube, link in the comments: it is well worth watching not just for the entertainment value but for what it teaches us about what it means to be human. BGR article: bgr.com/2187757/ai-passes-tu… PNAS article: pnas.org/doi/epdf/10.1073/pn… @brianchristian website: brianchristian.org/the-most-… @sfiscience lecture: youtube.com/watch?v=8Zs-GQ-E…

The clearest definition of the mysterious and often elusive "context layer". By Atlan's Prukalpa ⚡. contextandchaos.substack.com…

I know I am late to the party but I find this picture from @victorcoimbra really IS worth a thousand words to explain Jevons-like inflation in AI costs while token prices are free falling. victorcoimbra.life/blogs/tok…

𝗔𝗜 𝗔𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝗰𝗲 𝗥𝗲𝗱𝘂𝗰𝗲𝘀 𝗣𝗲𝗿𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝗲 𝗮𝗻𝗱 𝗛𝘂𝗿𝘁𝘀 𝗜𝗻𝗱𝗲𝗽𝗲𝗻𝗱𝗲𝗻𝘁 𝗣𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝗻𝗰𝗲, a preprint by a team from Carnegie Mellon University, University of Oxford, Massachusetts Institute of Technology and UCLA offers a well-designed "𝘳𝘢𝘯𝘥𝘰𝘮𝘪𝘻𝘦𝘥 𝘤𝘰𝘯𝘵𝘳𝘰𝘭𝘭𝘦𝘥 𝘦𝘹𝘱𝘦𝘳𝘪𝘮𝘦𝘯𝘵 (𝘕 = 354) 𝘰𝘯 𝘧𝘳𝘢𝘤𝘵𝘪𝘰𝘯-𝘴𝘰𝘭𝘷𝘪𝘯𝘨 𝘵𝘢𝘴𝘬𝘴. 𝘗𝘢𝘳𝘵𝘪𝘤𝘪𝘱𝘢𝘯𝘵𝘴 𝘸𝘦𝘳𝘦 𝘳𝘢𝘯𝘥𝘰𝘮𝘭𝘺 𝘢𝘴𝘴𝘪𝘨𝘯𝘦𝘥 𝘵𝘰 𝘢𝘯 𝘈𝘐 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯 𝘰𝘳 𝘢 𝘤𝘰𝘯𝘵𝘳𝘰𝘭 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯. 𝘐𝘯 𝘵𝘩𝘦 𝘈𝘐 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯, 𝘱𝘢𝘳𝘵𝘪𝘤𝘪𝘱𝘢𝘯𝘵𝘴 𝘴𝘰𝘭𝘷𝘦𝘥 12 𝘧𝘳𝘢𝘤𝘵𝘪𝘰𝘯 𝘱𝘳𝘰𝘣𝘭𝘦𝘮𝘴 𝘸𝘪𝘵𝘩 𝘢𝘯 𝘈𝘐 𝘢𝘴𝘴𝘪𝘴𝘵𝘢𝘯𝘵 (𝘎𝘗𝘛-5) 𝘢𝘷𝘢𝘪𝘭𝘢𝘣𝘭𝘦 𝘪𝘯 𝘢 𝘴𝘪𝘥𝘦𝘣𝘢𝘳. 𝘛𝘩𝘦 𝘈𝘐 𝘸𝘢𝘴 𝘵𝘩𝘦𝘯 𝘳𝘦𝘮𝘰𝘷𝘦𝘥 𝘸𝘪𝘵𝘩𝘰𝘶𝘵 𝘸𝘢𝘳𝘯𝘪𝘯𝘨, 𝘢𝘯𝘥 𝘢𝘭𝘭 𝘱𝘢𝘳𝘵𝘪𝘤𝘪𝘱𝘢𝘯𝘵𝘴 𝘴𝘰𝘭𝘷𝘦𝘥 3 𝘢𝘥𝘥𝘪𝘵𝘪𝘰𝘯𝘢𝘭 𝘵𝘦𝘴𝘵 𝘱𝘳𝘰𝘣𝘭𝘦𝘮𝘴 𝘪𝘯𝘥𝘦𝘱𝘦𝘯𝘥𝘦𝘯𝘵𝘭𝘺. 𝘗𝘢𝘳𝘵𝘪𝘤𝘪𝘱𝘢𝘯𝘵𝘴 𝘪𝘯 𝘵𝘩𝘦 𝘈𝘐 𝘤𝘰𝘯𝘥𝘪𝘵𝘪𝘰𝘯 𝘩𝘢𝘥 𝘢 𝘴𝘪𝘨𝘯𝘪𝘧𝘪𝘤𝘢𝘯𝘵𝘭𝘺 𝘭𝘰𝘸𝘦𝘳 𝘴𝘰𝘭𝘷𝘦 𝘳𝘢𝘵𝘦 (𝘮𝘦𝘢𝘯 0.57 𝘷𝘴. 0.73; 𝘱 < 0.001, 𝘊𝘰𝘩𝘦𝘯'𝘴 𝘥 = −0.42) 𝘢𝘯𝘥 𝘩𝘪𝘨𝘩𝘦𝘳 𝘴𝘬𝘪𝘱 𝘳𝘢𝘵𝘦 (𝘮𝘦𝘢𝘯 0.20 𝘷𝘴. 0.11; 𝘱 = 0.031, 𝘊𝘰𝘩𝘦𝘯'𝘴 𝘥 = 0.25) 𝘵𝘩𝘢𝘯 𝘤𝘰𝘯𝘵𝘳𝘰𝘭 𝘱𝘢𝘳𝘵𝘪𝘤𝘪𝘱𝘢𝘯𝘵𝘴." The authors took care of obvious confounders such as skill levels and similar effects were replicated for reading comprehension. The conclusion stated in the title seems warranted and supported by the evidence. One might argue that a 10-minute test where persistence is measured is of limited applicability, and indeed it is when taken literally (for example an interesting question would be do the effects persist after 1 week?). But I think the more interesting result is in the middle of the paper, when the authors analyze how the AI was used: "𝘗𝘦𝘳𝘴𝘪𝘴𝘵𝘦𝘯𝘤𝘦 𝘤𝘰𝘴𝘵𝘴 𝘸𝘦𝘳𝘦 𝘤𝘰𝘯𝘤𝘦𝘯𝘵𝘳𝘢𝘵𝘦𝘥 𝘢𝘮𝘰𝘯𝘨 𝘱𝘢𝘳𝘵𝘪𝘤𝘪𝘱𝘢𝘯𝘵𝘴 𝘸𝘩𝘰 𝘱𝘳𝘰𝘮𝘱𝘵𝘦𝘥 𝘈𝘐 𝘵𝘰 𝘴𝘰𝘭𝘷𝘦 𝘵𝘢𝘴𝘬𝘴 𝘧𝘰𝘳 𝘵𝘩𝘦𝘮 𝘥𝘪𝘳𝘦𝘤𝘵𝘭𝘺. 𝘜𝘴𝘪𝘯𝘨 𝘈𝘐 𝘧𝘰𝘳 𝘩𝘪𝘯𝘵𝘴 𝘰𝘳 𝘤𝘭𝘢𝘳𝘪𝘧𝘪𝘤𝘢𝘵𝘪𝘰𝘯𝘴 𝘥𝘪𝘥 𝘯𝘰𝘵 𝘱𝘳𝘰𝘥𝘶𝘤𝘦 𝘴𝘪𝘨𝘯𝘪𝘧𝘪𝘤𝘢𝘯𝘵 𝘪𝘮𝘱𝘢𝘪𝘳𝘮𝘦𝘯𝘵𝘴." In other words, and I think it matters a lot, 𝗵𝗼𝘄 𝘁𝗵𝗲 𝗔𝗜 𝘄𝗮𝘀 𝘂𝘀𝗲𝗱 𝘄𝗮𝘀 𝘁𝗵𝗲 𝗸𝗲𝘆 𝗱𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁𝗶𝗮𝘁𝗼𝗿, 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗔𝗜 𝘃𝘀 𝗻𝗼𝘁-𝗔𝗜, which goes to show that all these half-baked experiments that do not monitor or control how the AI is used and generate "AI is destroying your brain" headlines are doing the whole field a disservice. Using AI is not a sufficiently accurate description of an experiment. The biggest work that remains to be done is precisely to design the best way to use AI: here we see that using it correctly does not impede abilities. arxiv.org/abs/2604.04721

My post was censored, so here it is again.

Censorship on X? I did a test: after 3 days of paid "boosting", which promised an estimated 5,000 to 10,000 views per day, this post about the SpaceX IPO has... 128 views. With boosting, the number of views is not about how interesting the post is but how the X algorithm decides to serve your post. My $30 experiment for 128 views shows that the algorithm is hiding the post. @elonmusk

Yes you can. That's the point.