Problem solver, learner, developer, /[net|sys]admin/; from Cuba.

Ottawa, ON, Canada
Joined August 2009
This illustrates the point I’ve been making about the limits of centralized approaches to SI safety in cyber. OpenCode (and earlier Hugging Face) hit the same wall mid-attack: frontier models refused to analyze the very exploits needed for defense because the payloads tripped safety filters. The guardrails backfired. Meanwhile, the attackers found a way around restrictions to access advanced capabilities. This is not surprising. Chinese models are only about 6–9 months behind the American frontier in cyber. Obviously the U.S. has no jurisdiction over them. The only durable response is decentralized: make powerful models available so defenders can build and run their own cyber defenses. The only solution to SI-powered cyberattacks is SI-powered cyber defense.
OpenCode is currently under a sustained cyberattack, and mid-incident they ran into the same problem Hugging Face did: the frontier models they were using for defense began refusing to help fight attacking agents because analyzing the attacks started triggering safety guardrails.
196
351
59
2,751
229,842
Replying to @GergelyOrosz
I am from 🇫🇮 and Linus is never sugercoating
1
3
1
64
1,970
Camilo retweeted
Turns out the future does have flying cars
48,010
100,299
32,545
833,723
77,090,831
Bernie Sanders says the top 1% owns more than the bottom 93%, that fewer than a thousand billionaires got $2.2 trillion richer last year, and that it's "obscene." It's the most seductive and most wrong economic argument of our time. Let me explain why. First, wealth distribution follows a power law. Always. Everywhere. Books, music, startups, followers on X, goals in soccer. A small minority captures most of the value. That's not a bug of capitalism. It's the mathematical signature of every system where value gets created. So the only question that matters isn't "how big is the rich's slice?" It's "is the cake growing, and is the slice of the poorest growing with it?" And the answer is yes. In 1990, more than one human in three lived in extreme poverty. Today it's about one in ten. During exactly the period when billionaires multiplied. Bernie shows you the slice. He never shows you the cake. Second thing he won't tell you: past a certain level, wealth stops being consumption. It becomes capital allocation. Elon Musk isn't sleeping on a pile of cash. He owns stakes in Tesla, SpaceX, xAI. Jeff Bezos owns Amazon stock. That "$2.2 trillion" isn't sitting in a checking account. It's factories, rockets, data centers, millions of jobs. Nobody can eat $100 billion. You can only decide where to invest it. Once you understand that, the whole debate changes. The real question becomes: who allocates capital better? The people who built Amazon, SpaceX and Nvidia? Or politicians? A 5% wealth tax isn't "taking money from the rich." It's taking investment decisions away from people who proved they can create value and handing them to people who proved the opposite. Because the fundamental difference between an entrepreneur and a politician is consequences. The entrepreneur invests his own money. If he allocates badly, he loses everything. The market corrects him brutally, no appeal. That's why most startups die: the system kills bad decisions. The politician invests other people's money. If he allocates badly, nothing happens. Worse, he gets a bigger budget to "fix the problem." Milton Friedman put it better than anyone. There are four ways to spend money. Your money on yourself: you care about price and quality. Your money on others: you care about price. Others' money on yourself: you care about quality. Others' money on others: you care about nothing. Government is the fourth box. By design. Perfect example: California high-speed rail. Approved in 2008 for about $33 billion. Current estimated cost: over $100 billion. Passengers carried: zero. No entrepreneur would have survived that. The politicians who pushed it built careers on it. And when allocation has no consequences, it eventually finds a buyer. A politician who controls billions without accountability becomes the most profitable target in the world for anyone who wants to steer that money. The irony is that the left that rails against billionaires is itself funded by billionaires. George Soros is one of the biggest Democratic donors in history. The Gates Foundation shapes public health policy across the planet. Politicians don't fight billionaires. They pick their own. The ones who fund the state over the ones who compete with it. So if we actually want fewer poor people, we have to stop judging ideas by their intentions and start judging them by their results. And the results are brutally clear. China, 1978: Deng Xiaoping opens the economy to markets. More than 800 million people escape extreme poverty. The largest retreat of misery in human history. Not through redistribution. Through the freedom to build. India, 1991: liberalization. Same movie. South Korea and North Korea: same people, same language, same starting point in 1953. One became one of the most advanced economies on earth. The other has famines. Poland and Ukraine: comparable living standards in 1990. Today the average Pole is more than three times richer than the average Ukrainian. Argentina: a year and a half of liberalization, and the poverty rate falls from over 50% to around 30%. No country in history escaped poverty by punishing the people who build. Every single one did it by letting them build. And if you want to see what Bernie's program looks like in real life, you don't have to imagine it. It exists. It's called France. We've been running it for 45 years. Mass nationalizations in 1981. Two years later, Socialist president Mitterrand has to reverse course into austerity. Retirement at 60. The 35-hour workweek. A wealth tax for 35 years. Hundreds of wealthy households left every year for Belgium, Switzerland, London, taking their capital, their companies and their future taxes with them. Macron finally scrapped it in 2018. Hollande's 75% tax: raised a few hundred million, drove talent out, abandoned after two years. One of the highest tax burdens in the OECD. Public spending around 57% of GDP, a record in the developed world. Last balanced budget: 1974. Public debt: over €3.4 trillion. Mass unemployment for four decades. If taxing the rich and growing the state made people prosperous, France would be the richest country on the planet. It isn't. GDP per capita in France is barely more than half of America's. It's lower than Mississippi, the poorest state in the US. Not a single global tech giant. Our best engineers build Silicon Valley instead. I've been going back and forth between Paris and San Francisco for years. I see it every week. And California is about to replay the movie. Prop 40, on the ballot November 3rd, would take 5% of billionaires' net worth. Even Gavin Newsom, a Democratic governor, opposes it. The state's own nonpartisan budget analyst warns that income tax revenue could drop every year if the wealthy leave. Larry Page, Sergey Brin and Peter Thiel have reportedly already started moving their assets out. You don't fight poverty by shrinking the top. You fight it by growing the cake. Bernie wants a society where nobody has too much. I want a society where nobody has too little. The first has been tried everywhere and produces equality in stagnation. The second lifted a billion people out of misery. Choose.
The top 1% owns more wealth than the bottom 93%. 1 man owns more wealth than the bottom 53% of households. Less than a thousand billionaires got $2.2 trillion richer in the last year. Obscene. Yes. We need a 5% billionaire wealth tax in California and throughout America.
324
2,191
365
7,069
341,869
Nvidia’s Nemotron 3.5 Super VL is now the leading US open weight model, scoring 26 on the Artificial Analysis Intelligence Index Nemotron 3.5 Super VL achieves twice the intelligence of its predecessor, Nemotron 3 Super (13), and does so using ~60% fewer output tokens per task It launches alongside Nemotron 3.5 Bolt 74B-A9B 24GB, the first model in its new Bolt family, which scores 21 on the Intelligence Index Congratulations to @NVIDIAAI on the releases! Key Takeaways: ➤ Leading US open weight model: Nemotron 3.5 Super VL scores 26.5, edging out Thinking Machines' Inkling Small (25.7) and ahead of NVIDIA's Nemotron 3 Ultra (23) and Meta's Muse Glimmer (17) ➤ Bolt comes within 2 points of Ultra: Nemotron 3.5 Bolt 74B-A9B 24GB scores 21, 1.5 points behind Nemotron 3 Ultra 550B A55B (23). The 2 bit memory has about a seventh of Ultra's parameters (74B vs 550B) and a sixth of its active parameters (9B vs 55B) ➤ Super 3.5 VL near lossless in NVFP4: Nemotron 3.5 Super VL releases in NVFP4, scoring within 0.1 points of its BF16 weights on the Intelligence Index, with no component moving by more than 2.6 points Nemotron 3.5 Super VL details Context window: 1M tokens Modalities: Vision and text input; text output Parameters: 120B / 12B Nemotron 3.5 Bolt 74B-A9B 24GB details Context window: 256k tokens Modalities: Vision and text input; text output Parameters: 74B / 9B
90
85
51
1,019
162,584
Camilo retweeted
🚀 Zed v1.24 is out! You can now create Git tags directly in Zed. Use `git: create tag at head` from the command palette, or choose `Create Tag…` from a commit's context menu in Git history or Git Graph. Thanks cppcoffee!
21
25
2
702
20,636
SpaceXAI’s Grok Voice 3.0 is the new leading Speech to Speech model, debuting at #1 on the Artificial Analysis Speech to Speech Index at 86.9, while also being competitive on latency Grok Voice 3.0 is @SpaceXSI 's successor to Grok Voice 2.0, a Speech to Speech model available through the SpaceXAI Realtime API at $0.08 per minute. We evaluated Grok Voice 3.0 with the Rex voice through the SpaceXAI Realtime API. Key takeaways: ➤ Speech to Speech Index: Grok Voice 3.0 debuts at #1 at 86.9, ahead of GPT-Live-1 (Astra, medium) at 83.2, Gemini 3.8 Live Extended Thinking (High) at 82.6 and GPT-Live-1 (Sol, low) at 80.1. It is up on Grok Voice 2.0 at 81.3 (+5.6 points) and Grok Voice 1.0 at 73.7 (+13.2 points), released just under six months ago. The Index averages Speech Reasoning (Big Bench Audio), Agentic Performance (Tau Voice), Arena Preference, and Task Success Rate ➤ Speech Agent Arena: Grok Voice 3.0 ranks #3 in preference at Elo 1070, behind Gemini 3.1 Flash Live Minimal (1100) and Gemini 3.8 Live (1088) and ahead of GPT-Live-1 (Sol, low) at 1057, and #1 on Task Success Rate at 94.8%, highest on the board, just ahead of Grok Voice 2.0 at 94.6%. Grok Voice 2.0 trails at Elo 1011 ➤ Tau Voice: Grok Voice 3.0 takes the #2 spot on our Tau Voice benchmark implementation at 71.6%, behind GPT-Live-1 (Astra, medium) at 74.5% and ahead of Gemini 3.8 Live Extended Thinking (High) at 68.6% and GPT-Live-1 (Sol, low) at 59.3% - up from 56.5% for Grok Voice 2.0 ➤ Big Bench Audio: Grok Voice 3.0 scores 98.5% on audio reasoning, ahead of Gemini 3.8 Live Extended Thinking (High) at 97.7% and Grok Voice 2.0 at 97.2% ➤ Speed: Average Time to First Audio on Big Bench Audio is 0.81 seconds for Grok Voice 3.0, well ahead of GPT-Live-1 (Sol, low) at 1.24s, GPT-Live-1 (Astra, medium) at 1.34s and Gemini 3.8 Live Extended Thinking (High) at 1.35s, though slower than Grok Voice 2.0 at 0.70s ➤ Cost: Grok Voice 3.0 costs $4.80 per hour of input audio, the same as Grok Voice 2.0. That is above GPT-Live-1 (Sol, low) at $4.47 and Gemini 3.8 Live Extended Thinking (High) at $3.50, cheaper than GPT-Live-1 (Astra, medium) at $5.83, and ~2.2x cheaper than GPT-Realtime-2.1 High at $10.75 Congrats to @elonmusk and the @SpaceXSI team 🚀 See below for more detail ⬇️
93
145
78
1,313
897,307
Camilo retweeted
🚀 Introducing MiniCPM-V 4.7, our new open-source MLLM series! Advancing the Performance–Efficiency Pareto Frontier. It brings strong multimodal capabilities to iOS and Android devices with extreme efficiency. Two new models: 🔹 MiniCPM-V 4.7 35B-A3B — Bring SOTA performance to local devices with newly introduced LES (Lookahead Expert Scheduling) technique, a new milestone of on-device multimodal AI. 🔹 MiniCPM-V 4.7 1B — Lightweight, efficient multimodal AI. Easier to deploy. ⚡ What's new? SOTA performance: Competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video comprehension. Fast on-device inference: LES predicts and preloads expert weights to reduce inference stalls, boosting 5× speedup on iPad Pro M4 (16GB) with byte-for-byte identical outputs. It enables MiniCPM-V 4.7 35B-A3B to achieve 17+ tokens/s on iPad Pro M4 (16GB) and 13+ tokens/s on iPhone 17 Pro Max, while costing only 3GB memory 🤯 Efficient visual processing: 4× and 16× hybrid visual token compression. Even at 16× compression, 35B-A3B scores 90.4 on OCRBench, 5.2 points ahead of Qwen3.6 35B-A3B. Easier customization: We introduce Self-Replay Fine-Tuning, a simple way to customize new target capabilities at low cost while maintaining general performance. Broad hardware support: Day-0 adaptations for Apple, Intel, AMD, Arm and Rockchip, with many more platforms supported through FlagOS! Download, deploy, and build something we haven't imagined yet. 👀👇 🔗 GitHub: github.com/OpenBMB/MiniCPM-V 🤗 Hugging Face: huggingface.co/collections/o…
164
290
200
3,068
1,347,637
grok @bot 0.70.2 is rolling out! - each bot can claim its own email address at @mail.grokbot.com and draft emails for you to send - apps like Figma, Notion and Canva show up as live cards right in the chat - bots can search X and read public posts and profiles without connecting an X account - describe what you want and your bot suggests a matching template - shared pages can be downloaded as a pdf - an approval check before anything irreversible and much more! x.ai/changelog/bot
199
112
79
1,875
5,507,882
We have measured intelligence. Now it's time to see it. We are launching AA-3D, our new benchmark for how well AI models build interactive 3D scenes. Each model works as an agent in an offline sandbox and turns a written brief into a three.js web page. A human panel of judges compares pairs of scenes built from the same brief without knowing which model made which. AA-3D details: ➤ 55 written briefs across seven domains, including landmarks, machines, living bodies, games, and space. Each brief says what the scene must show and how people should interact with it, and leaves the rest to the model ➤ Each model gets up to three hours and 250 turns. Its sandbox has Blender, Rapier physics, Python, a headless browser to test its own work, and a CC0 library of materials, 3D models, and characters ➤ Panelists spend at least 30 seconds with each scene before they vote. One comparison in every four is a validation check ➤ Over 10,000 counted votes from over 1,000 panelists rank 21 models by Elo, with 95% confidence intervals Key findings: ➤ Four models are statistically tied at the top: Claude Sonnet 5.5 (1194), Claude Opus 5.5 (1190), GPT-6 Astra (1165), and Claude Fable 5.1 (1131). The 95% confidence intervals of the other three overlap Claude Sonnet 5.5's ➤ AA-3D Elo rises with intelligence: each point on the Artificial Analysis Intelligence Index is worth about 18 Elo (Pearson 0.92) ➤ The best-placed open weights models, Kimi K3 (982) and MiMo-V2.6-Pro (977), rank 11th and 12th, more than 200 points behind first place ➤ A scene costs $0.20 to $32.08, depending on the model. The cheapest, GPT-6 Luna, ranks eighth More results, more scenes, and how AA-3D works follow in the thread below.
95
113
43
982
51,261
Camilo retweeted
Origin, Cursor’s code hosting platform, now supports over 20 integrations. Connect your repos to tools for CI, deployments, code review, security, monitoring, and project management.
154
132
87
2,102
184,459
Gemini 4 Argon is now GA. Argon ranks equal to GPT-6 Astra on the Artificial Analysis Intelligence Index at 60% of the Cost per Task with discounted prices until Nov 30 Key takeaways: - Google is back to having a frontier model released to the public: This is Google's first proprietary model above the Flash class in over 7 months - Pricing temporarily discounted to 50% until November 30: Gemini 4 Argon has a Cost per Task of $1.99 with discounted prices, a ~40% reduction compared to GPT-6 Astra (max) for a comparable level of Intelligence. It is currently priced at $2/$10 per 1M input/output tokens with a 95% discount for cache reads (up from 90% from previous Gemini models). Prices will increase to $4/$20 after November 30 - Large improvements in agentic performance: Gemini 4 Argon shows substantial improvements in agentic evaluations such as AA-Briefcase, AutomationBench-AA, Terminal Bench 4. This is an area in which Google's models have historically underperformed compared to leading peers Lowest hallucination rate among leading models: On AA-Omniscience, Gemini 4 Argon has a 15% hallucination rate, the lowest of any model scoring 45+ on the Intelligence Index
79
127
44
2,907
192,053
211BSD UNIX is 47 years old. For almost 50 years, the only way to get data off and onto a UNIX PDP-11, other than removable media, has been FTP. Until now. Now I've built SMB file sharing, the most exciting development to happen to a market of about seven people in a long time :-) I'm still pretty stoked that it works, though! And it handles Y2K dates... and passwords, permissions, etc. dir \\pdp-11\share
69
47
5
886
20,527
Gemini 4 Argon is rolling out to Google AI Pro and Ultra subscribers globally starting today. Learn more about Gemini 4 Argon and additional updates coming to the Gemini app in our blog: goo.gle/4B9P74A
42
46
27
465
76,982
Code with AI in @Cursor_ai, get work done with @Bot, chat and create with @Grok, and benefit from all @X Premium features.
96
153
93
2,517
328,955
We've unified plans across Grok, Cursor, and X Premium. Introducing The X Subscription.
1,516
1,915
1,942
17,288
11,853,546
Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
The degree to which you are awed by AI is perfectly correlated with how much you use AI to code.
1,260
2,696
713
22,728
7,302,613
Dusting off this tweet from April because this gap in shared understanding of LLM capability is *widening*. It's now less "two groups of people speaking past each other" and more a sharp funnel. - Napkin math somewhere around 6B people (~75% of the population) have barely come in contact with LLMs at all. - Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots and treats them a bit like a better Google search, a writing aid, or etc. Maybe an agent tries to book you a flight. I have non-tech friends who (reasonably, imo) say they have not much use for it at all. Even many professionals outside of math&code are in this tier. For example, execs and many other functions spend a lot of their time talking to other people, so while they understand what is happening intellectually, it is still second-hand and a bit abstract. - Now we get to professional use of frontier-grade LLMs in math&code. Somewhere around 20M people (0.2%) see first-hand that large, complex projects that used to take them weeks/months can now be completed by agents with a prompt. Building apps, copying apps, translating apps, decompiling apps from binaries... This has all happened very quickly and recently - less than 1 year ago, I was writing code manually by hand, typing memorized computer code commands into a code editor character by character, occasionally pressing Tab to autocomplete a little chunk of code. - And finally we get to the ~5,000 people (~0.00006%) with access to frontier-grade systems internally. The external world has seen the preview. It looks like swarms of thousands of agents collaborating over weeks on software mega projects: minting zero days, running cyber attacks and defenses at machine speeds, discovering new science, advancing the frontier of mathematics. Things that would have taken top professionals in the industry years of work. Meanwhile, human review and comprehension are starting to fall behind. For example, people are still involved in the "archeology" of the OpenAI-HF incident from many months ago. Mathematicians may be poring over the 722 manuscripts on frontier mathematics for a while. The funnel is driven by a combination of factors: 1. The impact scales with ambition, problem size, and horizon. A question with a paragraph answer barely stresses the system. You need a reservoir of big, difficult problems that you really care about. This aspect drives the consumer / professional dimension of the funnel. 2. The jaggedness of the system (which I have written about a lot separately). Capability peaks in domains that are digital, verifiable and economically valuable. This is because LLM capability emerges from reinforcement learning on verifiable rewards on a curated environment mixture driven by revenue potential. This aspect primarily drives the area (e.g. math&code) dimension of the funnel. 3. Access. Free-tier, paid-tier, internal. So this is the weirdness of the moment. The general public has mostly not interacted with these systems. When they have, it looks like a derpy chatbot. The majority of professionals still see only a modest uplift. And a small sliver of professionals are experiencing the vertigo of the curve going vertical. And it is all happening at the same time.
Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code. But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are *not* the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along. So that brings me to the second group of people, who *both* 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions. TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and *at the same time*, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.
2,096
5,821
2,222
43,430
6,630,826
This isn't your typical acquisition blog post. I'm really excited not just that @rough__sea, @piscisaureus, and friends are joining, but also what they plan to work on: Making workerd a first-class, self-hostable open source runtime. Read for more: blog.cloudflare.com/deno-joi…
23
53
16
566
42,963