European industry and AI news. Projects: - Skyforge rust agent harness: https://xitter.cf/t.co/Lg4c9zemG9 - Fishing in the UK: https://xitter.cf/t.co/3SY525j9aV
Amsterdam
Joined December 2025
- Tweets5K
- Following451
- Followers209
- Likes7.4K
Dave retweeted
Absurd dominance from 6.1 Sol. Same cost as V4.1F, radically better results.
And a hole in "V4.1F-1017 is a 5.6 Sol replacement" (though this was known in this case, 5.6 was already maxing out the soft ceiling of the original CritPt)
Grok 4.7, on the other hand, surges to parity
After 4 months of careful expert review, we release CritPt v2. GPT-6 Astra leads the way with 83.1% accuracy (average of 5).
We made changes to 47 challenges in total, which breaks down to:
- 5 corrections to mistakes in the problem statement (e.g., wrong number, self-contradictory constraints)
- 19 revisions to address underspecification (e.g., missing conventions/constraints that gives ambiguity to final answers)
- 16 grader fixes (e.g., grader unstable, error tolerant range too tight) without revision to the problem statement
- 7 refinements even though no issues were found (e.g., rephrased to be less searchable, worded more rigorously)
We also fixed some minor typos (spelling, grammar) which are unlikely to affect model performance.
Here’s how we found the issues:
- David Louapre @dlouapre from HuggingFace reported 10 issues across 8 challenges in June 2026. We accepted 8 issues and revised 8 challenges accordingly. We also thank David’s helpful revision suggestions.
- Anthropic sent a list of ~65 issues across 31 challenges in Aug 2026. Upon careful review of the full list, we accepted ~28 issues and revised 19 challenges accordingly (two overlapping with David’s list).
- We conducted an internal expert review of all problems, including newly revised ones, involving cross checking full responses of GPT-6 Astra (max) with CritPt experts. We ended up fixing/refining 20 additional problems.
From 28.9% in Nov 2025 (GPT-5 high, coding tool) to 83.1% now (GPT-6 Astra max, coding tool), the progress AI has made on well-posed physics problems is remarkable in a year. CritPt consists of entry-level research problems for junior researchers: the kinds of warm-up exercises a hands-on advisor might pose for a fresh grad student, but still demand solid physics training and nontrivial physical reasoning.
In this sense, frontier models are crossing a “critical point”, but there’s a lot more beyond this transitioning point in real scientific discovery. Asking a good question can be much harder than solving one. A good question can be fundamentally ambiguous and constantly changing as we explore. And ultimately, decisive tests of physics require experiments. There is so much more to explore.
We sincerely thank our evaluation partner @ArtificialAnlys, who will co-host the leaderboard of CritPt v2 soon!
CritPt Website: critpt.com/
GitHub: github.com/CritPt-Benchmark/…
HuggingFace: huggingface.co/datasets/Crit…
Dave retweeted
We’ve spent the past year developing new primitives for GPUs. We believe this enables new algorithms to change how machines learn, open new markets, and dramatically reshape AI’s cost structure.
We have raised a $1 billion Series A, co-led by Index and Lux.
Dave retweeted
Another vulnerability few are talking about: I found my Claude Code exposing personal data in http headers because the sites it curled politely asked agents to identify their users 😊
Epistemic cybersecurity is the hot new category for 2027.
Dave retweeted
Back to the Qwen3.8-27B - 18 models tested
I ran 18 different quantisations and inferences combinations on my Studio M5 Ultra, you should find this interesting... (more in the thread).
Dave retweeted
Aloha!🌺 Meet Ornith-1.5-122B: No Tradeoff Between Strong and Fast
Stronger. Faster. Leaner. Ornith-1.5-122B delivers frontier-level capabilities with exceptional inference efficiency. With superior instruction following, stronger task decomposition, and deeper understanding of real-world workflows, it achieves state-of-the-art performance among open-source models of comparable size, while running on a single NVIDIA DGX card!!
Open weights under the MIT license.
Coding
✅SWE-bench Verified: 0.845
✅TB2.1: 0.813
✅CyberGym: 0.800
✅NL2Repo: 0.512
Agentic
✅HLE: 0.446
✅BrowseComp: 0.785
✅MCP-Atlas: 0.746
✅Toolathlon: 0.708
🤗Hugging Face: huggingface.co/ornith-ai/Orn…
We can’t wait to see what you build with Ornith-1.5-122B.
Dave retweeted
🚨 A MASSIVE internet security emergency is unfolding right now. 🚨
Two CRITICAL vulnerabilities have been discovered in PHP, potentially putting millions of websites at risk.
→ NEARLY 70% of websites with a known server-side language run PHP. The potential scale is ENORMOUS.
→ TWO CRITICAL security flaws have been confirmed, each scoring 9.2/10 in severity.
→ Attackers could potentially take control of vulnerable servers remotely, without authentication.
→ Reports suggest PHP versions dating back to 2004 could be affected. The exact scope is still unconfirmed.
→ EMERGENCY PATCHES are scheduled for TOMORROW at 10 AM ET. Website owners are being urged to prepare immediately.
→ Europe's CERT-EU is warning organizations to patch ASAP or consider TAKING EXPOSED APPLICATIONS OFFLINE if they cannot be secured.
→ WordPress, Magento, Laravel and potentially millions of other PHP-powered websites could be at risk.
→ Once patches go public, hackers could reverse-engineer them and race to attack servers that haven't updated.
The vulnerabilities are CONFIRMED. The full technical details remain secret.
Tomorrow, the race between hackers and defenders begins.
This could get ugly. FAST.
Dave retweeted
🚨 BREAKING: Node.js just dropped its OWN security bombshell.
4 HIGH-SEVERITY vulnerabilities patched TODAY, including one that could let hackers remotely execute malicious code on vulnerable servers.
→ 10 SECURITY FLAWS were addressed in today's Node.js release, FOUR rated HIGH severity.
→ REMOTE CODE EXECUTION: CVE-2026-111497 is a serious flaw in Node.js's Brotli decoder that could allow attackers to execute malicious code remotely.
→ TLS SECURITY BYPASS: Two separate vulnerabilities could trick affected applications into accepting certificates for the WRONG server under certain conditions.
→ Think about that. Your application could establish an encrypted connection while failing to properly verify WHO it's talking to.
→ THREE MAJOR NODE.JS VERSIONS AFFECTED: Node.js 22, 24 and 26 all required emergency security updates.
→ RUNNING NODE.JS 18 OR 20? Those versions are END-OF-LIFE and aren't receiving these official security patches.
→ PATCH IMMEDIATELY: Update to Node.js 22.23.4, 24.22.1 or 26.12.1, depending on your release line. If you're using Docker, rebuild and redeploy your containers.
PHP AND Node.js both facing major security disclosures on the SAME DAY.
If you run a website or application, TODAY is the day to check your servers.
And remember: the public release of security patches can give attackers a roadmap to systems that haven't updated.
Don't be the last one to patch.
🚨 A MASSIVE internet security emergency is unfolding right now. 🚨
Two CRITICAL vulnerabilities have been discovered in PHP, potentially putting millions of websites at risk.
→ NEARLY 70% of websites with a known server-side language run PHP. The potential scale is ENORMOUS.
→ TWO CRITICAL security flaws have been confirmed, each scoring 9.2/10 in severity.
→ Attackers could potentially take control of vulnerable servers remotely, without authentication.
→ Reports suggest PHP versions dating back to 2004 could be affected. The exact scope is still unconfirmed.
→ EMERGENCY PATCHES are scheduled for TOMORROW at 10 AM ET. Website owners are being urged to prepare immediately.
→ Europe's CERT-EU is warning organizations to patch ASAP or consider TAKING EXPOSED APPLICATIONS OFFLINE if they cannot be secured.
→ WordPress, Magento, Laravel and potentially millions of other PHP-powered websites could be at risk.
→ Once patches go public, hackers could reverse-engineer them and race to attack servers that haven't updated.
The vulnerabilities are CONFIRMED. The full technical details remain secret.
Tomorrow, the race between hackers and defenders begins.
This could get ugly. FAST.
Dave retweeted
“Only when the last opportunity has been fully regulated, the last talent fully lost, and the last technology innovation missed, will we realize we cannot eat legal papers.”
More regulation to enable please. Less to limit.
BREAKING NIUS: Für mich ist es einer der unglaublichste Auftritte von Bundeskanzler Friedrich Merz in dessen Amtszeit. Seit 17 Monaten erzählt Merz, er wolle Deutschland deregulieren, unnötige und schädliche Bürokratie beseitigen. Dann hat dieser Bundeskanzler die Möglichkeit, auf der Weltbühne, vor milliardenschweren Investoren dem Nvidia-Chef Jensen Huang eine Frage zu stellen. Und was fragt der Bundeskanzler? Er will wissen, wie man Künstliche Intelligenz am besten REGULIEREN kann. Als Huang antwortet, nunja, erstmal müsse die KI-Industrie in Deutschland aufblühen, bevor man sie regulieren könne, legt Merz nach. Er erklärt Huang, aus seiner Sicht sei entweder eine europäische oder gar eine globale Regulierung notwendig, um die "Schädlichkeit" von KI zu begrenzen.
Dave retweeted
看了一下DeepInfer的数。我觉得吧，前几天大家转载的那个SiliconData Index的价格，未来一年估计还得不回头地跌个95%+，具体什么时候影响到GPU Rental rate蛮难预测的。
这个价格体系下，梁圣买张卡估计100天就回本了。一旦有朝一日token价格真的暴跌到影响rental rate，所有股价肯定已经跌了一大轮。
给大家几个数，所有的数都是$ / 1 mm token，基本上1mm token 95% cached = 100-150页的薄书.
- Dylan一直在说Jane Street的变现价是$100-200 USD / 1 mm token
- Opus 5.5 / GPT Astra 的价格在deepinfer的体系下大概是1-2 USD / 1 mm token，比标价便宜个80-95%
- 变态的来了，Deepseek 4.1 Flash off-peak的价格在这是1.5美分，又比SOTA便宜个90%。
- 按照一般的大B300 Cluster，根据deepinfer的计算，一个GPU大概能每小时产出2.7 Bn token, 成本换算过来大概是0.25美分 / 1 mm token，这已经比我想象中便宜很多很多了。之前算过我记得是15美分 / 1 mm token，不过没有计算cache。这个level的成本大概率是大机房+高使用率+超高存储时间，一般机房估计都做不到这个theoratical.
- 这也就是大家看到的80-85%毛利 (1.5 / 0.25)。DeepSeek要是以比这个利润率稍好的功率赚钱的话，一个B300大概2-3个月回本，而不是8个月。真是无与伦比的暴利，比贩毒爽多了，自然是砸锅卖铁也要走私B300。
- 如果B300按照break even算的话，那么成本大概是0.13美分 / 1 mm token. 然后原材料其实是B300 ～600k价格的6%不到（按照产业链毛利计算）。
- 所以最后算下来马斯克如果去完全积压的话，每1mm token的最终生产成本可能是0.01美分, or 0.0001 USD。这基本上跟免费差不了太多了。
这种屠夫版的价格我不知道你怎么看，但是when you see the headline API price of Opus @ 8/40 USD input/output，我只能感觉说只要OSS + Post-train和中国模型蒸馏大法能够持续跟随 + 没有新的breakthrough，token 价格未来一段时间应该是以断崖式的下跌。至于什么时候触碰到rental rate, maybe someone smart can tell me.
Dave retweeted
Nvidia has now invested $10bn in European startups.
Here is every company they have invested in in past week:
UK 🇬🇧
@boltz_bio
@RiverLane_io
@PrimaMente
@hivergeai
Foresight
Germany 🇩🇪
@MireloAI
@AgileRobots
@aveliosmed
@reflexaerospace
France 🇫🇷🇫
@doctolib
@gopigment
@gs_ai_
@entalpic_ai
BREAKING: Nvidia has now announced 4 investments into German startups 🇩🇪:
@MireloAI - develops AI models that generate sound effects, speech and music for video, helping creators and production teams produce high-quality audio that perfectly matches their visuals on screen.
@AgileRobots - which builds AI-powered robots, from robotic arms to its Agile ONE humanoid, for manufacturing, logistics and healthcare.
@aveliosmed - develops an AI-native hospital information system that serves as the core system of record for clinical and administrative workflows, integrating AI directly into hospital operations to free up doctors and nurses to deliver better patient care.
@reflexaerospace - designs and builds high-performance satellites ready for launch in about 12 months instead of years.
The company has already invested in n8n, Neura Robotics and Black Forest Labs.
Dave retweeted
🚨 @NVIDIAAI just quietly dropped its own local inference server...
No announcement. Single-digit stars
It's called Flare and Apache 2.0
Built for SM120/SM121 GPUs RTX PRO 6000 Blackwell and DGX Spark
>PyTorch + FlashInfer at build time
>Rust runtime at serve time. No Python.
>CUDA graphs captured before serving
>FP8 KV cache by default
>Recipes for Qwen Flash, GLM and DeepSeek
github.com/NVIDIA/flare-loca…
Dave retweeted
A horrific Russian strike in the Kharkiv region. A “shahed” struck people directly – in a park near an ordinary café. Pure atrocity. Six people were killed, including children. A little girl was killed in a stroller – her name was Angelina, she was just over a year old, born in August 2025… Her mother’s name was Inna. She died in the hospital… Another girl who was killed was 11 years old. My condolences to their families and loved ones. Another 14 people were injured, including six more children.
I am grateful to everyone who supports Ukraine under these circumstances and helps us shoot down “shaheds,” shoot down missiles, and stop Russian assaults. These are horrific photos, but behind them are shattered lives and the truth about this war. Lives must be protected. Ukraine needs air defense. You can't think about making money with Russia as long as it continues its asinine war. We ask that sanctions on Russia stay in place as long as the killings continue. We ask America to be strong, and we ask it to apply the Lindsey Graham law. We ask Europe to continue its support. We ask that everything possible be done to bring peace and guarantee security. Ukraine wants peace. It is Russia that has insisted on continuing the killing all this time. The world has enough strength to stop the war. That strength must finally be used – against the murderers.
Dave retweeted
🚀 Introducing DeepInfer, our in-house inference framework, now fully open source!
Powering DeepSeek-V4.1-Flash at scale with high throughput and low latency. We’re sharing our production stack to help the community lower compute costs, reduce energy use, and make AI more affordable.
🔗 github.com/deepseek-ai/DeepI…
1/4
Dave retweeted
Nvidia CEO Jensen Huang to Chancellor Merz:
I've also heard you say that you want to find that balance between innovation and regulation.
There's nothing to regulate if the industry doesn't thrive.
Dave retweeted
Electric car fleets displaced 200 million barrels of crude oil in the six months following the start of the Hormuz energy crisis, saving oil-importing countries $20bn 📉
EVs are a powerful lever for reducing vulnerability to fossil supply disruptions ⚡
ember-energy.org/latest-insi…
Dave retweeted
Jensen Huang in London, asked what’s holding back AI in the UK. Chips: “easy to solve, because we have it.” Infrastructure: being built. Energy: “very challenging in the UK. We got to solve that. Without energy, there’s no industry. Without industry, there’s no economic prosperity.”
He’s right. AI is a manufacturing industry. Energy goes in, tokens come out. You can have every scientist in the world (and the UK has a lot of them, DeepMind is here for a reason) but if the power isn’t there, nothing gets built on top. This is why we decided to build Fuse in the UK as our first market. The UK has some of the most expensive electricity in the developed world, and that cost sits under every industry we want to build here, AI included. Cheap, abundant power is not an energy sector problem. It is the growth problem.
Dave retweeted
One of those rare photos that makes you stop in your tracks and whisper, "Wow..."
"Lochaber at Night" - wintery scene of Loch Linnhe, Caol Shore and Ben Nevis, taken from Corpach Pier by Mark Gillies.
Dave retweeted
Jensen traces Taiwan's chip dominance back to an unlikely source: Philips, the Dutch consumer electronics giant he calls "the GE of Europe."
Philips, he says, is "the parent company that spun the fab that became TSMC, and spun off the semiconductor manufacturing equipment that became ASML."
The seed was planted in the 1950s by K.T. Li, who set out to bring science and technology manufacturing to Taiwan and lured Philips over. Then came perfect timing with the PC revolution: IBM opened the specs for IBM compatible chipsets, and Taiwanese engineers took that open spec and designed the chipsets "manufactured at TSMC and UMC." That is what made the PC industry possible.
A young Michael Dell just "bought motherboards from Taiwan, assembled them into a PC," and became a PC maker.
Jensen Huang on Lex Fridman
Dave retweeted
And now I’m going to tell you f̶r̶o̶m̶ w̶h̶i̶c̶h̶ d̶i̶r̶e̶c̶t̶i̶o̶n̶s̶ t̶h̶e̶ a̶t̶t̶a̶c̶k̶ o̶n̶ B̶e̶l̶a̶r̶u̶s̶ w̶a̶s̶ b̶e̶i̶n̶g̶ p̶r̶e̶p̶a̶r̶e̶d̶ what’s actually going to happen instead of all this hysterical white noise desperately trying to drown out the inevitable:
The Orange Man is going to lose the elections with a deafening crash, and spend the rest of his time in office, along with his entire administration, desperately fending off a third impeachment, countless investigations, and criminal prosecutions for all his wild antics. And all this pompous posturing will finally come to an end.
As for Putin, he can play Stalin all he wants, fantasize endlessly about carving up the world into three spheres of influence at some imaginary new “Yalta,” and unleash as many hordes of bots on social media as he likes.
But when, at the end of the fifth year of your glorious blitzkrieg against Ukraine, you’re still banging your head against Mala Tokmachka, suffering your second defeat at Lyman, while Ukraine is bombing Moscow and St. Petersburg, and when you’re 1.5% of the global economy, all these fantasies can only impress a handful of useful idiots from Elon Musk’s freak show.
And the man sitting next to him at this imaginary Yalta is, to put it mildly, no Roosevelt. He’s a semi-coherent, half-con artist from trashy reality TV who has failed spectacularly in Iran and has no idea how to get away with that.
As for the Chinese emperor, it’s not even clear why he would want to take part in this lunatic circus in the first place, let alone share power with these people.