Dan Rapp

As systems with single-node ephemeral engines over #S3 become mainstream (hello bauplan!), a new wave of "format aware" caching tools is inevitable. Our friends at University of Wisconsin-Madison and InfluxData just released LiquidCache, a cloud-native cache built on Apache DataFusion, a disaggregated cache based on logical data representation with incredible performances: go and add a star to Xiangpeng Hao repo now! While bauplan cache works without custom formats (it's specifically built to be native for Apache Iceberg and Apache Arrow), it shares many of the same ideas: extracting #S3 scans from queries, re-using entries on overlapping requests and more importantly, representing scans with *logical* filters, supporting relational algebra inference across languages. Our paper from last year with Ryan Curtin is available here (and perhaps some new papers are brewing as well, who knows 😎): https://lnkd.in/eCmedBgf! Congrats to our friends Xiangpeng Hao, Andrew Lamb and Remzi Arpaci-Dusseau for the incredible work: look forward to geeking out on caching the next time we all chat! #iceberg #s3 #arrow #cache

282

5 Comments

Resonance has joined a $25 million initiative led by Roadrunner Venture Studios to build a quantum campus and venture studio in Albuquerque’s Innovation District, aiming to accelerate quantum commercialization in New Mexico. The coalition brings together national labs, the University of New Mexico, startups such as QuEra Computing Inc., Maybell Quantum, and Qunnect, and investors including DCVC, Playground Global, and Quantonation to launch companies, attract capital, and strengthen the region’s leadership in quantum technology. Planned infrastructure includes a multi-node quantum network, testbeds, prototyping labs, and a dedicated venture studio, with measurable goals through 2026 such as creating startups, deploying private capital, and increasing institutional collaboration. https://lnkd.in/eZwwJcbX

11

✨ GPT-5 crossed a major threshold: over the last two months, we’ve heard repeated examples of scientists successfully directing GPT-5 to do novel research in math, physics, biology, CS, and more. If you have an example to share, please reply below! Prizes for the best ones 👇 To see what I mean by GPT-5 being directed to do novel research, see my original OpenAI for Science tweet thread here: https://lnkd.in/gTxs9P6C But don’t just believe me. Here's Terry Tao: https://lnkd.in/gB9hbPyc And here's Scott Aaronson: https://lnkd.in/gUqTgQqb There are many other prominent scientists who’ve shared privately. I’m NOT claiming GPT-5 is ready to prove the Riemann Hypothesis. It’s more at the “Lemma” stage today—when guided by an expert, it can do bounded chunks of novel science—things that maybe would have taken an professor or her postdoc a few days or a week to work through. But this IS the beginning of accelerating science. Because if each path takes a week, you can only explore so many of them. If it takes 20 minutes with ChatGPT Pro, and you can run them in parallel, suddenly you can explore far more. And remember, the model you’re using today is the worst it’ll ever be for the rest of your life. The idea that ChatGPT could do novel science sounded crazy a year ago, but here we are. Think where we’ll be in 3 months, 6 months, 12 months. If you’re a scientist of any kind doing novel research with ChatGPT and are willing to share your examples, I’d love to see them. ✨ I’ve got something special for the best ones, as judged by myself and my colleagues. ✨ Reply here with a short explanation and a ChatGPT share link, or DM if you prefer. Extra context is great: I'd love to know what worked and didn't, how you're using it successfully, and what you'd love to see from the next model from a scientific PoV. Excited for the replies! Lots more to come on this front. We're inspired by the successful examples we're seeing, and motivated by the ones that don't yet work.

626

77 Comments

Earlier this week I spoke with ACM@Mines at Colorado School of Mines. No slides. Instead: if you haven’t tried Tailscale, pull out your phone and install it now. Several people were online within a few minutes, which is the only metric that really matters. Networking is the part of computing where simple ideas go to get buried under NATs, firewalls, and the modern sport of “expose a port to the internet and see what happens.” Tailscale is the opposite approach: identity and encryption from the start, direct connections when possible, and, crucially, you only share with the people, services, and devices you actually intend to share with. The Internet doesn’t get a vote. We talked about how you keep a mesh network secure (keys + identity + explicit sharing turns out to be a pretty good recipe), why IPv4 vs IPv6 is still mostly an incentives problem, and a bit about startups: the boring secret remains “talk to real people, find a real problem, build the thing.” Thanks to ACM@Mines for having me. And to anyone who installed Tailscale during the talk: you’re halfway done. Add a second device. One node is a nice idea; two nodes is a network. #Tailscale #networking #security #computerscience

36

3 Comments

The Energy Department’s Pacific Northwest National Laboratory is pushing forward on its project to use better data and artificial intelligence to speed up infrastructure permitting. Last week, the team released an updated dataset of over 120,000 permitting documents, and they’re continuing to test AI tools to help users make sense of this data.

8

𝐒𝐭𝐨𝐩 𝐋𝐋𝐌 𝐇𝐚𝐥𝐥𝐮𝐜𝐢𝐧𝐚𝐭𝐢𝐨𝐧𝐬 1️⃣ Prompt Engineering: Craft precise instructions to guide responses. 2️⃣ Retrieval-Augmented Generation (RAG): Use external verified knowledge sources. 3️⃣ Constitutional AI: Embed explicit truthfulness rules during training. 4️⃣ Self-Consistency: Generate multiple responses and select the most coherent. 5️⃣ Chain-of-Thought Reasoning: Break down tasks into logical steps. 6️⃣ Fact-Checking: Cross-verify outputs with trusted references. 7️⃣ Model Calibration: Adjust confidence thresholds and temperature settings. 8️⃣ Knowledge Grounding: Define clear boundaries of the model's knowledge. 9️⃣ Fine-Tuning on Domain Data: Train with specific datasets for accuracy and alignment. Follow the AIKaDoctor (Free AI & Data Science Resources)  channel on WhatsApp: https://lnkd.in/dCTCEKKc Follow Dr. Habib Shaikh, PhD (AI) For more such content.

125

12 Comments

Gemini 3 benchmarks are impressive. But here’s what matters: Google built this on TPUs, not Nvidia chips. If they open commercial access at competitive pricing, the entire AI infrastructure market reshapes. Watch the hardware play, not just the model.

423

14 Comments

When I worked in HPC we used to describe computational capacity in FLOPS. More recently that has shifted to GigaWatts. And now people talk about it in $AUM. I find this shifting vocabulary supremely interesting and quite telling of the underlying drivers, scarcities and business models

58

6 Comments

A new article from Adrian Bridgwater introduces a new term, "SBOM flakiness." I think there's a kernel of truth here, which is the need to integrate SBOM data (and other metadata) into operational platforms for both AppSec and supply chain risk management teams. SBOM has always been necessary, but not sufficient. Its absence is a key indicator that something may not be right, but the presence alone doesn't guarantee that nothing is wrong. https://lnkd.in/eDNqinSA

47

8 Comments

AI is making it dramatically easier to contribute to Open Source. But that is not making life easier for Open Source maintainers. I wrote about the "asymmetric pressure" this creates, and how we might reduce maintainer burden without slowing innovation. https://lnkd.in/eA2vZfxE #Drupal #curl #OpenSource Daniel Stenberg

119

30 Comments