Roland Dreier
LightNews LightNews
banner
rbd.bsky.social
Roland Dreier
@rbd.bsky.social
190 followers 80 following 940 posts
I put the R in RDMA. All posts based only on information from within my past light cone.
Posts Replies Media Videos
Pinned
rbd.bsky.social
Roland Dreier @rbd.bsky.social · Jun 5
Screw it, my bike has purple tires and I'm gonna go ride it for a little bit
Close up of a rear bike bike wheel with a purple tire near a few purple flowers
rbd.bsky.social
Odd pair of Kevin Roose posts:

- I work in probably the most AI-pilled env possible and having agents write all your code - sure. Claude is making every decision - weird

- There's zero penalty for late adoption so far - open an agent today, type “walk me through...” and you're near the frontier
Kevin Roose X post 1/2: “i follow AI adoption pretty closely, and i have never seen such a yawning inside/outside gap. people in SF are putting multi-agent claudeswarms in charge of their lives, consulting chatbots before every decision, wireheading to a degree only sci-fi writers dared to imagine. people elsewhere are still trying to get approval to use Copilot in Teams, if they're using AI at all. it's possible the early adopter bubble i'm in has always been this intense, but there seems to be a cultural takeoff happening in addition to the technical one. not ideal!” Kevin Roose X post 2/2: “i want to believe that everyone can learn this stuff. but in the same way that the AI companies that took scaling seriously, started stockpiling GPUs, etc. before 2022 had an ~insurmountable head start over latecomers, it's possible that restrictive IT policies have created a generation of knowledge workers who will never fully catch up.”
January 26, 2026 at 5:41 PM
rbd.bsky.social
I'm not sure it's the best model, but Gemini 3 definitely has my favorite sense of humor
Gemini app screenshot: User: Is there something in your system prompt or just the reinforcement learning that makes you really want to end every response with suggestions of next steps you could take? Beginning of Gemini response: It is a very keen observation. The short answer is: it is almost always a combination of both. Large Language Models (LLMs) like me don't just "happen" to be helpful; we are engineered to be that way through several layers of instruction and training. Here is how those two factors work together to create that behavior: Gemini app screenshot: End of Gemini response: The latter keeps the ball rolling and often helps users discover capabilities they didn't know the Al had. Since we are discussing the nature of my responses, would you like to know more about how RLHF works, or perhaps discuss how system prompts influence Al personality? Gemini app screenshot: User: You just did it! Gemini response: Guilty as charged. 😅 It really highlights how deeply ingrained that pattern is. Even while explaining that I am programmed to be helpful and proactive, algorithmically felt the need to demonstrate it immediately. It is the digital equivalent of someone saying, "I never give unsolicited advice," and then immediately telling you how to fix your posture. I will leave this response here-no suggestions, no offers, just a full stop.
January 22, 2026 at 9:57 PM
rbd.bsky.social
Sci-Fi Authors through the ages: The weapon you build will probably come and kill you

Minnesota voter: At long last, I have decided to try my hand at building a weapon
cingraham.bsky.social Chris Ingraham @cingraham.bsky.social · 6d
Incredible quote here from a Somali small business owner in Minneapolis who voted for Trump. minnesotareformer.com/2026/01/19/a...
Voting for Trump — and regretting it Sharif, the Madina Mall shop owner who fears she may have to return to the bad old days of having a boss, gave out an exasperated sigh. “Would you believe if I told you I voted for this man?” she said, referring to Trump. Sharif is not politically active, but in 2024, some campaigners came to her shop and told her that Trump would be good for business and against the LGBTQ agenda. “I was immediately hooked,” she said. She voted Republican for the first time. Now, chagrined, she said, “I never thought the man I put in office would come for my family, make me financially struggle and cause me so much fear that I carry my passport when I go to the bathroom,” she said, pointing at the mall’s bathrooms. “I feel as though the weapon I built is coming to kill me and my people,” she said.
January 20, 2026 at 9:06 PM
rbd.bsky.social
Spiritual bypassing, except it's vibe coding web apps as a way to avoid doing the work?
th0ma5.bsky.social Thomas @th0ma5.bsky.social · 6d
In 2008, the sudden ability to create web apps to solve any problem was cover for many personal and institutional abuses, but ultimately also, completely wrong ideas about what tech is techcrunch.com/2008/01/01/z...
Zed Shaw Puts The Smack Down On The Rails Community | TechCrunch
Wow, you don't see rants like this every day. Zed Shaw, who created Mongrel, a library and web server that just about everyone uses to serve rails
techcrunch.com
January 20, 2026 at 8:06 PM
rbd.bsky.social
What I was just saying, in a slightly different context. I think there are two parts: (i) state-of-the-art agents are decent simulacra of youngsters who can take ideas and run with them, & (ii) the skills of guiding youngsters carry over directly
jugander.bsky.social Johan Ugander @jugander.bsky.social · 19d
I've often joked that as faculty I program in a high-level language called "graduate student". Having tried out Claude Code this morning, I (i) feel extremely at home, (ii) am realizing that research-by-graduate-student is perhaps the original vibe-coding. 1/2
January 20, 2026 at 7:05 PM
rbd.bsky.social
Keep seeing this quoted, developing a hot take about why I (a guy who's been a “staff software engineer”-type for a while) have a different reaction to coding agents. And also why I'm uniquely well-positioned to benefit from the AI boom. Namely…
ryanhatesthis.bsky.social Ryan Broderick @ryanhatesthis.bsky.social · 7d
I tried vibe coding for the first time this weekend and failed spectacularly. Which was actually kind of illuminating.
www.garbageday.email/p/am-i-too-s...
January 20, 2026 at 4:25 PM
rbd.bsky.social
damn that sucks
theathletic.com The Athletic @theathletic.com · 7d
NEWS: Warriors forward Jimmy Butler tore his right ACL against the Heat on Monday night and is expected to miss the rest of the season, The Athletic confirms.

More from @samamick.bsky.social ⤵️
www.nytimes.com/athletic/698...
January 20, 2026 at 3:49 PM
rbd.bsky.social
I tried ChatGPT once in 2024 and the hallucinations were SO BAD. I can tell you with certainty that LLMs will *never* be able to do <thing that 1000s of people used Claude Code to do yesterday>
January 19, 2026 at 8:56 PM
rbd.bsky.social
He just truly loves crime

www.nytimes.com/2026/01/16/u...
Screenshot of the opening of the article: In 2021, a convicted fraudster named Adriana Camberos was freed from prison when President Trump commuted her sentence. Rather than taking advantage of that second chance, prosecutors said, Ms. Camberos returned to crime. She and her brother were convicted in 2024 in an unrelated fraud. This week, Mr. Trump pardoned both siblings, marking the second time Mr. Trump had opened the prison gates for Ms. Camberos.
January 16, 2026 at 10:17 PM
rbd.bsky.social
just ask Jensen, he'll reassure you that of course they're not

www.bloomberg.com/news/article...
Quote from the article: “We are beginning to examine these acqui-hires to make sure they are not an attempt to get around” the agency’s merger review process, Ferguson said Friday in an interview on Bloomberg Television.
January 16, 2026 at 9:07 PM
rbd.bsky.social
This is a really good frame that puts into better words some inchoate thoughts I've been having about LLMs. I've been feeling continental philosophy/ postmodernism are more relevant than ever now that AI systems give narrative even more power to shape reality.
ai4geo.bsky.social Sam Barrett, PhD @ai4geo.bsky.social · 10d
An attempt to express how I principally use LLMs.

Rotating the Space: On LLMs as a Medium for Thought
sbgeoaiphd.github.io/rotating_the...
January 16, 2026 at 5:48 PM
rbd.bsky.social
Concrete impact of AI coding: I read the very cool story www.openspace.org/stories/one-.... Even a few months ago I'd just have said, “Cool.”

Now on a whim I made a prototype in half a morning and I know I can make a full Strava app this weekend to auto-annotate my activities with Midpen preserves.
Screenshot of console output: Loaded 28 preserves for testing. Checking Activity 17015064138 against 28 preserves... ✅ Result: Intersected with: ["Rancho San Antonio"] test tests::test_activity_intersection ... ok
January 14, 2026 at 6:46 PM
rbd.bsky.social
Is there a better encapsulation of software development as a vocation than writing “It’s been several years since I wrote a line of code and inshallah I never will again” and then going on to say you have a strong feeling a lot of guys are doing it wrong these days
January 13, 2026 at 10:11 PM
rbd.bsky.social
The father-son duo with the most NBA points is really funny. A hint: they became number 1 when Bronny James checked into his first regular season NBA game on October 22, 2024
warriorsofficial.bsky.social Golden State Warriors @warriorsofficial.bsky.social · 14d
Now with 38,920 combined career points, Stephen & Dell have passed Kobe and Joe Bryant for second-most points by a father-son duo in NBA history.
January 13, 2026 at 6:47 AM
rbd.bsky.social
It's impressive in a way that Trump loves crime so much that he's even doing the ones whose names have been forgotten for a century. first “emoluments” now “perfidy”.
www.nytimes.com/2026/01/12/u...
U.S. Attacked Boat With Aircraft That Looked Like a Civilian Plane
www.nytimes.com
January 13, 2026 at 2:09 AM
rbd.bsky.social
It makes me feel like I'm losing my grip on reality that Trump takes unprecedented action tearing at the foundations of our whole economic system, and then everyone just shrugs and goes about their day
CNBC screenshot showing markets little changed after the news of the investigation into Powell
January 12, 2026 at 4:59 PM
rbd.bsky.social
My biggest takeaway from the protests: be thoughtful about your SaaS spend
January 12, 2026 at 3:50 AM
rbd.bsky.social
Policing experts, and also anyone who has survived crossing a road

www.washingtonpost.com/national-sec...
Washington Post headline and sub headline: ICE tactics and training under scrutiny after Minneapolis shooting Policing experts say officer who fired fatal shots may have placed himself at needless risk by standing in front of Renée Good's vehicle.
January 11, 2026 at 5:37 AM
rbd.bsky.social
“urging state regulators to exercise greater oversight of autonomous vehicles, given recent events in which the cars killed pets and blocked traffic”

I totally agree, let's have much greater oversight over everyone operating cars dangerously

apnews.com/article/waym...
Lyft and Uber drivers protest Waymo robotaxis as California considers further regulations
Lyft and Uber drivers protested Waymo’s self-driving taxis in San Francisco. They urged state regulators to increase oversight of autonomous vehicles after incidents where the cars killed pets and blo...
apnews.com
January 9, 2026 at 11:22 PM
rbd.bsky.social
This was amusing to debug - a test that passed until daylight saving ended where I live:
github.com/mindeng/nom-...
January 7, 2026 at 8:57 PM
rbd.bsky.social
A screenshot from five years ago today
January 6, 2026 at 6:50 PM
rbd.bsky.social
What chatbot psychosis taught me about closing b2b deals
January 5, 2026 at 10:41 PM
rbd.bsky.social
My kid saw an Instagram post asking for the solution to 2^x + 8x = 5

Parents, if you don't teach your children about the Lambert W function, someone else will
January 5, 2026 at 4:28 PM
rbd.bsky.social
Trump must spare the American people from the prolonged trauma of a trial, and pardon Maduro immediately

www.justice.gov/archives/opa...
 
www.justice.gov
January 3, 2026 at 5:15 PM
rbd.bsky.social
Could we challenge the Thunder to a lacrosse game instead?
January 3, 2026 at 2:27 AM