CTO of @InfluxDB (YC W13), founder of NYC Machine Learning, series editor for Addison Wesley's Data & Analytics, author of Service Oriented Design with Ruby.
New York City, NY
Joined February 2008
- Tweets6.7K
- Following1.1K
- Followers10.6K
- Likes1.1K
I was inspired to write a personal post about getting my start in tech. I can't imagine what it would be like if I had AI at my fingertips back then, but I'm certain I would have gotten farther, faster. pauldix.com/ai-is-the-greate…
Paul Dix retweeted
Up to 8x faster point queries with @ApacheDataFusio with the right tuning: datafusion.apache.org/blog/2…
Great stuff
I built bbtree, the foundation of databases and filesystems, and the fastest concurrent btree.
Code: github.com/XiangpengHao/bbtr…
Blog post & demo: bbtree.xiangpeng.systems/
It’s getting wild out there. Gotta have an always on agent security force.
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Agree with Parker on this. Not useful to juice your top line revenue by subsidizing tokens just so you can hit your next fundraise. You're setting up for a big fall later when it all comes due.
Many fell into this trap in the big data/infrastructure game. Subsidize the cloud infra costs (which scale linearly with customer usage, like tokens) to sign up customers and then find a little ways down the road that you're operating a low margin/negative margin business that you really can't just optimize your way out of.
I have a bit of lived experience with some parts of this.
Good follow up from Christina:
Hi Parker. You and I might disagree on other things, but I don't think we disagree on this (eg see this other excerpt from the same interview.)
Gross margin is good! We are in a strange era right now where investors encourage trading off gross margin for incremental growth; I suspect that will not last (when has it lasted, in the history of software investing?)
I found this to be an amazing read: anthropic.com/engineering/re….
It points to the future of software engineering. Anthropic has the unique position of an unlimited token budget with the best frontier models. I think the rest of us peasants will likely have this kind of capability within a reasonable budget by this time next year.
The trick is to create an SDLC that enables this kind of work.
Same, going to take this approach. And then send a swarm of agents to clean up the existing code base. Time to throw more AI at it to deslop the older, dumber AIs
We've released @InfluxDB 3.12! The big add is distributed compaction in beta in Enterprise. This allows operators to dynamically scale a cluster's compaction resources up and down in smaller increments with more VMs, rather than just scaling vertically.
We also made bulk Parquet imports roughly 3x faster by doing more work in parallel, added optional per-database schema enforcement, and made RBAC generally available.
influxdata.com/blog/whats-ne…
And it seems the post URL changed, so here it is again: influxdata.com/blog/influxdb…
LOL yesssss, this is the AI nonsense I'm here for. We can't slow this down, we're approaching the nonsense singularity. Full speed ahead!
Paul Dix retweeted
The contagion of fear
bcantrill.dtrace.org/2026/09…
I think the embedded evaluators formalized by government need some sort of name so we can refer to them.
Hmmmm.
Perhaps The Turing Police?
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
Called this three months ago: xitter.cf/pauldix/status/2067994…
Paul Dix retweeted
Boston’s @ApacheDataFusio meetup last week was a blast. Huge thanks to @andrewlamb1111 and to Pratibha Rana + @datadoghq for bringing the community together.
A special thank-you to @pauldix for years of visionary support for DF (and for believing in @Bauplan_labs!): review in 🧵
Looks like my "End of Programming" post got some coverage by @ThePrimeagen. I think it's fair coverage and I agree with his points. From the end of his video "technical expertise is going to be a gigantic differentiator".
I couldn't agree more. Agentic AI currently provides leverage for the humans operating them. The more knowledge and expertise you have, the greater the impact of that leverage.
A senior programmer with specific knowledge about a hard problem they're solving will run circles around a junior programmer (or non-programmer) vibe coding without any idea what's going on.
Although what constitutes hard will continue to be a moving target as the AIs get better at building all the things.
Here's the full video: youtube.com/watch?v=Eg5bygoM…
Here's my post if you haven't read it yet: pauldix.com/the-end-of-progr…
Oof, these code examples are rough. Maybe we still need to review that code, at least from Astra.
Astra is really, really cool but I cannot current trust it for my present day engineering. At least until I have adjusted. Some random thoughts including code samples of what I recovered from my traces. lucumr.pocoo.org/2026/9/7/as…
Hey @Cloudflare it looks like other people in your Discord are having the same problem. You broke your paying customer onboarding. You've charged me, but still have me on a free plan. I really would like to not dispute the CC charge.
My post: discord.com/channels/5953179…
@Cloudflare help, I've tried to sign up for a CF account and upgrade to paid, but can't. Your AI says I can't ask for human help because I'm not on a paid plan and I can't get on a paid plan because your flow broke and is now in some inconsistent state where I can't do anything to actually sign up to give you money. I have a valid payment method entered. No clue why this failed as it was on a new account signup and all I tried to do was upgrade to paid.