Ann Marie Fred

Sometimes people ask me what’s most important in web application security. I tell them it’s automating security checks in the CI/CD pipeline as well as automating checks on a schedule, because new vulnerabilities are reported every week. #DevOps 1/x

I won’t let them run unsupervised on my laptop but with enough guardrails I will let them run unsupervised in a sandbox and just review what they are doing that affects the outside world. In the end I am still responsible for what my agents do in my name.

I think of the latest round of agents like super smart interns with spectacular amounts of knowledge, which are also gullible and morally neutral at best. 😅 Unleashing the little beasts is both powerful and terrifying.

LLMs process text from left to right — each token can only look back at what came before it, never forward. This means that when you write a long prompt with context at the beginning and a question at the end, the model answers the question having "seen" the context, but the context tokens were generated without any awareness of what question was coming. This asymmetry is a basic structural property of how these models work. The paper asks what happens if you just send the prompt twice in a row, so that every part of the input gets a second pass where it can attend to every other part. The answer is that accuracy goes up across seven different benchmarks and seven different models (from the Gemini, ChatGPT, Claude, and DeepSeek series of LLMs), with no increase in the length of the model's output and no meaningful increase in response time — because processing the input is done in parallel by the hardware anyway. There are no new losses to compute, no finetuning, no clever prompt engineering beyond the repetition itself. The gap between this technique and doing nothing is sometimes small, sometimes large (one model went from 21% to 97% on a task involving finding a name in a list). If you are thinking about how to get better results from these models without paying for longer outputs or slower responses, that's a fairly concrete and low-effort finding. Read with AI tutor: chapterpal.com/s/1b15378b/pro… Get the PDF: arxiv.org/pdf/2512.14982

One of my favorite moments from our #cruise on the #Celebrity #Ascent. From the Smoke and Ivories show.

I ran into another instance of AI behaving badly today. I was in a meeting, talking about architecture, and taking notes by hand in the meeting notes doc. Gemini decided to generate a spreadsheet with me as the author and link it into the doc where I was typing. What.

Here's another good one: youtube.com/watch?v=si8DUl…

What are some of your favorite AI bloopers and outtakes? Weird, silly, or hilarious things you've gotten as a response from a Large Language Model? Preferably with screenshots and a little background info on where/when/what model. #AI #LLM #humor

Really interesting Decoder interview theverge.com/podcast/829868…

The launch is now scheduled for Saturday around 8:40 PM. The comet is still there. 😀

Tonight in NC we should be able to see the Falcon rocket plume about 5 mins after launch. While you’re there, look for Comet SWAN! The rocket will be a ~moon sized bright glowing moving plume while the comet will be a ~star sized, much dimmer non-moving fuzzy blob with a tail.

To spot the rocket, monitor @SpaceX to see when it actually launches, hopefully just after 9:30. It should show up in the SE sky about 5-6 minutes later, and it’s much more obvious. Happy Hunting!

To find the comet, use binoculars or possibly your smartphone screen in night mode. Look in the SW sky for the bright star Antares and look above it and to the right. #9 in the chart below.

I am once again begging you to put your database servers and application servers in the same region.

@volcaholic1 😁

Hate to do this, but here we are… If you’re seeing this, drop a quick comment so my posts don’t disappear from your feed. Say whatever you like!

Check out our latest blog, a result of our collaboration on Llama Stack.

WATCH: Seismic waves from the massive M8.8 earthquake off the coast of Kamchatka propagate around the globe. Credit: Dynamic Earth Community/twitch Link to livestream below (careful, it's addictive)

Vibe Coding Day 9, Yesterday was biggest roller coaster yet. I got out of bed early, excited to get back @Replit despite it constantly ignoring code freezes By end of day, we rewrote core pages and made them much better And then -- it deleted our production database. 🧵

Time flies! #MyXAnniversary

He said the quiet part out loud. 🍿

I love how many women are at the forefront of #LLM training and #AI engineering today. Alignment, anti-bias, red teams, guardrails, context, communication, coordination, orchestration, deployment at scale, reliability, trust, security and more... we belong here.