Matt Sayar

Dox with Grok After learning that researchers can de-anonymize users by pointing LLMs to datasets, I wondered if they can do it with just prompts. I happen to have an account I post with that I don't generally want...

You're not anonymous

mattsayar.com/dox-with-grok/

I accidentally broke my website. Claude fixed it in minutes For a few weeks there, my website was missing articles. It's not a huge deal, but it broke in a very silent, sneaky way. Fixing it by hand would have taken several hours of concentrated effort, but I ...

I accidentally broke my website, but used Claude Cowork to fix it in minutes. So happy with the amount of time I saved.

mattsayar.com/i-accidental...

Saturday Morning Breakfast Cereal - Never Saturday Morning Breakfast Cereal - Never

Haha you're so accurate with your predictions! So what happens five years from now? Lemme check today's comic

www.smbc-comics.com/comic/never-2

....

I wonder what's happening between now and October that will make this API available again

Spotify's Page Match executes of the first idea I ever had for LLMs I started reading Seveneves years ago. Halfway through, I put it down and read something else. A couple years later, I picked it up again. Unfortunately, I didn't remember anything from before, and I ...

I like Spotify's new audiobook feature mattsayar.com/spotifys-pag...

Orchestrating AI Agents for Job Searching I got tired of all the manual processes involved with job searching and wrote a Claude Cowork plugin to automate the boring parts. My typical search process looks like this: Working with my career coa...

Been using Claude pretty extensively to job search

mattsayar.com/orchestratin...

ChatGPT made me strong The fitness wars with AI have escalated. It started four months ago when OpenAI put out an ad showing how you can use ChatGPT for fitness. Anthropic recently responded by making fun of OpenAI for putt...

I recently wrote about how I used ChatGPT to get fit

mattsayar.com/chatgpt-made...

Tankless water heaters are basically magic I feel like I understand computers pretty well, but when I do some maintenance on my tankless water heater, this is the best understanding I have: My tankless water heater is a Navien NPE-240A natural...

I wrote this for myself for the next time I had to clean my water heater and it was SO HELPFUL. Easily cut my active time by half.

mattsayar.com/tankless-wat...

I agree it's good, but I think we will be a minority opinion on a long enough timeline. Eventually everyone could conform to it, and people may even prefer it

How many prompts am I allowed to use now that the LLM saved me some energy?

Let AI catalog your house for insurance If my house ever burned down, I like to think I would get over it pretty quickly. It's just a box holding a bunch of my stuff, right? Insurance will rebuild it, and I'll fill it with stuff again. Over time, it's become less of…

I took a video of my house and made Gemini put it all in a big list for me

https://mattsayar.com/let-ai-catalog-your-house-for-insurance/

An AI summary about an AI spokesman banning AI

I'll be driving home from skiing while the @denver-broncos.bsky.social are playing Saturday. I don't get consistent cell signal through the mountains so radio is my best bet, but I don't know which radio stations carry the game. What a perfect use case to make AI look it up!

Calendar on wall

Ok I'm using the bird calendar

Calendar on wall

Ok I'm using the bird calendar

Just copy and paste this with a 2026 timestamp and it's just as relevant today

Good to know! It used to be helpful for earlier LLMs but it's nice to see that models don't need those kinds of prompts anymore

I'm curious how the risk factor on my smartwatch is higher than my physical card.

Just tried to pay for a large purchase with tap to pay on my smartwatch. Declined. Used the same card from my wallet and the transaction went through. The smartwatch is multifactor auth (something you have, something you know (to unlock it)) while the card is only single factor auth.

Book called the Quick Guide to Prompt Engineering

This is a THICK book but reading the word "Bard" on the cover immediately makes it feel dated.

What would be the most effective argument against it? I can understand someone selfishly enjoying the product vs an abstract threat against people you don't know. I don't agree, but I understand.

Found a dark pattern while trying to cancel LinkedIn Premium on my mobile browser. I actually planned to do it later before I remembered one checkbox. Really annoying

— -Wing

But then how will everyone know how big of a nerd you are??

I do it because I don't know where else to put my stickers

Or just own the reputation as being a terrible speller. Then everyone will be confused when your non-dictated messages are typo-free. And then you'll be accused of writing with ChatGPT

Gemini 3 feels like a big next step Before vibe-coding had a name, I made a 'game' to help my kiddo learn left from right, and then I was able to drastically improve it with better LLM models. And then today, Google reveals Gemini 3 whi...

It used to take a couple hours of iteration to make a simple game with older models, and now Gemini 3 is doing it in one prompt. The pace of acceleration is dizzying

mattsayar.com/gemini-3-rea...

At AI's current level of capability, none of these companies are trusting AI with high-risk automated config changes. Today's outage appears to be caused by a file that got too big. There's still plenty of ways for humans to mess things up!

I have no idea what you're talking about!