Dad of five. Principal SWE, AI at Epic Games. My Apps: https://xitter.cf/t.co/XJ2dpY40NU, https://xitter.cf/t.co/nyK1a2Plx7, https://xitter.cf/t.co/90ZPuoJfUZ
Venice, Florida
Joined February 2008
- Tweets86
- Following58
- Followers768
- Likes1.1K
It's hilarious that every major lab's safeguard detection blocks your query for removing spyware from your TV ("hacking") but we have agents hacking bank accounts; hundreds of thousands of agents exploiting popular web frameworks, etc. tldr; Poor you? You can't even ask about deep seek flash obliterated in Claude without safeguards stopping it so good luck fighting back.
The next business school case study will be Delta’s decision not to select Starlink for inflight wifi.
What is fascinating to me is it’s an unforced error and they still didn’t self-correct even after significant public backlash.
They perfectly served up an opportunity for their competitors and now United is running an entire campaign around this.
At some point in the near future, the decline of listings on amazon.com will be something business school students are required to do a case study on. It's just amazing how fast it's turned into an herculean effort to wade through all the slop listings. The amount of junk on the first page of results is insane.
Explained to my kids about 9/11 today. I was in high school. I still remember the day very clearly including the fact we stood in a field while a helicopter flew over us to take a picture (unrelated to 9/11). I told them that despite actual evil, good prevailed. United we stand.
Despite its shortcomings, Astra is pretty slick. It loves to use python; it’s fairly meh at some coding tasks; otherwise amazing. Surprised.
Ultimate agy (antigravity) statusline: github.com/tweibley/agy-stat…
I’m running a small AI summer camp for kids / teens. If any company in the space wants to donate swag, or do a virtual speaking engagement / show and tell, or even donate services please let me now. RTs appreciated.
Https://srqaikids.com (hello@srqaikids.com)
I cannot overstate how big of a workflow hack this is for large new features/major new work... 1.) Draft plan, using Claude Opus, 2.) Have Gemini 3 flash review and suggest changes, 3.) Have Claude revise plan, 4.) Have Gemini review again. 5.) Tell Claude "JUST PRETEND (DO NOT ACTUALLY IMPLEMENT) you are going through the implementation phases of our <@theplan.md> plan. What issues do you spot?" 6.) Have it document the issues in a separate .md file. 7.) Have gemini 3 flash suggest solutions inline, where possible. 8.) Have Claude review and integrate. (Optionally, ask it again to simulate and revise.)
This probably explains why it works well. It’s really the repeating context and question…
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/pr…
Get the PDF: arxiv.org/pdf/2512.14982
Also before everyone tells me "that's too many turns with too many AIs" ... I'm talking about work spanning multiple codebases / repos, etc. Also Gemini 3 Flash is goated for implementing UI elements.
I talked with Matt about this role and it sounds very exciting: a strong team building the future with significant influence internally at Apple.
My team at Apple is hiring a software engineer! What might you be doing on this team? Building the future of software development at Apple. Prototyping cutting edge tooling with the latest advancements in AI. Becoming an expert in how Apple builds software
jobs.apple.com/en-us/details…
I was very, very, early to cloud computing and bet/built my career on it.
Candidly though, I view a lot of things in our industry with healthy skepticism. Lately, I’ve been skeptical about the actual benefits of AI beyond “another tool like a calculator” or fun image generator.
However, I believe, with conviction, there’s a whole new paradigm unfolding that will forever change our industry for the long term.
The new software development process is driven by anyone with an idea. Their qualifications: something they want to see on the screen, and persistence to keep prompting.
I’m calling this The Beginner Expert or maybe better coined as the Unskilled Software Prompter.
If you pay attention, there’s already breadcrumbs of this persona on social media – like the impressive 8 year old girl who built a Harry Potter AI web app in 45 minutes.
There’s no buying an O'Reilly or "Prag prog" book on the language you intend to use; learning about Docker, serverless or the latest Javascript framework.
Instead, the Unskilled Software Prompter begins with a loose description of the UI in V0 and then refines each component through chatting with Claude in Cursor.
It’s a loop that I think is best described as “Naive Software Development” – if you get an error just paste the error back into the prompt and let Claude figure it out. Seriously.
Though it sounds too good to be true, it just works. And when it doesn’t on the first try, Naive Software Developers simply ask Claude to help them understand the error, so they can resolve it and move on.
This is all working and available – today.
Anyone who has written code using a major framework uses code snippets, called packages (Ruby Gems, Node (JS) / Go Packages etc.), that make writing code a lot easier!
The unskilled software developer now uses a new kind of package: the prewritten prompt. For example one that offers authentication and user management. Or one for connecting to Stripe to accept payments.
There might be a prompt for getting files from Google Drive, data in BigQuery, or leads via the Salesforce API, etc. And since none of these are unique to any one enterprise business, someone will already have written a “good enough” prompt they can begin with.
The Naive Developer can take screenshots of their live project and include it back in their prompt, “I want the area circled in red to have rounded corners” and let AI do the work.
And when it’s good enough they just share the repl.it link with a coworker.
You may say, ”yeah dude! anyone can make a simple CRUD site, but companies have brand standards!” I say, don’t forget a prompt can include images, so maybe there will be style guides, samples of existing apps, etc. to accelerate seamless development and design.
The new Shadow IT is The Beginning Expert prompting away with his AI buddy.
Verifying myself: I am themcgruff on Keybase.io. F0Q_AnfF09JYg9yIjJT8IWpzBsuGnSy7zi6H / keybase.io/themcgruff/sigs/F…