Obligatory: the views and opinions expressed in this post are my own and do not represent the views and opinions of my employer. In light of all the hype going around about ChatGPT, I wanted to offer my “hot take” on what the next 2-5 years
We evaluate three leading AI models—Claude Sonnet 4.5, Gemini 2.5 Pro, and GPT-5—on their ability to solve Google reCAPTCHA v2 challenges and find significant performance differences, with success rates ranging from 28% to 60%.
Testing Andrew Ng's document extraction service reveals over 50% hallucinated values in financial tables. | Pulse AI
Before we go any further: I hate to ask you to do this, but I need your help — I'm up for this year's Webbys for the best business podcast award. I know it's a pain in the ass, but can you sign up and vote for Better
The workflow that's changing how software gets built.
Frameworks for LLM Agents have been surprisingly disappointing. I want to offer some principles for building agents based on our own trial & error, and…
I’ve been sitting on this draft for a few days now, partly because I thought it would turn down the bitterness, and partly because I kept asking myself ...
Random tech notes
Let me get one thing out of the way immediately: LLMs are helpful. This isn't about whether LLMs can write code. They can. It's about why vibe coding might be the worst career investment you'll make.
Why clear communication—not clever prompts—is your real AI superpower
While my final ranking in the Google AI Contest was disappointing ( 280th ), it was a very educational experience and was totally offset by Gábor Melis' dominating win using Common Lisp as well.
From silly cat costumes to world-changing innovations, OpenAI's latest release marks the beginning of something extraordinary. The fascinating world of visual reasoning is emerging, where AI models will soon think in pictures and solve complex spatial puzzles, transforming how machines understand and interact with the…
How local, learnable routers can reduce token overhead, lower costs, and bring structure back to agentic workflows.
Today we’re sharing a human rights impact assessment on AI/ML to help Wikimedia understand effects on human rights.
We introduce and open-source an automated evaluation for political even-handedness, testing six models across 1,350 prompt pairs.
We’re here with a couple save the dates to reserve some space on your calendars.
New discovery underscores security implications of AI-powered development and the rise of Vibe Coding Platforms.
Delegate investigative work to AI agents, from L1 to L3 reviews, with consistency, auditability and traceability.
Personal blog of Stephen Diehl - Software engineer writing about technology, programming, and the future
QWEN CHAT GITHUB HUGGING FACE MODELSCOPE DISCORD Introduction Last December, we launched QVQ-72B-Preview as an exploratory model, but it had many issues. Today, we are officially releasing the first version of QVQ-Max, our visual reasoning model. This model can not only “understand” the content in images and videos…
I recently needed to use a microcontroller to estimate the pose (translation and orientation) of an object using readings from six different sensors. Since the readings were non-linear and coupled with each other, an explicit analytical solution was out of the question.
AI won’t crash—it will burn. Like every tech cycle, the fire will clear the brush, redistribute talent, and leave infrastructure to power what comes next. The question is: what kind of plant are you?
Vibe coding (or vibeware) is making rounds on X now. To the best of my knowledge Andrej Karpathy started the “meme” in this X entry . I find it well written and hilarious and it seems to have taken off.
AMD is sending us the two MI300X boxes we asked for. They are in the mail.
Most browser agents forget what they’ve done. This one remembers, adapts, and learns, redefining how AI agents get work done.
And implications for RLVR progress
My life is not a math Olympiad
Success! There is machine learning happening on my Apple ][+.
