Does Temperature 0 Guarantee Deterministic LLM Outputs?
Large Language Models (LLMs) are often used in applications where consistency matters. Developers frequently assume that setting the temperature to zero will make the…
Read postAll LLM posts
Everything in this archive, newest first.
You’re Probably Using AI Reasoning Models Wrong
You might have tried one of those new AI reasoning models, maybe O1-Pro, and thought "this thing just doesn't get it." I've been there.…
Read postSilicon Valley Started in Garages. AI Could Too – With State-Funded GPUs
Something fascinating is happening in the AI landscape. While OpenAI chases commercialization with ChatGPT, Anthropic emphasizes their careful approach, and Google and Meta integrate…
Read postAI Isn’t Cheating
"No AI tools allowed." You've probably seen this phrase in job postings, workplace policies, journalist requests for comments, etc. Companies and organizations are trying…
Read postCursor is Secretly Running a Weaker Version of Claude 3.5 Sonnet
Last week I wrote about how Cursor is maxing out Anthropic's GPU capacity. Now it looks like they're giving us a worse version of…
Read postAI Agents in Enterprise IT: Why CIOs and IT Teams Disagree About Implementation
According to PagerDuty's 2025 State of Digital Operations Report reveals an interesting divide in how different levels of IT organizations view AI agents. While…
Read postCan AI-Generated Text be Reliably Detected?
AI detection tools don't work. And they're not going to work. Not because we haven't found the right algorithm yet, but because the entire…
Read postBuilding an AI-Friendly Codebase
This post explores practical strategies for making your codebase more AI-friendly. The goal isn't to constrain your architecture to AI limitations, but to find…
Read postPick Your Fights: A Pragmatic Approach to Coding with AI
Parents learn to "pick their fights," understanding that not every small disagreement warrants a battle. This wisdom translates remarkably well to the realm of…
Read postAI in 2025: Two Major Shifts
The ability to train competitive AI models at 1/45th the traditional cost is fundamentally changing who can develop AI systems and what they can…
Read postDeepSeek and the Effects of GPU Export Controls
Last week, DeepSeek unveiled their V3 model, trained on just 2,048 H800 GPUs - a fraction of the hardware used by OpenAI or Meta.…
Read postTrae AI IDE: ByteDance’s Answer to Cursor
ByteDance recently launched Trae, a new AI-powered IDE that's generating buzz in the developer community. As someone who has extensively analyzed various AI coding…
Read post