Human-Like Memory for LLMs
TL;DR I wrote a manifesto-style essay about a memory model for LLMs that is as close as possible to human memory and lets the system build a relationship history and long‑term memory of the user. If y…
Stories tagged “AI”
Indie hacking, startups, resilient systems - and staying sane while building a small company
TL;DR I wrote a manifesto-style essay about a memory model for LLMs that is as close as possible to human memory and lets the system build a relationship history and long‑term memory of the user. If y…
Two kinds of startups When you spin inside a certain field long enough, you start to forget that some concepts and terms are not obvious to people around you. That happened to me with the term 'startu…
All right, I’ll still write a bit about the 2025 results, but for now only about Fullyst. Some stats: 1. Subscriber growth for the Fullyst news channel: +716 people. And this is a channel with news on…
I crap-coded, sorry, vibe-coded – that is what they call it now. As an experiment to see whether a top Codex-like model can build something complete and working on top of an existing codebase, I decid…
I wonder how much longer we will pretend the world has not gone mad. On one side, a new era is coming, where AI is not just a loud acronym but a technology that is genuinely changing our usual ways of…
A mini update on Fullyst: Recently I tightened the limits on AI spam detection, added chat notifications when the limit is reached, and even introduced a way to gift a plan with higher limits to your …
I never liked writing email templates. They always broke my sense of beauty (the same way their responsiveness breaks in different clients): 1. Monstrous inline CSS 2. Lots of restrictions on which CS…
Improvy OÜ has received a grant in AssemblyAI’s startup program! 🎉 We’ve been awarded 200,000 hours of Speech-to-Text (worth roughly $30,000 (!)) with no expiration date. The quota is spli…
Web Summit 2025: First Impressions This is my third year in a row at Web Summit Lisbon, and this one feels especially unusual. It seems to me that Web Summit has really rethought how to make networkin…
Yesterday OpenAI introduced GPT-5, and the new model is already leading in most benchmarks. My subjective take after a few conversations with GPT-5: it is so good that I honestly cannot imagine how we…
Here is the funny part: I sometimes write about AI, I keep up with almost all the updates from major players like OpenAI and Anthropic, yet yesterday I was complaining about a task that was routine an…
A funny fact: no OpenAI model, in any format, is able to write a full-featured usage statistics handler for the endpoints at https://api.openai.com/v1/organization/usage/ Even if you feed it the entir…
Yesterday, along with the new models, OpenAI released Codex CLI, a full-fledged coding agent you can use right from the terminal. To me it looks like a perfect complement and, in some ways, a replacem…
A bit about AI The day before yesterday OpenAI introduced a new toolkit for building AI agents, and today I finally got around to reading the release notes and the description of what is actually ther…
About DeepSeek In the last few days the entire internet has been buzzing about a new OpenAI competitor, the Chinese company DeepSeek, which claims that its models are much more powerful than anything …