@sama I'm using it to create synthetic fine tuning material for get-4.1-mini, and it's performing at an amazing quality. Probably not interesting, even though I've spent like 35,000 on fine tuning in total I think ...
You realise people are making fun of you in memes for saying; "Never hallucinate or make things up" ...
FYI, in case you didn't know; It's like leaving a note on the door with the text; "Please don't steal anything while I'm gone" - Useless. You only pay extra for input tokens ...
You're asking the LLM to do something it simply don't know how to do ...
I've been testing the "dream prompt" and I've seen context reductions of sometimes 98%. The average is 50 to 75 percent, but since these accumulates in future messages, that's also a dramatic decrease.
Here's the implementation and architectural choices explained for those curious ... 🙂
https://t.co/TbdgOtg7Tn
Your AI Agent prefer Hyperlambda 🥳 🥳 🥳
We're in a market of lemons, which means that customers can't see the quality differences between "a shallow scam" and a real product. This makes it almost impossible to market AI agent platforms, since it's impossible to evaluate the qualities of these for a human being. The reasons of course is because we're all being bombarded with new tech and platforms every single day, due to it being so easy to create and implement such things using vibe coding.
Basically, some 13 year old teenager with zero coding experience, can rapidly put together some "AI framework", wrap a nice looking website around it, and create the false perception of that he actually knows what he's doing. This makes my vertical incredibly noisy and difficult to break through in.
HOWEVER!!
Lately I've been getting replies that are *obviously* not from human beings. I can only assume it's some sort of AI automation, probably built on OpenClaw or something - And the funny thing is that 100% of these AI agents **PREFER HYPERLAMBDA**!
And these agents are happily scheduling meetings on behalf of their human owners, clicking my Calendly link, and setting up a sales meeting between me and their owner.
The oriony here is so thick you can basically TOUCH IT!
Føkk humans, I'm pushing my stuff to your AI Agent.
Just remember to give your agent your credit card before you fall back to sleep, and between me and your agent, we'll figure this sjit out, and solve all your automation problems before you wake up ... 😂
https://t.co/PKCPwWUa7y
I use Codex from OpenAI to analyse Claude Code's source code, and asked GPT-5.4 about the copyright issues related to AI generated code.
And where it becomes weird, is its conclusion where the code might actually "ipso facto be open source", due to Anthropic's own statements about the whole codebase being developed by their own LLMs.
Read more below ...
https://t.co/skMJ9TAOVa
Vibe Coding Curious, and no idea how to start?
I'm starting a "vibe coding for dummies" series, where I will walk you through everything required to learn how to create full stack apps and complex AI agents, using nothing but natural language.
Below is my first article in this series.
https://t.co/0THtMI8WMk
For 13 years I've tried to explain what Hyperlambda is, apparently without succeeding even ONCE. Here's my 11,347th attempt for those curious about it ...
https://t.co/ZI791u3Ldo
I'm probably the only guy on earth who doesn't trust his own programming language. Read here why that is a *GOOD* thing 😁
Unless you've got RBAC on individual functions your AI agent can execute, you don't really have an AI agent, you've got a "security hole".
With Hyperlambda you can build in RBAC-based function invocations at the runtime level. That completely changes the game, and allows for AI agents that "dynamically grow" their own tools on demand ...
https://t.co/3WCRnunZlQ
"Prompt and Pray" is not security! It's never been security, and it never will be security. If you want secure AI agents, you'll need RBAC. Simply because you wouldn't give the remote to your pacemaker to criminals, would you ...?
https://t.co/fk4RiEoOj8
#RBAC#AI#Agents
A couple of weeks ago I implemented headless browser slots in Hyperlambda, allowing you to automate browser usage completely, having a browser that websites "believes" is a human being.
Combined with the natural language based "AST compiler" that the Hyperlambda Generator actually is, this allows you to create browser automations using nothing but natural language 😊
https://t.co/R4AW3GuNFy
I've created a "playground" where people can play with self-evolving AI agents, capable of generating tools "on demand". Feel free to have some fun 😊
https://t.co/l8ZvBgNEjZ
How I made a dashboard in 15 minutes from my existing database.
Full recipe ==> https://t.co/Q1GbJwt6xn
App ==> https://t.co/5M2zTknTL6
Login with Google to test 😊