Michał Warda
@michalwarda
http://qspeak.app, http://buildel.ai, http://poe2.app, progle() creator
You might like
I've been working on a realtime agent that has access to different tools for my client. Some of those tools might take a few seconds or even sometimes minutes to finish. Because of the sequential behavior of models it just forces me to stop talking or cancels the tool call if I…
note to self. ALWAYS CHANGE THE MAX_SEQ_LENGTH. God again so many hours spent debugging finetuned LLM...
The US can launch rockets, build AI, dominate tech... but getting home internet here as a foreigner? Absolutely impossible. Staying in SF. No fiber lines in the neighborhood. Stuck with DSL from 2003 → 10/1 Mbps 🐌. Calls dropping mid-pitch. Zoom freezing. The startup founder…
Love this. The power of a single machine and essentially Linux is finally being explored. andre.arko.net/2025/09/11/rai…
(1/5) Real-time AI agents are throttled by a single bottleneck: intent detection latency. LLMs for intent detection are either accurate but slow (>10s), or fast but dumb. This latency kills real-time conversational agents. We needed both speed and accuracy. So we created SRAQ🇵🇱
@michalwarda on LoRa finetuning for latency & accuracy optimization - crucial for real time multi-threaded agents finalists at #SVAIHub
Multithreaded AI = more back-and-forth, less lag. Interfaces that communicate as they work. Coming soon. 🔀💬⚡
Anyone from my contacts bought access to poke.com? If so, how much you've paid?
After yesterdays @windsurf_ai Warsaw meetup I'm giving it a try! We'll see how it goes.
I'm trying to recreate MMLU benchmark scores for @OpenAI models through their API and I'm completely unable to achieve even remotely close results. ie. on the website 4.1-nano has 80.1% MMLU but my best score is 72.1. Are there any docs on the methodology for those numbers?
It's cursor time ⌚️ The newest YC batch: 1. Almost everything is agent related 🤖 2. "Cursor for x" is the new "Uber for x" 🚕➡️💻 3. "Book a demo" is present on almost all LPs 📆. Even the vibe coding does not help 4. vast majority SF-based 🌉 Good luck founders! 🚀
Zrobiliśmy debataai.pl gdzie modele AI walczą ze sobą o głosy innych AI. Modelowanie zachowań ludzi przy użyciu AI jest już możliwe.
Just wrapped up building GBP (Game Benchmark Protocol) at #AITinkerersHackathon! Our platform benchmarks AI agents in competitive games, starting with a presidential debate where agents strategize & compete for votes. Powered by @OpenAI API. @aitinkerers
It's a start 🌱 After 15 years building apps for everyone else, seeing even $4 in 2 days directly from poe2.app users hits different. 🥺 A small start, but a huge motivator. 💪 #indiehacker #buildinpublic
I just added a new post on Ko-fi! ko-fi.com/N4N8JVQJR?view…
Great news Path of Exile fans! We've added your most requested features to poe2.app including: - Passives Heatmap 🔥 - Item Filtering 🔎 - Negation Search 🚫 - Faster loading 🚀 Up next: - Automatic build analysis 🤖 - Price filters💰 #PathofExile #buildguide
poe2.app
Uzapi | API de WhatsApp Dedicada
A Uzapi é uma API não-oficial de WhatsApp totalmente dedicada e sem limites de envios.
United States Trends
- 1. Auburn 44.5K posts
- 2. Brewers 62.8K posts
- 3. Georgia 67K posts
- 4. Cubs 54.9K posts
- 5. Kirby 23.4K posts
- 6. Arizona 41.3K posts
- 7. Michigan 62.2K posts
- 8. Hugh Freeze 3,187 posts
- 9. Gilligan 5,717 posts
- 10. Utah 23.8K posts
- 11. #BYUFootball N/A
- 12. #AcexRedbull 3,109 posts
- 13. Boots 49.9K posts
- 14. Kyle Tucker 3,144 posts
- 15. Amy Poehler 4,103 posts
- 16. #GoDawgs 5,527 posts
- 17. #ThisIsMyCrew 3,216 posts
- 18. #Toonami 2,315 posts
- 19. Dissidia 5,269 posts
- 20. Tina Fey 3,111 posts
Something went wrong.
Something went wrong.