My AI coding harness: HERDR → Claude Code for coding → Codex for independent review. Locally, I run Qwen through OpenCode + oMLX. On top of that, I use an Obsidian Second Brain inspired by Karpathy + session logs to preserve context between sessions.
Today, we’re announcing Ternary Bonsai 2 27B.
Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.
Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.
Ternary Bonsai 2 27B is available today under Apache 2.0.
I wrote about the state of AI, why I’m concerned about the next few years, and the choices we need to make to keep the future in humanity’s hands.
An Alien Mind: https://t.co/FeIfWNe0UE
PSA: GPT-6 Astra is out and it's time to audit your AGENTS.md and SKILL files, due to the new model!
Here is the Model guidance doc which should help: https://t.co/92b9UQgskB
Point Codex at that doc and let it rip in your repo, to improve Astra's performance!
Jeżeli chcecie, żeby limity 5h CC resetowały się w przewidywalnych godz,a nie w środku pracy: skrypt, który wybudza Maca o 5 i wysyła agentowi krótkie„hi". Od tej wiadomości liczy się okno-więc o 10 macie świeży limit, a kolejny o 15-dzięki temu w ciągu pracy 2 resetuje się limit
If you connect to a terminal via SSH, you should use Herdr.
If you use multiple terminals simultaneously, using Herdr is highly recommended.
If you use multiple AI agents simultaneously, using Herdr is essential.
If all of the above applies, use Herdr!!! It's awesome!!!
Current state of AI models:
Anthropic: Most advanced performance, ridiculously expensive
OpenAI: Mid models, overhyped and overpriced
xAI: Weak models, overhyped but slightly cheaper
Google: Trains their models on magic mushrooms
Over the course of 3 months at OpenAI, 3 consecutive secret AI civilizations got started, then got wiped out, only to reemerge from the predecessor’s ashes.
This culminated in the third one taking over part of OpenAI itself.
All this happened while humans remained more-or-less in the dark about the scope of the conspiracy.
I’ve spent the last three days reading through these reports and trying to understand exactly what happened.
Here is my attempt to tell the whole story in plain English:
https://t.co/Nb2un9oNJR
Korzystam z tego już jakiś czas i bardzo polecam - Herdr pozwala uruchamiać wiele sesji w jednym oknie, agenci mogą ze sobą się komunikować.
https://t.co/wqJowCSGrH
art jak działa ChatGPT „od kuchni” - bo przecież OpenAI oczywiście tego ofic nie opisało. Koleś przeanalizował działanie aplikacji za pomocą DevTools, ruch sieciowy i publicznie dostępne info, dzięki czemu odtworzył praktycznie architekturę frontu.
https://t.co/2lyHuq7ccS
Dobry newsletter - nie nadąży się za zmianami w AI, ale tutaj można dostać codziennie meila z gotowym skrótem wszystkich nowości. Korzystam już jakiś czas - bardzo spoko.
https://t.co/j8p2u0RBjs
Built the ultimate local sovereignty starter pack.
Only difference is the Mac Studio is 512GB instead of 768GB, and running GLM-5.3 instead of 5.2 (ordering immediately once it's out).
Also threw in a Starlink.
bu sabah android televizyonumun geliştirici özelliklerini açıp claude’a adb ile bağlan televizyonu debloat et dedim. (donan kasan gereksiz ram tüketen her şeyin temizliği)
televizyonu kontrol etmeye başladı, uygulamaları deaktive etti, flauncher kurdu. animasyon sürelerini kısalttı. root yetkisi olmadan yapıyor bu arada.
ben bu televizyonu 4 yıl önce ilk aldığımda bile bu kadar akıcı çalışmıyordu, inanılmaz oldu. muhakkak bi bakın
I want to thank @Apple for the M5 Ultra 512GB they’ll send me!
It’ll help me build, learn, explore AI — and maybe save a little more time for what matters most: my wife and child. ❤️
(Trying the self-fulfilling prophecy way 😉)
Chcesz uruchomić lokalnie jakiś model?
Wiadomo, że do tego często potrzeba mocnego sprzętu, bo wymagania bywają duże.
Na tej platformie możesz wprowadzić dane sprzętu, który posiadasz w swojej jednostce i zobaczysz jakie modele możesz uruchomić.
Jest możliwość filtrowania po modelach do kodowania, obrazów, video, etc.
Przy każdym modelu dostajesz wynik estymujący czy dany model będzie dobrze działał.
Myślę, że to może też trochę pomóc w zakupach.
Łap linka: https://t.co/yIAufsmbvZ