One of the biggest AI sceptics on Youtube (Mo Bitar) https://www.youtube.com/@atmoio got AI pilled.
https://x.com/atmoio/status/2106089460316397857/
Video is long so i can't fully upload it here.
TL:DR
He used Claude Opus 5.5 on UltraCode and it built him flawlessly working app with 2700 lines of spec in 7 hours and it was a turning point for him. Just 6 months ago he was fully anti AI and said AI is useless slop.
Yellowstone national park.
Huge gray wolf and Grizzly bear trying to share a Bison carcass
Basically, nothing new. AI is FORCE AMPLIFIER. If you were already good, it makes you good.
If you sucked before, your AI output will still suck.
This guy was Anti AI when i first started watching his streams, he coded Linux drivers by hand using C.
GPT 6 / "Astra" is released next week. Leo is 100% reliable leaker and has insider info on model releases, gets it right most of the time!
https://x.com/synthwavedd/status/2085365276640702915
Basically new model will be BIG, so called "Mythos/Fable" level pre-train. Current GPT models are very good and almost 'Fable level', but they reach this with amazing post-training. With new pretrain model will be "smarter" by default like Fable and great for interactions on architecture and discussions, big picture analysis, driving GPT 5.6 models for coding, etc..
And with OpenAI's capacity it might also be suitable for coding, unlike Fable, which is limited.
Its over they are using AI for systems C code in FFMPEG we are cooked
its now slop
Source: https://x.com/synthwavedd/status/2074886230018568582
Leo is very reliable AI insider. His information is usually mostly correct, not LARPer. You guys can double check all his info and make sure.
So basically we are getting GPT-6 next month, new pretrain and "big model".
I switched to CODEX and GPT back in October because it was not lazy, it did not take shortcuts, it did not hallucinate as much as Claude and was my workhorse. Before it was mainstream even. 5.2 XHIGH was my goat.
Last days though it acts like Claude and i am very disappointed. Instead of doing work it takes shortcuts, implements half-assed features, documents everything as "blocker" instead of adding missing pieces, even when i directly tell it to adhere to instructions and implement missing parts.
When i tell it that missing parts must be added instead of declared blocker, it then says oooh right you are absolutely right. It acts like Claude. Has been like this for last several days.
Please, OpenAI. Don't turn it into Claude. I beg you.
P.s
Rate limits on other hand are very good for me. Spamming XHIGH/HIGH on fast mode and i cant reach rate limits even on big ass project for few days now
Hi there guys. I was experimenting around for few weeks and made this new TS/JS runtime.
It is not production ready yet. But you can play around and see how it is different from Node, Bun and Deno.
https://github.com/RtlZeroMemory/Sloppy
Here is short TL:DR
Sloppy is trying to bring a more serious, batteries included development experience to the JavaScript ecosystem, closer to what developers expect from .NET or Java/Spring.
It is not trying to be another Node/Bun/Deno clone. The idea is different: runtime + framework + compiler-aware tooling, built together.
Sloppy includes, or is actively building toward:
* pre-compilation and static runtime metadata
* compile-time checks for routes, contracts, ORM models, and framework metadata
* HTTP routing and request/response handling
* OpenAPI/contracts
* authentication and authorization
* logging, health checks, metrics, and management endpoints
* first-party ORM and migrations
* background jobs / scheduler
* cache and output cache
* WebSocket / realtime primitives
* WebHooks
* HTTP client factory with resilience policies
* Streaming API's
* Cursor based data streaming from database
* Redis integration
* TestHost and TestServices for serious integration testing
Ask any questions and clarifications in comments, i will gladly answer.
This project will keep growing/improving and being maintained.
Some feedback would be appreciated