@Ahwhwh2740394 Really wanted a RC Helicopter as a kid. Never got one. Made myself a simulator with the very first public beta of Unity, then it was love...
@mantinband50375 @jahooma Up to some extent, the immense amount of data the subsidized inference brings can make up for some of those costs. Deepseek v4 with that ridiculously low price is probably gathering so many codebases to train on and analyse...
@bijanbowen@robinebers It is weird. I had to have deepseek v4 pro babysit long tasks with 3.5 flash, which is the first time for me where the orchestrator model is cheaper than the worker agent. With that setup i got the intended work done and done well and very fast, but... weird.
Midnight brain dump (with some screenshots) on the progress of my way too ambitious (but super fun to work on) project Synthasia.
A game and narrative engine that wants to tame LLMs and exploit their strong points.
#screenshotsaturday#indiedev#indiegame#gamedev#AI#LLM
@OfficialLoganK There is something really wrong with the price, effectively same cost of gpt 5.5 medium but way worst output on the same tasks. Cool at one-shotting some simple stuff but crumbles soon where other pull through. A weird model with some strong points but not worth the price.
๐จ The REMAIN AT YOUR DESK demo is live on Steam now!! ๐จ
Fake work by day.
Hack your corrupt company by night.
Hereโs the new trailer, link below ๐
#indiegame#cyberpunk#superdevday#trailertuesday
Need a tool for background removal in images for you or for your agents? Seek no more!
https://t.co/l0adE2SuLb
A little tool i have been using for some time now, made it for me to speed up some workflows and ended up being used by my agents (MCP).
It works, its fast, its free
All this said, it is a flash category model, maybe currently the best one at that, BUT safe for the exceptional speed, not astonishing. It's a worker model not an architect and gets easily lost on long implementations plans. A lot of the wow comes from how shitty the old one was.
Been coding, HARD, with brand new Google Gemini Flash model (3.2 or 3.5 however they are going to call it) for 24 h straight now. Some impressions :
- Lightyears better than previous flash model
- Fast, like scaring fast, like so fast it makes you doubt it actually did the job ๐งต
Oh, almost forgot, it is absurdly... Positive ? Don't know how to describe it, he is a cheerleader but also, it feels like it is constantly amazed by what it discovers. Had him perform a long task and it would have to connect to LMStudio, when he got the model list, he was in awe
@Cryptonoob17 Antigravity, still the old unusable for me flash in the cli. I would have big models prepare meticulous plans and then something as flash implement. BUT, even this new flash ain't a big thinking model, so you have to brake down the plans into small bits or it lies and drifts
Need a tool for background removal in images for you or for your agents? Seek no more!
https://t.co/l0adE2SuLb
A little tool i have been using for some time now, made it for me to speed up some workflows and ended up being used by my agents (MCP).
It works, its fast, its free
@AndrewCurran_ Been playing with the new Gemini flash model, and well... it is an extraordinary flash class model, and it can be so fast it feels weird. And it doesn't delete code randomly which is a huge plus. So for the pro to meaningfully better, it might crush the other SOTAs after all
My 4 years daughter is a vibe coder #gamedev!
Google playground, new Flash model, microphone active. And ... SHE WAS SO HAPPY ! Second prompt: "I don't want to lose"
(have a custom MCP based tool that my agents connect to for image background removal, will open source it soon)