24, sde @ stealth startup, performative nerd
hyderabad
- Tweets 11,665
- Following 907
- Followers 4,346
- Likes 273,218
I tried the First Contact demo for the first time on my Quest 3 last year and it's really sad that basically nothing has the level of polish that older studios under Meta did, what a polished demo, that aged so well that I found no flaws on my Quest 3, almost 10 years later.
The strongest “I’m living in the future” moment was the First Contact demo on the Oculus CV1 in ~2016 and I’ve been chasing it ever since.
Yeah yeah we’ve got AI and self driving cars and stuff now, but it doesn’t hit the same.
Shame that the space kind of fizzled out…but here’s to hoping there’s a resurgence soon.
There’s got to be other technologies that had an extreme “wow future” moment that never panned out. What would your “oculus” moment be in the 90s, 80s, 70s?
8
1,074
ruddr is a tool that lets your Codex or Claude Code spin up other instances of any model harness with any model, letting you set up model-agnostic workflows from any harness
write your code with Claude Code, have it reviewed by Codex, just tell your agent to use ruddr CLI
ruddr also has a pretty cool 'session viewer' for all your agents, (Factory Droid, Pi, Codex, Claude Code, OpenCode are all supported) so you can see all your sessions and their diffs and what changes were made all in one place LIVE! (EXTREMELY fast btw)
I was tired of opening my IDE to view session-specific git diffs, it is now way easier, I can ask Opus 5.5 to spin up a GPT-6.1 Sol subagent using ruddr to review it's work, and your agent can even steer the agents spun up by ruddr and look at their progress live and provide additional context at anytime, and so can you with the ruddr TUI at anytime, including a live diff for every session, using ruddr or not :D
Claude Opus 5.5 rewrote this entire project from Go + Bun to entirely in Rust in about 2 hours (besides the web dashboard which is TypeScript) with 5 subagents and one final review agent, before using ruddr itself to review it's work using GPT-6.1 Sol, below are the stats:
TUI memory: 119 MB → 7 MB
npm install: 512 MB, 174 packages → 18 MB, 1 package
runtime: needed Bun → none
binary: 6.5 MB → 18 MB (now includes the TUI, web dashboard, and agent adapters, MORE features than before)
11
1
31
3,226
Google AI Pro will get Deep Think soon!
🚨BREAKING: Google is about to make Gemini’s best models much more exclusive.
Starting October 9, free users lose both Flash and Pro and get Flash-Lite only. AI Plus users lose Pro too. Only AI Pro and Ultra keep access to all three model tiers.
And finally: Deep Think will also be available to AI Pro users. Until now, it has been exclusive to AI Ultra subscribers.
4
129
6,705
Reminder that Google can infact wipe out marketshare whenever they want to, they just choose not to
Now that generally capable models are coming to Antigravity that are known for efficiency and quality, and Google can collect data on them, I think Google could make a comeback
🚨Antigravity is getting Claude Opus 5.5 and Sonnet 5.5 on Pro and Ultra plans !!
GPT-OSS 120B is getting removed(finally)
This might be a hint that Gemini 4 Argon's rollout will be happening pretty soon..
Though I don't know why does it say that it won't be available on my plan as I'm already on Pro.. can some clarification be provided here? @_mohansolo @rodydavis .. thanks!!
6
77
4,106
cheaty retweeted
Here you go! The Elder Scrolls III: Morrowind running natively on PS5.
1
3
23
1,628
how long until someone runs this on GPT-6.1 Sol/6 Astra/Opus 5.5 and finds out it's already saturated/generalized?
Can LMs discover & fix bugs if you don't tell them what went wrong? To proactively maintain a repo, agents need to find problems before users do.
SWE-sweep benchmarks this on 100 repos, 22 languages, 4k real bugs (numpy, php interpreter, lean kernel). Top models get <5%.
6
1
71
7,388
that's right, AI detectors also go in the LLM transformer hole
How well do AI Detectors work? And are general-purpose language models catching up?
Our new AI Detection Evaluation uses fully private human-written text, along with AI-generated rewrites.
We find that the age of specialized AI detectors like Pangram may be coming to an end.
4
54
2,478
Gemini 4 Neon would go so hard
@GoogleDeepMind take notes
Gemini 4 Helium (Flash-lite)
Gemini 4 Neon (Flash)
Gemini 4 Argon (Pro)
Gemini 4 Krypton (Ultra)
You heard it here first!
4
55
3,631
i am luna level speeds #proud
2
14
1,099
FLUX 3 Image is out!
It supports bounding box JSON image generation just like Ideogram, and there will be an open-weight variant coming out soon :)
Introducing FLUX 3 Image.
Control every pixel.
Make precise multi-turn edits without changing any other pixel.
Lay out the image exactly how you want using bounding boxes.
Generate in up to 4K to preserve details.
Use up to 10 references to compose an image.
Commercial Weights available for companies running image generation at scale.
Open Weights version of FLUX 3 Image is launching in the coming weeks.
17
1,771
life is actually really fucking boring once you've automated it all away
i should be grateful i learnt this lesson at only 24 years old
i don't want to imagine what this is going to feel like when im 30 years old
im going to read a book for starters
4
14
853
