Sean Uddin (@Metafintek) header
Sean Uddin (@Metafintek) profile picture

Serial Founder | Builder | ENTP Author of "Architects: Building the Exit to Techno-feudalism" 0-1 Systems Practical Harness Engineer https://nitter.gq/t.co/DYskegP1I6

Metaverse
Joined January 2022
Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)Photo by Sean Uddin (@Metafintek)
Replying to @Metafintek
AI’s future is not predetermined. It will be shaped by those who design its collaborative incentivised co-ownership models now. That is the real battleground. 📘 Architects: Building the Exit from Techno-Feudalism amazon.co.uk/dp/B0GFXPP9Y6 #ArchitectsBook #Technofeudalism #Decentralisation #AI #davos
1
2
1
9
1,234
"Checked, and the answer is worse than "wrong door". The talking-head path was designed and never built. The design note is dated 20 Nov, 21:30 SGT, written by me from your ruling that evening: start from podcast audio, a script, or both. Its status line reads "DESIGN proposal, nothing built". It sat as wave 5 of the fix plan. I then spent that night on six review rounds of a picture-ownership rebuild you never asked for, and wave 5 never started. So tonight I pushed speech through the music video planner because the correct lane does not exist, and I did not tell you that when you said "make the videos". I should have said it at 21:13." it snow 1. am,. ever since fable 5.2 released and you somehow nerfed opus 5.5 you have wasted my fucking time, time and time again. @AnthropicAI 1.5 trillion my arse. how do i get these tokens back??
14
Another "holo video". I think I have perfected it now. I can take a selfie video and it motion-captures all the movements and maps it onto my holo-renderer. Pretty neat. Made with @googleaidevs Gemini Neon and @Scenario_gg.
9
1
31
1,064
Same idea, made locally. @everywearos 's AI Director, on one PC: Left: from a one-line brief. Right: from @MayorKingAI's 19-shot prompt. Score composed in Gener8. No cloud. everywear.id #OneClickCreate #OwnTheMeansOfCreation
Asked Claude to recreate the entire history of the world in 30 seconds It wrote me a 19-shot prompt. i ran it through seedance 2.5 in 4k on @dreamina_ai This is what came out Prompt below
1
2
38
Sean Uddin retweeted
WARDOGS 0.31 makes the Verba $600, cuts $500 off the PKM and drops .50 Cal ammo to $200. The T-72B3 now unlocks at level 27 and the Gepard at 35. All seven price cuts and four unlock changes, plus GhostGaming’s breakdown. #WARDOGS wingmait.xyz/wardogs/guides/…
2
3
98
They fucked claude opus 5.5. Useless for over night runs, with all permissions given, then just invents a new permission or condition and stops costs 8 hours of productivity.
1
2
28
So now a small 300MB adapter makes your local 27B coding model dramatically better! Hugging Face released an RL-trained LoRA adapter for Qwen3.8-27B, specifically optimized for OpenCode. And the reported results are impressive! SWE-bench Pro (hard 51 tasks): 🔴 Original → 35.3% 🟢 With LoRA → 51.0% 👈 Terminal-Bench 2.1: 🔴 Original → 46.4% 🟢 With LoRA → 52.4% 👈 Median agent turns on tasks both solved: 🐌 Original → 48 turns 🚀 With LoRA → 13 turns 👈 That's a 15.7% improvement on the selected SWE-bench Pro tasks and 73% fewer agent turns on shared solves. With thinking disabled, the adapted model reportedly approaches or exceeds the original model running at maximum reasoning effort, using 6–8× fewer generated tokens per task in the team's comparison. The best part? 💾 Adapter → ~300MB 🧠 Base model → Qwen3.8-27B 🔓 License → Apache 2.0 💻 Local inference → llama.cpp 🛠️ Coding harness → OpenCode 📦 GGUF adapter available Just a small adapter teaching an existing model to use coding tools more effectively. Video source: @HuggingFace
72
170
18
2,076
128,364
Shes not wrong.
This woman is intelligently funny 😂😂
6
Intriguing
Introducing: Self-Improving Memory 🧠✨📈 We let the model discover its own memory skills, then train them into its weights: - Meta-optimizes memory from the model's own runs with large gains. - Distilling these gains works too. Blog post: 📝 facebookresearch.github.io/R… 🧵1/6
30
Excited to share our new Behind the Scenes mini-doc today, showing the huge human effort behind Gods Don't Give Gifts. Our special early preview screenings with Director Q&As on 12/4 (LA) and 12/5 (SF) sold out instantly—with only a few tickets left for NYC (12/10), we have big news for everyone else: Due to incredible global demand, we’ve partnered with our international distributor AGC to secure a simultaneous WORLDWIDE THEATRICAL RELEASE beginning JANUARY 22 to ensure fans everywhere can see it on the big screen!
93
134
47
1,113
73,183
Not wrong.
dear 8gb, 12gb, 16gb and 24gb vram owners, i know people in the local ai space might discourage you, saying don't use q4 and q4 kv, it trades away quality and sucks and all that, but hear me out loud, i'm going to be very clear the ones saying that own datacenter grade chips and can afford full precision, and as a small gpu owner you don't have a choice i'm out here in the field running everything from 8gb cards to 256gb of unified memory, and i know what i'm saying don't get sold, try it and find out, it's not that bad at all
8
GM everyone! Time to bring Jensen back!!!
2
2
2
25
1,140
Sean Uddin retweeted
Took a bit longer due to getting flooded! Monsoon Season!
1
1
23
How quaint Google catches up to me :D I published MyMory’s sleep/dream-cycle design in May 2025, including collation, human approval and LoRA training. Google’s SLEEP research explores the same broad principle through a different, experimentally evaluated implementation: AI needs time to consolidate what it encounters. github.com/BAIS1C/mymory/blo…
It’s happening. Google built an AI that rewrite its own weights to get smarter.. and it does it by sleeping. Right now, every AI model you use has a fatal flaw. When you train a model on new data, it suffers from "catastrophic forgetting." It forcefully overwrites its old knowledge to make room for the new. If you try to bypass this by using a massive context window, the AI simply forgets everything the second you close the chat. Once a model is deployed, its brain is effectively frozen in time.. But Google Research published a paper introducing a biological "Sleep Framework" for LLMs. They split the AI's existence into two distinct states. 1. The Wake Phase: AI interacts with you. It solves problems, processes data, and stores new information in a fragile, short-term memory bank. 2. The Sleep Phase: When the system goes offline, the AI literally goes to sleep. It unlocks dormant parameters and actively transfers its short-term experiences into permanent, long-term memory. It learns forever without overwriting the past. But then, the AI starts dreaming. During the sleep phase, the model deliberately activates random, unusual neural pathways. It generates synthetic scenarios, literally dreaming up fake data, and tests itself to improve its own reasoning and uncover hidden patterns. The results are staggering. • It retained perfect knowledge across a massive 10-million token window. • It learned entirely new, unseen languages with zero performance drop. • It crushed elite math benchmarks (AIME), jumping to nearly 80% accuracy. Zero catastrophic forgetting. Continuous, autonomous self-improvement. We’ve spent billions trying to build bigger context windows so AI can hold more information at once. But the future of AI isn't about cramming more data into a single prompt. It's about letting the machine process its experiences after you log off.
18
Which one feeds the wife and kids ?
who needs a pre seed round when opencode is literally acting as an angel investor for weekend projects 12 free models, 0% dilution, and enough free inference to ship an MVP before monday morning - GLM-5.3-Flash - DeepSeek V4.1 Flash - Hy4 flash - Inkling 2 Small - MiMo-V2.6-Flash - Flint 40B - Nemotron 3.5 Super VL - Nemotron 3.5 Lightning - Big Pickle - LongCat 2.5 - Iceman - Beam which one is actually carrying your workflow rn?
1
1
27
Great stuff
Everywear 0.1.40 is out for Windows. 3nvizen is free and now has an Edit tab: bring in clips from your Vault, pick takes, trim on a real timeline and remake any shot with a retake. Plus desktop widgets you can place anywhere and a steadier sign-in. everywear.id/download #LocalAI #OwnTheMeansOfCreation #ItJustWorks #NoSubs #NoSaaS #OneClickAI
1
83
couple of hours video rendering.
Replying to @NostromoCore
Damn, I wanted to hear the whole song. I was just starting to get into it.
15
Yah man, ai totally can't do longer shots.
AI’s biggest weakness and easiest tell is that it can’t hold a shot. Nobody is going to watch a film/show that cuts every 2 seconds.
332
56
45
1,359
369,925
Sean Uddin retweeted
It’s honestly incredible to see a film that was made with AI being posted by Fandango. Years in the making and the moment is here. This will open the door for so many AI filmmakers. Congrats to @GossipGoblin on the success.
Watch the new trailer for Gossip Goblin's #GodsDontGiveGifts, coming to theaters this December.
12
5
2
114
5,483
which way western man
Anthropic & OpenAI wants to own intelligence. Open models want to free it
40
74
1
668
11,762