as (@AshikHafi) header
as (@AshikHafi) profile picture

https://teapawt.net/t.co/Q10zgd9lB6

Chicago, IL
Joined May 2016
achieved AGI at what costs 馃ゲ
2
62
Omg odegaard what a goal! #arsenal
1
4
112
We are just getting started!
this summer was crazy... - discussing AI policy with actual ministers - meeting @maxjunestrand at @ycombinator stockholm - flying to SF on 24 hour notice for a meeting - launching Stratson, met with overwhelming response here's to an awesome Q4 馃嵕
1
1
3
220
yo wtf is this woke half time interview shit 馃槶馃槶 premier league is finished man just let them play football
2
68
Oh my days martin 酶degaard!!
2
52
GPT 5.3 Codex Spark got stuck in a compaction loop After context compaction, it seems to be repeatedly entering a loop: "I'm doing one more audit pass" "I'm checking a few more files" "I'll summarize findings shortly" @thsottiaux @jxnlco
1
1
100
Then context compacts again and it restarts the same audit plan. Result: ~36 minutes of runtime dozens of file reads/searches multiple compactions never produced the actual findings
1
1
29
fuck! claude code is soo back.
1
29
I fucking love this club
This belongs to all of us.
1
23
Our FA Cup journey ends here. Attentions turn to Sporting CP on Tuesday.
1
33
Moltbots in disguise
2
24
OMGGGG Benefica #UCL
1
95
thats the pass we gotta feed Gy枚keres. LFG
34
that was an insane attempt from saliba. #CHEARS
353
馃У 1/10 Just read DeepSeek-AI鈥檚 new paper: mHC (Manifold-Constrained Hyper-Connections). It fixes a real problem with Hyper-Connections: they scale poorly due to unconstrained residual mixing. mHC adds a simple constraint that makes them stable at scale. Thread 猡碉笍 #AI #DeepLearning arxiv.org/abs/2512.24880
1
1
91
2/10 First, the basics: In deep nets (e.g., Transformers), "residual connections" let info skip layers, like shortcuts in a highway. This keeps training stable as models get deeper. Think of it as adding the input directly to the output: x_next = x + f(x).
1
30
Wooo. What a start to 2026!!
We believe the next breakthrough in long-horizon agents is training models to manage their own context. Introducing our new research direction on Recursive Language Models. We are sharing our initial experiments showing the promise of RLMs. primeintellect.ai/blog/rlm
2
74
let it all work out.
20
Cursor cooked with the debug mode. @cursor_ai
41
Why we scoring worldies today. Wtf 馃槼 #arsenal
43