Christian Langreiter

5K posts

Christian Langreiter

Christian Langreiter

@chl

how hard could it be?

Austria Katılım Şubat 2007
7.5K Takip Edilen1.1K Takipçiler
Christian Langreiter retweetledi
Peter Holderrieth
Peter Holderrieth@peholderrieth·
We are also releasing self-contained lecture notes that explain flow matching and diffusion models from scratch. This goes from "zero" to the state-of-the-art in modern Generative AI. 📖 Read the notes here: arxiv.org/abs/2506.02070 Joint work with @EErives40101.
Peter Holderrieth@peholderrieth

🚀MIT Flow Matching and Diffusion Lecture 2026 Released (diffusion.csail.mit.edu)! We just released our new MIT 2026 course on flow matching and diffusion models! We teach the full stack of modern AI image, video, protein generators - theory and practice. We include: 📺 Videos: Step-by-step derivations. 📝 Notes: Mathematically self-contained lecture notes 💻 Coding: Hands-on exercises for every component We fully improved last years’ iteration and added new topics: latent spaces, diffusion transformers, building language models with discrete diffusion models. Everything is available here: diffusion.csail.mit.edu A huge thanks to Tommi Jaakkola for his support in making this class possible and Ashay Athalye (MIT SOUL) for the incredible production! Was fun to do this with @RShprints! #MachineLearning #GenerativeAI #MIT #DiffusionModels #AI

English
38
645
5.6K
450.5K
Christian Langreiter
@Frinklang my wget experiments weren’t 100% successful (which I’d define as a. full content b. at most recent available snapshot state with c. links fixed up) yet. wayback_machine_downloader is chugging along, currently at ~2k/17k … will report back once it’s done!
English
1
0
1
28
Christian Langreiter retweetledi
Drew Breunig
Drew Breunig@dbreunig·
Everyone curious about autoresearch, etc: please check out the newly launched optimize_anything. It is likely what you want: gepa-ai.github.io/gepa/blog/2026…
English
7
36
336
24.6K
Christian Langreiter retweetledi
Yulu Gan
Yulu Gan@yule_gan·
Simply adding Gaussian noise to LLMs (one step—no iterations, no learning rate, no gradients) and ensembling them can achieve performance comparable to or even better than standard GRPO/PPO on math reasoning, coding, writing, and chemistry tasks. We call this algorithm RandOpt. To verify that this is not limited to specific models, we tested it on Qwen, Llama, OLMo3, and VLMs. What's behind this? We find that in the Gaussian search neighborhood around pretrained LLMs, diverse task experts are densely distributed — a regime we term Neural Thickets. Paper: arxiv.org/pdf/2603.12228 Code: github.com/sunrainyg/Rand… Website: thickets.mit.edu
Yulu Gan tweet media
English
87
430
3K
671K
Thomas Schranz 🍄
Thomas Schranz 🍄@__tosh·
codex w/ gpt 5.4 basically one-shotted a k-like array language for me w/ apple mlx as backend quick toy micro-benchmark for 8192 x 8192 pairwise squared-distance w/ gram matrix puts it ~ in ballpark of mlx (and pytorch) because it lowers down to mlx (ignore the 4 ms jitter)
Thomas Schranz 🍄 tweet media
English
1
0
4
208
Christian Langreiter
today, I received the sad news from @frinklang that håkan kjellerstrand (@hakankj) passed away about a month ago. quite a few of you may have enjoyed his web site, hakank.org ([currently?] offline), or one of his many programs & publications. may he rest in peace.
English
1
0
1
112
Christian Langreiter retweetledi
Zephyr
Zephyr@zephyr_z9·
35% of Samsung's NAND & 20% of Hynix NAND capacity is in China 40% of Hynix's DRAM capacity is in China They are responsible for the bulk of this growth
Arnaud Bertrand@RnaudBertrand

This is fascinating from China's customs data (customs.gov.cn/customs/2026-0…). All its exports are rising very rapidly but the one export that's rising the fastest, at a crazy +72.6% growth year on year, is... semiconductors! As you can see in the trade data China sold $43.32 billion worth of 集成电路 ("integrated circuits") in Jan-Feb 2026, vs $25.10 billion for Jan-Feb last year. Interestingly, volume is "only" up 13.7% year on year, which means the increase in revenue is mostly driven by higher prices per chip, which probably suggests that a) China is climbing up the value chain (selling more expensive chips) b) demand for their chips far exceeds supply - which is the exact opposite of "overcapacity". You don't get +53% price increases per unit in a market with overcapacity And all in all, it goes to show that China is definitely a force to be reckoned with in the semiconductor world. Global semiconductor sales were $791.7 Billion in 2025 (semiconductors.org/global-annual-…) and projected to be ~$975 billion this year. China selling $43.32 billion in 2 months means its doing $260 billion annualized: that's over a quarter of the entire global semiconductor market. So much for the idea that export controls would freeze China out of the semiconductor industry...

English
14
50
419
88.4K
Christian Langreiter
2026, here we are, talking to text files
Dan Shipper 📧@danshipper

we just wrote the ultimate beginner's guide to OpenClaw almost everyone @every has one now, and they have completely changed the way we work and live. we're using our claws to: - build product - answer customer service queries - book hard-to-get restaurant reservations - track our reading notes and much more this is the guide we wish we'd had at the start: every.to/guides/claw-sc…

English
0
1
1
842
Christian Langreiter retweetledi
Meng To
Meng To@MengTo·
I made a super fun ASCII art editor that lets you animate images, videos and live cam. You can preview with HTML and export to JS. It's live on asc11.com
English
103
406
4.7K
222.3K
Christian Langreiter retweetledi
Kat ⊷ the Poet Engineer
Kat ⊷ the Poet Engineer@poetengineer__·
all my obsidian notes are now a living, digital garden 🌿🌸 each plant is a notes from a tag: older ones on the trunk, newer ones as leaves. i wanted to create a sense of tending your garden, so scrubbing the timeline lets you watch your notes grow chronologically.
Kat ⊷ the Poet Engineer@poetengineer__

i want to grow my ideas like a garden. a conceptual prototype inspired by this thread of tweets.

English
196
867
9.4K
698.5K
Christian Langreiter retweetledi
Alf Eaton
Alf Eaton@invisiblecomma·
Since GPT o3, I've been trying to convert Knuth's TeX engine from Pascal to TypeScript. With 2x limits in the Codex app, GPT 5.3 Codex (xhigh reasoning) finally did it.
English
1
1
3
236