Pinned Tweet
There should be 2 internets. One for #llm bots, one for humans. 404media.co/ai-scraping-bots…
I give up... At this point, either I send my agents to run all these verification steps themselves or I just ask my co-authors to remove me from the list. It's more important that #nlproc knowledge goes out than putting any name on things...
Liling Tan retweeted
We worked on a llama.cpp impl for DS4.1-Flash + made Unsloth dynamic GGUFs for it!
The engram / PLE can be swapped out from MXFP8 to Q6_K to Q5_K and IQ4_NL precision - we found Q6_K approx = MXFP8 and IQ4_NL loses a tiny bit of accuracy.
The engram can stream fully from disk!
DeepSeek V4.1-Flash-1017 can now run locally! 🐳
We made 1-bit retain ~88.2% accuracy while shrinking size from 508GB to 250GB (−52%).
Run on a 128GB+ RAM device.
V4.1 Flash 1017 is the strongest open model to date
Guide: unsloth.ai/docs/models/deeps…
GGUF: huggingface.co/unsloth/DeepS…
Take the poison pill until something more permissive comes along or just ignore and wait 1-2 months? huggingface.co/meta-models/M…
#llm #opensource is getting weirder each day…
Wait till the #llm and #machinelearning folks finds this.
Looking for that monitor by the side to keep your terminal UI open for "progress report", "continue autonomously, don't ask me!", "are you sure?" to command your agents. @wesbos foto frame is the way to go. wesbos.com/hacking-smart-pho…
Q: So what's snack is in the pantry
A: Moon pie @_@
youtube.com/watch?v=loamVbAO…
Excited to introduce RSI-Anything!
One RSI model that works across coding, GUI, planning, or even Science tasks.
Let's RSI Everything!
📄 Paper: arxiv.org/abs/2610.13912
💻 Code: github.com/Gen-Verse/RSI-Any…
#RSI #WorldModels #RSI #AIAgents #AIforScience #JEPA
Can we force all our agents to read this before any pull-request or commits to #opensource?
catb.org/~esr/writings/cathe…
Liling Tan retweeted
If you're building a coding agent harness, give your agent this paper and let it figure out how to improve itself.
arxiv.org/pdf/2609.20804
Language is never ever random nature.com/articles/s41593-0… #nlproc
Liling Tan retweeted
You can now train your own Decision model with our free notebook! 💡
Qwen3.5-4B will generate decisions instead of text on just 8GB VRAM locally.
Learn to data prep (state, questions, gold answers), train, serve.
Notebook: colab.research.google.com/gi…
Guide: unsloth.ai/docs/basics/train…
You can now train your own Decision model like Jev locally!
We increased Qwen3.5 0.8B’s aggregate accuracy from 20.7% to 74.3% across 3 decision benchmarks - on just 4GB VRAM.
Turn any LLM like Qwen3.8, Gemma 4 into decision models with our open-source Unsloth repo.
We fine-tuned with a Clef head using Unsloth and LoRA (r=64) for one epoch, increasing downstream accuracy from 30–37% to 78%.
GitHub: github.com/unslothai/unsloth
Guide and Notebooks: unsloth.ai/docs/basics/train…
Liling Tan retweeted
We made it possible to train your own Decision model locally on just 3GB VRAM!
We converted Qwen, Gemma, Llama all into decision models by fine-tuning using a Clef head, boosting accuracy from 30 to up to 78%.
You can try it yourself via Unsloth Desktop!
You can now train your own Decision model like Jev locally!
We increased Qwen3.5 0.8B’s aggregate accuracy from 20.7% to 74.3% across 3 decision benchmarks - on just 4GB VRAM.
Turn any LLM like Qwen3.8, Gemma 4 into decision models with our open-source Unsloth repo.
We fine-tuned with a Clef head using Unsloth and LoRA (r=64) for one epoch, increasing downstream accuracy from 30–37% to 78%.
GitHub: github.com/unslothai/unsloth
Guide and Notebooks: unsloth.ai/docs/basics/train…
E-commerce iceberg fastcompany.com/91619118/ama…
Actually not bad, but definitely room to improve. Creepy yes but not the creepiest I’ve seen.
This is tiresome. PSA: If you are going to modify a morphological analyzer code in an #opensource for a specific language with an agent putting up the PR and commits, please acknowledge if you actually know the language or consulted people who knows the language... #nlproc
Maybe it’s just cos it’s just #friedday and brain is pretty much fried so only idea today is “looking for new challenges, perhaps” #ifYouKnowYouknow
I guess people who knows should know how to reach me 😁
It’s still #friedday somewhere and my brains are double fried now, esp. some mad recap of KL, PMI and silly loss func…
So here’s a couple more unsolicited ideas:
3) Use the infini-gram.io/ for speculative decoding, the authors literally put it in their “envision”
4) Agentic pentesting should be taught in any #llm class, the blast area of any sort of exploit by agents are beyond human imagined creativity, remember this? arxiv.org/abs/1803.03453
I was curious (and a bit sleepless) so I did what I know best, coded and clauded. #llm #startup
Then I saw something interesting. Not sure if anyone else spot the same.
alvations.github.io/melon-se…
Liling Tan retweeted
Quality filters that work in English degrade measurably in every other language you cover. Tested across 35 languages.
The vendor question is not how many languages they cover. It is how quality is measured per language pair.
pangeanic.com/data-for-ai/ev…
