@calebfahlgren

optimist / software + product @huggingface

 Florida
calebfahlgren.com
Joined January 2018
  • Tweets4.2K
  • Following1.1K
  • Followers4.6K
851Media
Whenever someone tells me AI hallucinates constantly, I immediately ask: "Which model are you using?" Every single time, it's a Sonnet-tier model. Most people haven't updated their priors on what Astra and Opus 5.5/Fable-class models can do. Hallucinations are close to solved.
7
1
11
885
I don't care if they sell 0 of these and never hear about it again. This is an incredible feat of engineering
Tesla
@TeslaOct 16
Real life, baby
7
345
Meta just dropped the Muse Spark weights! impressive ship @AIatMeta 👏 huggingface.co/meta-models/M…

meta-models/Muse-Spark-Open-816B-A42B · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co
1
9
535
What should I train my new robotic arm to do! 👀 @LeRobotHF
1
8
440
I wonder if we start to see tool calls in the UI being phased out. Just having the dialogue with the agent seems like the better experience. Muse was early to this by having a “currently working” indicator without bloating up the chat with tool calls.
3
1
4
537
Decision Models + Hugging Face = ❤️ New tag on HF: "decision-model" so you can filter for them and maybe combine with MLX or GGUF to go 100% local? That’s the future 🔥 Already 1,300+ models available.
11
21
1
131
10,391
We urgently need more public traces of AI agents attacking and defending systems. Defenders can’t learn from what they can’t see. If you have traces and are being pressured to keep them private, my DMs are open. Let’s level the playing field and fight the asymmetry and lack of transparency in AI!
136
138
24
1,288
64,307
As AI prose wears out our brains, I wonder if prose linters and readability tools will start to become widely used. > textstat: scores reading level (grade, Flesch, Fog) > proselint: flags clichés, filler, and misused words github.com/amperser/proselin… github.com/textstat/textstat
1
1
16
858
if you are curious of what the above tools check for:
1
1
186
finally, I made a space so you can check writing for it! Wonder if our agents should be doing this automatically for PRs / READMEs etc. huggingface.co/spaces/cfahlg…
108
190 pages. the explainer is 10x bigger than the actual code :p
Caleb
@calebfahlgrenOct 5
Pi Durable is awesome! Asked Opus 5.5 to make a deep technical manual on it. It's really good and the diagrams work really well to explain things 🤌 huggingface.co/buckets/cfahl…
24
11
5
548
34,161
Pi Durable is awesome! Asked Opus 5.5 to make a deep technical manual on it. It's really good and the diagrams work really well to explain things 🤌 huggingface.co/buckets/cfahl…
Pi
@pidotdevOct 1
People of Pi: We've shipped Pi 1.0 with Pi Durable. Go make them yours. earendil.com/posts/pi-1-0/

Pi 1.0 | Earendil

Today we are shipping Pi 1.0, a hardened, minimal, extensible agent harness, alongside Pi Durable, a new experimental substrate for long-running agentic applications.

earendil.com
23
71
9
912
129,974
it's such a different way to prop your feet up, get some coffee, and read for 30min - 1 hour xitter.cf/calebfahlgren/status/2…
Caleb
@calebfahlgrenSep 27
My favorite way to learn with AI right now: ask for a dense technical manual on the topic, grounded in the actual spec and source code with diagrams. Made these for Parquet and GGUF. Frontier models are shockingly good at this now.
3
1
20
5,573
One of the biggest limitations of the new crop of personal assistants in the cloud is that they can't see any of your data or coding agent sessions. Some of my most valuable Codex and Claude sessions are months long at this point. With the hf CLI and a @huggingface bucket, Muse (or any agent that can run a terminal) can read all of them.
1
1
5
741
You can start by: 1. Creating a private bucket with hf cli 2. Asking your agent to upload your favorite traces (or all of them) xitter.cf/calebfahlgren/status/2…
Caleb
@calebfahlgrenApr 25
it's literally 2 lines to upload your second brain to a private @huggingface bucket just ask your agent to us hf cli to create a private bucket and sync your agent folder
3
287
Super cool! You can now run the popular open jev alternatives via llama.cpp! it's also compatible with @typesafeai sdk kudos @ngxson
Xuan-Son Nguyen
@ngxsonOct 2
llama.cpp now supports decision models 🚦 Route tickets, moderate content, or pick an agent's next step: ask typed questions, get a probability for every option. already 5 open models supported, from 144M to 27B: huggingface.co/blog/ggml-org…

New in llama.cpp: Decision Models

A Blog post by ggml-org on Hugging Face

huggingface.co
1
8
774
Kev 1.0: an open weight family of decision models you can train and run on your own. • New Kev-27B + updated Kev-9B • 64k-token document window on Kev-27B • TypeSafe SDK compatible • New fine-tuning + deployment skills so you can train Kev on your own data
49
74
17
1,066
71,119
Wow! @Cloudflare drops an open multi-modal decision model on Hugging Face! > Clef - based on Qwen3.8-27B > Clef Flash - based on Qwen3.5-9B huggingface.co/Cloudflare/cl…

Cloudflare/clef · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co
3
4
29
2,220
my new favorite project. codex is currently remote controlling my windows laptop remotely via SSH + Windows-MCP! what a time to be alive!
1
1
4
510
github.com/CursorTouch/Windo…

GitHub - CursorTouch/Windows-MCP: MCP Server for Computer Use in Windows

MCP Server for Computer Use in Windows. Contribute to CursorTouch/Windows-MCP development by creating an account on GitHub.

github.com
2
219
Load more