mactop v2.1.5 is now available
- ANE usage support on macOS 27 Beta+ @anemll
- New layouts @anemll
- Bug fixes
brew upgrade mactop
https://t.co/IsoMYMHOVv
We now support rich formatting for all chatbots.
Tables, nested lists, inline media, formulas, headers and more — right in Telegram messages.
🔨 Start building! Docs: https://t.co/zgzPOOUJF5
Most of the internet pages support apple pay, but when you need to pay something in the Developer portal from @Apple You need to fcking fill in ALL your contact information, credit card, and all shit that APple Pay abstracts. But hey, we have nice Emojis
You can now fine-tune free Qwen3.5 on Colab,
Need 5GB VRAM to train Qwen3.5-2B LoRA locally,
1.5x faster with 50% less VRAM.
Qwen3.5-4B Colab: -/colab.research.google.com/github/unslothai/notebooks/blob/main/nb/Qwen3_5_(4B)_Vision.ipynb
Gemma 4 now runs 2x faster with MTP GGUFs! Run locally on just 6GB RAM. ⚡️
MTP enables Google Gemma 4 run ~1.4–2.2× faster with no accuracy loss.
Gemma 4 12B MTP can run at 162 t/s vs. 52 t/s without MTP. 31B reaches 101 t/s.
GGUFs + Guide: https://t.co/c4gAUlb6YE
Recently, we purchased one of each Anthropic/OpenAI subscription plan and randomly ran long horizon coding tasks until we exhausted the weekly limit. It's widely believed that a $200/month plan maxes out at ~$2000/month worth of tokens (assuming API pricing). However, we found that the subscriptions are actually far more generous. (2/4)
"Trajectory-Refined Distillation"
This paper shows a core failure in on-policy distillation.
When a student takes a wrong reasoning path, the teacher is forced to supervise from that broken prefix, so token-level KL becomes noisy.
So they introduced TRD that fixes this by refining the whole rollout first.
What it does is the teacher rewrites the student trajectory into a better reasoning path, then the student distills from that corrected trajectory.
While this is a small change, it still moves distillation from token fixes to trajectory fixes, giving cleaner supervision and stronger reasoning gains.
📣 NVIDIA CUDA 13.3 is here!
CUDA 13.3 brings new capabilities and performance optimizations to developers across the CUDA ecosystem.
Includes enterprise-grade feature parity across CUDA Python and C++ with CUDA Python 1.0, CUDA Tile for C++, and CUDA Tile coverage.
🔗 Read the tech blog: https://t.co/PpsjrDnoSA
Apple finally did it.
Its new framework, Core AI, runs models entirely on Apple silicon, so inference happens on the user's device with zero server calls and zero token bills.
That means Qwen, Mistral, and SAM3 running natively across iPhone, iPad, Mac, and Vision Pro.
It's a memory-safe Swift API that compiles models ahead of time for near-instant load. Pulling one in takes a few lines:
let segmenter = try await ImageSegmenter(resourcesAt: sam3ModelURL)
let response = try await segmenter.segment(image: inputImage, prompt: "flower")
The launch goes beyond the runtime, though.
It ships curated open models packaged for Swift, PyTorch extensions to convert your own, and an optimizer that shrinks models layer by layer with minimal accuracy loss.
There's also a macOS debugger that profiles performance and traces behavior back to your original Python, plus Xcode tools to validate models before they ship.
For any team that wanted real on-device AI without a cloud bill attached to every user, this is the answer.
Models repo: https://t.co/yZCIhrM6YQ
Claude Fable 5 is our first generally available Mythos-class model.
It ships with new safety classifiers that may flag certain prompts in dual-use domains like cyber and bio.
We've added fallbacks: a refused request retries on Claude Opus 4.8 instead of dead-ending.
Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use.
Its capabilities exceed those of any model we’ve ever made generally available.
‼️🚨 BREAKING: ServiceNow has been breached. Customers are reporting unauthorised access to their instances.
One customer states their security team reported this vulnerability to them, and they closed the case twice, saying they had already known since the 7th of April.