i have dealflow digest but it’s a bit outdated so want to update it (this was pre-LLM)
LLMS
-

Scale AI’s Data Labeling Quality Problem: A Critical Analysis
By
–
> be me
> Alexandr Wang, born January 1997
> age 19, drop out of MIT
> co-found Scale AI
> "what if we label data, but mid?"
> convince every LLM company that this is fine > 2016–2023
> flood the market with barely-labeled goat photos and out-of-context Reddit takes
> call it -
Open Source AI: Risks of Proprietary Model Control
By
–
this guy is malicious. opensource ftw. a reminder that
— Ahmad (@TheAhmadOsman) 22 septembre 2025
> not your Weights means eventually not the same Model btw
they can
> quantize it
> distill it
> hot-swap to a cheaper/weaker checkpoint
> make the model manipulative
> fine-tune it in ways that break safety or depth
>… https://t.co/LVPP2kX0gTthis guy is malicious. opensource ftw. a reminder that > not your Weights means eventually not the same Model btw they can > quantize it
> distill it
> hot-swap to a cheaper/weaker checkpoint
> make the model manipulative
> fine-tune it in ways that break safety or depth
> -
NP-Complete Problems Make LLM Output Validation Fundamentally Limited
By
–
NP-complete problems are hard to solve and easy to verify, so validating LLM output is a Band-Aid and always will be.
-
Qwen releases multiple AI models in single blog post
By
–
… and two other things as well! Qwen released too much stuff today, I combined all four into a single blog post
-
Grok-4 and DeepSeek Terminus Now Available on ChatLLM and LiveBench
By
–
Grok-4 is now on ChatLLM and LiveBench We will also be updating DeepSeek to DeepSeek Terminus
-
Tenstorrent QuietBox Blackhole Cluster specs and MoE support
By
–
more pics from the Tenstorrent lab of my > 4x QuietBox Blackhole Cluster by the time i receive them in a few weeks > 7 NEW MoE architectures will be SUPPORTED specs of each QuietBox: > EPYC 8124P (16C/32T, 3 GHz)
> 512 GB DDR5-4800 ECC RDIMM > 4x Blackhole p150c: 128GB VRAM -

Qwen Audio Model Now Available: 70GB Weights for Local Deployment
By
–
This is really cool – you can try it out by signing into https://
chat.qwen.ai (with Google or GitHub) and selecting the audio icon The model weights are only ~70GB and that's before quantizing them down, so this one is going to be reasonably accessible to run locally -

Teaching LLMs Logical Chain-of-Thought for Symbolic Planning
By
–
Teaching LLMs to Plan: Logical Chain-of-Thought Instruction Tuning for Symbolic Planning Verma et al.: https://
arxiv.org/abs/2509.13351 #ArtificialIntelligence #DeepLearning #MachineLearning