It's terrible for most regular people. We are memory poor — MoE models make this even worse. It may be cute, but it's totally impractical. Dense models are far more usable in practice.
LLMS
-
ChatLLLM: Boost Productivity with AI-Powered Automation Tools
By
–
ChatLLLM drafts emails, manages projects, creates images & videos, and automates workflows. You can 10x your productivity without requiring heavy investment in technical skills! Check it out at:
-
Grok 4 builds Gemma-3-270M chatbot with transformers.js
By
–
Grok 4 one shots building a gemma-3-270m chatbot with transformers.js
— AK (@_akhaliq) 15 août 2025
one click deploy in anycoder pic.twitter.com/zciwWazuEFGrok 4 one shots building a gemma-3-270m chatbot with transformers.js one click deploy in anycoder
-
New Vulnerability Enables Pre-Fine-Tuning Weight Recovery Attack
By
–
— dr. jack morris (@jxmnop) 15 août 2025
We uncover a new vulnerability- Pre-Fine-Tuning Weight Recovery
With a few LoRA fine-tuned models we recover the pre-fine-tuning weightsof SoTA models, undoing Stable Diffusion personalization training and Mistral alignment
Project: http://
vision.huji.ac.il/spectral_detun
ing
… -
Working on Qwen: Extraction technique and model development
By
–
that's a good point; probably should have said 'we extracted one' not 'it'? working on qwen tonight
-
Clarifying Base Model Distribution Recovery in AI Experiments
By
–
some people are getting confused about the experiment – we didn't recover the base model's *weights*. that might not even be possible. we recovered the base model's *distribution*, with some error. an important question is how much. trying to figure that out right now…
-
OpenAI releases gpt-oss models on Databricks platform
By
–
Last week, we announced @OpenAI
’s new gpt-oss-20B and gpt-oss-120B models are now live on Databricks! They bring chain-of-thought reasoning and tool use, best-in-class latency and cost efficiency through Mixture-of-Experts, plus a 131k context window for long docs and RAG. -
Gempix Model and Precise Reference Toggle Spotted on AI-Powered Whisk
By
–
BREAKING : Gempix model is spotted on Whisk along with a new "Precise Reference" toggle. This feature is already available to trusted testers where Whisk is powered by Imagen 4. Pixel event should be fun
-
Benchmarking AI Models for Psychological Safety Risks
By
–
This is a much needed first attempt at a benchmark to measure how much given AI models will play along with users pushing them in delusional or potentially psychologically dangerous directions. Some early signal that full GPT-5 (not chat) is a less psychologically risky model.
-
OpenAI Brings Back GPT-4o After GPT-5 Backlash
By
–
GPT-4o is Back
After GPT-5 backlash, OpenAI brings back the fan favorite.
– GPT-4o for Plus users
– GPT-5 limits doubled
User experience > tech hype.