I keep seeing this advice to quantize the KVCache to 4-bit and save on memory Please don’t do that KV Cache quantization beyond FP8 usually is asking for a nerfed and incoherent model
LLMS
-
Stanford AI Index: Model race over per Elo ratings
By
–
The AI model race is over. Most people won’t realize it for another six months. Stanford’s 2026 AI Index published the numbers two weeks ago. Arena Elo ratings across every major lab: Anthropic 1,503. xAI 1,495. Google 1,494. OpenAI 1,481. Alibaba 1,449. DeepSeek 1,424.
-
Self-reflection and prompts improve AI output quality
By
–
Well yes, if you lie to yourself and to your AI, then the output will indeed be low quality. It’s a product of self-reflection. And relatedly, the prompt you give Claude if you want it to interview you to build it. If folks don’t want to write their own, I created prompts
-
Glean Launches Waldo Agentic Search Model
By
–
Glean just got smarter and faster with Waldo.🧐
— NVIDIA AI (@NVIDIAAI) 28 avril 2026
Congrats to the @Glean team on rolling out their first agentic search model. https://t.co/szfWB2VEuuGlean just got smarter and faster with Waldo. Congrats to the @Glean team on rolling out their first agentic search model.
-
Anthropic Models Use More Tokens in Spanish Translation
By
–
This is interesting: Anthropic's models suffer more from "speaking" other languages in terms of token usage. In Spanish, the number of tokens used multiplies by 1.62x compared to OpenAI's model in English.
-
Can LLMs trained on 1930s text learn to code?
By
–
Codex is doing AI research for me on Talkie using its in-app browser research question: can an LLM trained only on 1930s text learn to code?
-
1000+ Local AI Models Platform for Offline Agents
By
–
That's a wrap > 1,000+ models including Qwen and Llama, MLX and everything to power your agents with local models running agent
> Completely offline
> Blazing fast thanks to Google TurboQuant Try it yourself → http://
atomic.chat -
Running 35B LLM Locally on MacBook with Impressive Speed
By
–
HOLY MOLY running a 35B model locally on a MacBook shouldn’t be THIS FAST 🤯
— Charly Wargnier (@DataChaz) 28 avril 2026
Spent my weekend in @atomic_chat_hq testing Qwen 35B vs. Qwen 27B on my local machine.
I had them generate a fully animated HTML/Canvas car mini-game (demo below),
… and both models breezed through… pic.twitter.com/WXboJCSDZnHOLY MOLY running a 35B model locally on a MacBook shouldn’t be THIS FAST Spent my weekend in @atomic_chat_hq testing Qwen 35B vs. Qwen 27B on my local machine. I had them generate a fully animated HTML/Canvas car mini-game (demo below), … and both models breezed through
-
Testing Time LLM from Industry Legends
By
–
Test of Time LLM from some legends…pretty fun! https://t.co/ZsH7oLkUSM
— Soumith Chintala (@soumithchintala) 28 avril 2026Test of Time LLM from some legends…pretty fun!
-
Daily drops on LLMs agents and data workflows
By
–
♻️ If this sparked an idea, hit repost so others can catch it too!
— Charly Wargnier (@DataChaz) 28 avril 2026
Follow me → @datachaz for daily drops on LLMs, agents, and data workflows 🦾 https://t.co/SDIX3Rm3tdIf this sparked an idea, hit repost so others can catch it too! Follow me → @datachaz for daily drops on LLMs, agents, and data workflows