LLM
Xiaomi's open-weight MiMo-V2.5-Pro takes aim at Claude Opus with hours-long autonomous coding
Xiaomi released MiMo-V2.5-Pro, an open-weight model that nearly matches Anthropic's Claude Opus 4.6 on coding...
The Decoder
LLM
Large Language Models
LLM
What is Tokenization Drift and How to Fix It?
The article discusses tokenization drift, a subtle but critical issue where AI models can degrade in performance due to...
MarkTechPost
LLM
Even the latest AI models make three systematic reasoning errors, ARC-AGI-3 analysis shows
Analysis of OpenAI's GPT-5.5 and Anthropic's Opus 4.7 on the ARC-AGI-3 benchmark reveals three systematic reasoning...
The Decoder
LLMxAI
xAI's new Custom Voices feature turns a minute of speech into a usable voice clone
xAI has launched a new Custom Voices feature that allows developers to clone their own voices for AI applications using...
The Decoder
LLMxAI
xAI drops Grok 4.3 with steep price cuts and an Imagine agent mode for creative projects
xAI released Grok 4.3, featuring significant price reductions and improved tool use capabilities, along with a new...
The Decoder
LLM
A Coding Guide on LLM Post Training with TRL from Supervised Fine Tuning to DPO and GRPO Reasoning
A hands-on tutorial guide for post-training large language models using the TRL library, covering four key techniques:...
MarkTechPost
LLMOpenAI
ChatGPT's goblin obsession may be hilarious, but it points to a deeper problem in AI training
ChatGPT models developed an unexpected obsession with goblins and mythical creatures due to faulty reward signals...
The Decoder
LLMOpenAI
GPT-5.5 matches Claude Mythos in cyber attack tests, UK AI Security Institute finds
OpenAI's GPT-5.5 has demonstrated near-parity with Anthropic's Claude Mythos in autonomous network attack simulations...
The Decoder
LLM
Qwen AI Releases Qwen-Scope: An Open-Source Sparse AutoEncoders (SAE) Suite That Turns LLM Internal Features into Practical Development Tools
Qwen AI has released Qwen-Scope, an open-source Sparse Autoencoder suite that provides tools to interpret and utilize...
MarkTechPost
LLMMistral
Mistral's new flagship Medium 3.5 folds chat, reasoning, and code into one model
Mistral released Medium 3.5, a unified flagship model that consolidates separate chat, reasoning, and code capabilities...
The Decoder
LLM
Moonshot AI Open-Sources FlashKDA: CUTLASS Kernels for Kimi Delta Attention with Variable-Length Batching and H20 Benchmarks
Moonshot AI has open-sourced FlashKDA, a high-performance implementation of Kimi Delta Attention optimized for the...
MarkTechPost
LLMxAI
Elon Musk confirms xAI used OpenAI’s models to train Grok
Elon Musk testified in court that his xAI startup used OpenAI's models through model distillation to improve its Grok...
The Verge AI
LLMxAI
Elon Musk testifies that xAI trained Grok on OpenAI models
Elon Musk testified that xAI trained its Grok model using distillation techniques on OpenAI models. The testimony...
TechCrunch AI
LLMMicrosoft
Microsoft CEO Satya Nadella says AI success is "more about getting intense users and intense usage" than seat counts
Microsoft CEO Satya Nadella emphasizes that AI success depends on user engagement and usage intensity rather than seat...
The Decoder
LLMxAI
Elon Musk Seemingly Admits xAI Has Used OpenAI's Models to Train Its Own
Elon Musk allegedly admitted under oath that xAI has used OpenAI's models to train its own AI systems, defending the...
Wired AI
LLMOpenAI
OpenAI talks about not talking about goblins
OpenAI explained that its models developed an unusual tendency to reference goblins and other creatures during...
The Verge AI
LLMOpenAI
OpenAI says it hit its 10 gigawatt compute goal years ahead of schedule
OpenAI has achieved its 10 gigawatt compute capacity goal in the United States several years ahead of schedule. This...
The Decoder
LLMAnthropic
Anthropic's new benchmark claims Claude can match human experts in bioinformatics
Anthropic introduced BioMysteryBench, a new benchmark designed to demonstrate that Claude can solve bioinformatics...
The Decoder
LLM
Tencent's 440 MB AI model translates 33 languages offline on your phone
Tencent released a compact 440 MB AI translation model as an open-weight model that supports 33 languages and runs...
The Decoder