Large Language Models

Reverse Turing test illustrated as a scarecrow on a wooden post with a hat, built to pass as human

AIs Fail to Reliably Detect AIs in Reverse Turing Tests

In a reverse Turing test run at Paris Cité University, seven leading chatbots questioned hidden respondents and caught only three AIs in 98 tries, rating the machines more human than real people. We explain how the experiment worked, the numbers behind it, why the AI judges failed and what it means for bot screening, hiring and content provenance.

Read more
ai gender bias womens language workplace prompts a gramophone with a flared horn on a square box body

AI Might Be Making Women Sound Bad at Work

Johns Hopkins researchers took 427 real workplace writing prompts, rewrote each into a women-associated and a men-associated version, and ran both through GPT-4, Llama, Mistral and Gemma. Every model returned shorter, plainer, less formal documents for the women-coded phrasing. Adding a male or female sign-off name changed nothing. Linear probes decode the linguistic register at 0.988 accuracy by layer 5 while name gender reaches only 0.717, and activation patching puts the causal weight in layers 0 to 7. This article covers the method, the per-model results, the two controls that rule out mirroring, and why the authors say users cannot fix it themselves.

Read more
Grok 4.8 - grok 4 8 2 5t model cpp software stack a block stacking tower with one bar pushed out

Grok 4.8 to Feature 2.5T Parameters and New C++ Software Stack

Elon Musk says Grok 4.8 is a 2.5-trillion-parameter model trained with xAI’s new C++ software stack and will start reinforcement learning this week. We check the claim against every parameter count and stack post Musk has made, the delayed Grok 4.7, Grok 4.6’s published benchmarks, and what a 2.5T model means for memory, speed and price.

Read more
hy4 preview tencent open weight moe 1m context a solid dodecahedron

Tencent Releases Hy4 preview: An Open-Weight MoE Model With a 1M Context Window

Tencent open-sourced Hy4 preview on 28 August 2026 under Apache 2.0: a 770B Mixture-of-Experts model that activates just 49B parameters per token and reads a one-million-token context window. This breakdown covers the full architecture, from 256 routed experts per layer to Gated DeepSeek Sparse Attention and the built-in speculative decoding layer; every benchmark figure Tencent published, including the 163-expert blind evaluation against GLM 5.3 and Kimi K3; the API price list against GPT-5.6 Sol; the eight-GPU serving recipes; and the four caveats worth naming before any of it reaches production.

Read more
CHAT