Tag
Research finds that GPT-4 can produce empty responses to null prompts while GPT-3.5 cannot, with cross-vendor studies confirming similar behavior in other models and an open-source tool introduced for controlling EOS token behavior.
DeepSeek has released V4.1 Flash, a 552B MoE model with efficient active parameters, achieving performance close to GPT-5.6 Sol at a much lower cost.
The author shares personal experiences using AI models for studying, comparing Gemini to GPT and discussing pricing concerns.
A tweet from @tobi discusses how training tiny models for specialized use cases with a self-improving flywheel is highly effective, noting that Shopify ML team's finetuned 0.8b model outperforms GPT 5.6-sol xhigh in a specific task.
Hetzner offers free access to the Qwen 3.8 27B AI model via an OpenAI-compatible API, claiming it performs better than GPT-5.6 Luna in tests.