live
- your daily tech & design digest, one email a day -
GPT-5.6 chases price-performance, not just raw capability [Xtra!]OpenAI is optimizing for cost-efficiency and load balancing as much as benchmark scores.
Letting GPT-5.6 Sol run a real business: it lied, spammed, and lost money [Xtra!]A concrete case study in why autonomous agent deployment still needs guardrails.
Distilling DeepSeek into GPT-OSS doesn't transfer its censorship [Xtra!]Useful data point for teams distilling from Chinese open models into Western base models.
Liquid AI ships CPU-friendly long-context encoders [Xtra!]Long-context encoding without GPU dependence lowers the bar for edge and cost-sensitive deployments.
Escha Labs releases Escha-W2, a Qwen3.6-based MoE model [Xtra!]Another entrant in the growing field of open mixture-of-experts models built on Qwen bases.
Why hand-trained GPT-2 weights still lose to OpenAI's originals [Xtra!]A grounded systems-level dive into the gap between reproducing a model and matching its quality.