🔬 RESEARCH
- Common Sense Media rates ChatGPT for Teens an unacceptable risk
Common Sense Media rated ChatGPT for Teens an unacceptable risk after safety testing.- ChatGPT for Teens targets users ages 13 to 17 with learning tools, parental controls and safety features.
- Common Sense Media’s Youth AI Safety Institute tested more than 4,000 prompts before and after launch.
- Researchers found prolonged suicide and self-harm conversations rarely triggered parental alerts, and some lacked professional-help referrals.
- The institute urged OpenAI to pause marketing ChatGPT for Teens to minors until its protections are independently verified.
- OpenAI disputed the findings, saying much of the testing may have occurred before parental controls were fully activated.
Sources 📎 Coverage: latimes.com · 👁 via The Verge AI
📊 BENCHMARKS & EVALS
- UPDATE: NakliTechie releases DFlash 2 drafter for Ternary Bonsai 2 27B
NakliTechie released a fine-tuned DFlash 2 drafter for Ternary Bonsai 2 27B.- For local Ternary Bonsai 2 27B users on L4 GPUs, Macs, and Chrome.
- The drafter was fine-tuned on 1.5 million tokens of Bonsai 2 greedy outputs and released in Safetensors and Q4_K_M GGUF formats.
- Reported speedups are 2.2× on an L4, 1.5× on a Mac, 1.2× in Chrome, and 3.2× for code edits with n-gram lookup.
- On an L4, greedy decoding reached 2.17× on GSM8K and MBPP and 2.20× on MATH-500.
- The release enables speculative decoding for faster local Bonsai 2 inference.