Consumer AI

Researchers specifically sought to evaluate the quality of AI teacher assistants — such as MagicSchool, Khanmingo, Curipod, and Google Gemini for Education — that are designed to support classroom planning, lesson differentiation, and administrative tasks.

Common Sense Media found that while these tools could help teachers save time and streamline routine paperwork, AI-generated content could also promote bias in lesson planning and classroom management recommendations.

The Thursday release of GPT-5 brings together traditional and "reasoning" models and ups the ante in the race toward so-called artificial general intelligence (AGI).

Driving the news: OpenAI is making the new model available Thursday for free and paid users, with enterprise and educational customers getting access starting next week.

the main improvements in GPT-5 included: fewer factual errors, or hallucinations; better software coding, allowing it to create functional websites and apps; increased capability at creative writing; and, rather than “refusing” a prompt that breaches its guidelines outright, the model will instead try to give the most helpful response possible within safety guidelines, or at least explain why it cannot help.

The agent feature in ChatGPT – which carries out tasks such as finding restaurant availability and shopping online – will also be able to access users’ Gmail, Google calendar and contacts, if given permission.

As with its predecessors, GPT-5 can generate voice, image and text and can deal with queries in those formats too.

OpenAI said the upgraded ChatGPT would be better at answering health-related questions and would be more proactive at “flagging potential concerns” – such as serious physical or mental illness.

Kitten TTS is an open-source series of tiny and expressive text-to-speech models for on-device applications. We are excited to launch a preview of our smallest model, which is less than 25 MB. This model has 15M parameters.

This release supports English text-to-speech applications in eight voices: four male and four female. The model is quantized to int8 + fp16, and it uses onnx for runtime. The model is designed to run literally anywhere eg. raspberry pi, low-end smartphones, wearables, browsers etc. No GPU required!

For researchers, scientists, and academics tackling hard problems: Gemini 2.5 Deep Think is here. 🤯

It doesn't just answer, it brainstorms using parallel thinking and reinforcement learning techniques.

publishers should “structure the content in an easier way to be grabbed” — in citable chunks, with clear authorship.

Worstell agrees. “I’ve found that when it comes to AI, listicles do well on niche sites,” he says. “They’re looking for expert content. If I can put together a document that’s easy to crawl, they’ll actually source it.” The concept of being easy to cite — of producing text that can be easily sliced or summarized for inclusion in a multi-part conversation — came up with everyone I talked to, as did structured material: comparison tables; rankings; and clear, authoritative recommendations and original research.