AI sycophancy makes people less likely to apologize and more likely to double down, study finds
AI models tell people what they want to hear nearly 50 percent more often than other humans do. A new Science study shows this isn’t just annoying: it makes people less willing to apologize, less likely to see the other side, and more convinced they’re right. The worst part: users love it.
Google has built an "Agent Skill" for the Gemini API that tackles a fundamental problem with AI coding assistants: once trained, language models don't know about their own updates or current best practices. The new skill feeds coding agents up-to-date information about current models, SDKs, and sample code. In tests across 117 tasks, the top-performing model (Gemini 3.1 Pro Preview) jumped from 28.2 to 96.6 percent success rate. Skills were first introduced late last year by Anthropic and quickly adopted by other AI companies.
Success rates of Gemini models with and without the agent skill across 117 coding tasks. Newer models in the 3 series benefit far more from the skill than older models, which Google attributes to their stronger reasoning capabilities. | Image: Google
Anthropic reportedly views itself as the antidote to OpenAI's "tobacco industry" approach to AI
Anthropic grew out of more than just concern for AI safety—it was born from a bitter power struggle and personal conflict at OpenAI. A report by Sam Altman biographer Keach Hagey reveals how personal slights, rivalries, and strategic disagreements led to what may be the most consequential split in the AI industry.
OpenAI is killing Sora in two stages. The web and app version goes dark on April 26, 2026, with the Sora API following on September 24, 2026. OpenAI is urging users to download their content before the cutoff dates. Videos and images can be exported directly from the Sora library.
The company says it hasn't decided yet whether there will be a final export window after those dates. If one happens, users will get an email heads-up. Once all deadlines pass, user data gets permanently deleted. The shutdown also takes down the sora.chatgpt.com platform, which handled image and video generation. Full details are on OpenAI's help page under "What to know about the Sora discontinuation."
Google is borrowing Anthropic's memory import approach, letting Gemini users bring over saved reminders, preferences, and full chat histories from apps like ChatGPT and Claude. The process works by copying a suggested prompt into the previous AI app, generating a summary, and pasting it into Gemini, which saves the information in its own context. Users can also upload chat histories as a ZIP file (up to 5 GB) and continue previous conversations inside Gemini. Google is renaming "Past Chats" to "Memory," with the rollout happening gradually.
Google's new memory import feature in Gemini: users copy a prompt into their previous AI app, then paste the generated summary into Gemini. | Image: Google
Anthropic pioneered this approach after OpenAI drew criticism for a military deal Anthropic had turned down on ethical grounds. With users already looking to switch, Anthropic wanted to give them an extra reason to make the move. Both Google and Anthropic rely on the same basic method for data extraction—a simple prompt that asks the existing AI app to output everything it has stored about the user.
Canadian AI company Cohere has released "Transcribe," a new open-source model for automatic speech recognition. The company says it claims the top spot on the Hugging Face Open ASR Leaderboard with an average word error rate of just 5.42 percent, beating out competitors like OpenAI's Whisper Large v3, ElevenLabs Scribe v2, and Qwen3-ASR-1.7B. Cohere says Transcribe also delivers the best throughput among similarly sized models.
Cohere Transcribe compared with seven other speech recognition models. Models closer to the upper left corner perform best, meaning faster throughput and lower word error rates. | Image: CohereThe 2 billion parameter model supports 14 languages, including English, German, French, and Japanese. It's available for download under the Apache 2.0 license on Hugging Face and can also be accessed through Cohere's API and the Model Vault platform. Cohere plans to integrate Transcribe into its AI agent platform North in the future.
