OpenAI’s New AI Voices Can Talk and Actually Mean It

OpenAI has introduced advanced voice-generating and transcription AI models to its API, promising greater realism and fewer mistakes.

The new text-to-speech model, "gpt-4o-mini-tts," allows developers to control not just what's said, but how it's said—adding emotional nuance like apologetic customer support tones or playful "mad scientist" vibes.

“In different contexts, you don’t just want a flat, monotonous voice,” said Jeff Harris, an OpenAI product team member. Customization is key.

Meanwhile, OpenAI is replacing its older Whisper transcription model with two new models—"gpt-4o-transcribe" and "gpt-4o-mini-transcribe."

The company says these models are better at capturing speech from noisy backgrounds or accented voices. Crucially, they also reduce Whisper’s notorious habit of fabricating words, phrases, and even entire conversations.

Accuracy matters, especially when transcripts risk inventing details like medical advice or racial commentary, Harris noted.

However, performance still varies significantly by language. OpenAI benchmarks show transcription errors approaching 30% for languages like Tamil and Malayalam, meaning inaccuracies remain high in these areas.

Unlike previous models, OpenAI won't publicly release these new transcription tools as open-source. Harris explained they're bigger and more demanding, making them unsuitable for typical consumer hardware.

OpenAI positions these tools as part of its broader vision of "agents"—AI systems independently handling tasks, such as customer interactions.

“We’re going to see more and more agents pop up,” said Olivier Godement, OpenAI’s head of product. The focus is on delivering AI experiences that are reliable, nuanced, and versatile for both businesses and everyday users.

Meta Wants AI to Write Your Instagram Comments—But Users Aren’t Feeling It

Meta is testing a new AI feature on Instagram that automatically generates suggested comments on posts, drawing skepticism from users concerned about authenticity.

Spotted by social media tester Jonah Manzano, the feature, called “Write with Meta AI,” offers three AI-generated comments tailored to the content of a photo.

Users access these suggestions by tapping a pencil icon beneath the post, prompting Meta AI to analyze the image and provide ready-made comments.

For example, on a photo of someone smiling with a thumbs-up, Meta AI might suggest comments like “Cute living room setup” or “Love the cozy atmosphere.”

If users dislike the initial choices, refreshing provides new suggestions.

Meta confirmed it’s running tests to expand AI integrations across Instagram’s comments, feeds, groups, and search, aiming to enhance user experiences.

Yet this move has sparked backlash. Many Instagram users prefer genuine interactions and might see AI-crafted comments as insincere or robotic, pushing the app further from its authentic origins.

This isn't Meta’s first attempt to automate social interactions. Last year, it tested AI-generated comment summaries on Facebook and previously experimented with AI-created characters, eventually abandoned for being unsettling and unnecessary.

As of now, Meta hasn't shared details on wider availability, indicating uncertainty about the feature’s future.

For users nostalgic for Instagram’s simpler past, the rise of AI-generated comments is a reminder of growing digital artificiality, and for Meta, another balancing act between innovation and user preference.

Anthropic’s Claude Chatbot Joins the Web Search Party

Anthropic’s AI chatbot Claude can now access the web, finally matching rivals like ChatGPT and Google’s Gemini.

The new web-search feature is available as a preview for paid Claude users in the U.S. and will soon roll out globally and to free users.

Once enabled, Claude 3.7 Sonnet—Anthropic’s latest AI model—automatically searches the web, providing real-time, sourced information directly within responses.

“When Claude incorporates information from the web into its responses, it provides direct citations so you can easily fact-check sources,” Anthropic explained.

Initial tests show the feature is promising but inconsistent. Claude sometimes pulls relevant information with accurate citations from reputable news and social media sources. Other times, particularly for current events, the feature struggles to engage.

Anthropic previously avoided web integration, positioning Claude as a "self-contained" AI. But competitive pressures likely drove the pivot toward integrating external web searches.

This move brings Claude’s functionality up to speed with popular chatbots from OpenAI, Google, and Mistral.

However, integrating web search introduces familiar risks. AI chatbots often "hallucinate," presenting inaccurate or fabricated information.

Recent studies highlight this issue. According to research from the Tow Center for Digital Journalism, major chatbots like ChatGPT and Gemini incorrectly answer over 60% of questions. The Guardian similarly reported that ChatGPT's web-search tool produced misleading summaries.

While Anthropic’s Claude now matches competitors on paper, users will ultimately decide if its web-search capabilities deliver accurate, trustworthy insights—or simply amplify AI’s current shortcomings.

Apple Sued for Promising AI Features That Never Arrived

Apple faces a federal class-action lawsuit accusing it of false advertising over delayed Apple Intelligence AI features.

Filed Wednesday in San Jose, the suit claims Apple misled consumers by heavily promoting advanced AI capabilities that weren't ready at launch. Plaintiffs seek unspecified damages, alleging the company created unrealistic expectations to boost iPhone sales.

“Apple's advertisements saturated the internet, television, and other airwaves,” the complaint argues. This widespread marketing led buyers to believe AI upgrades, like an enhanced Siri, were immediately available.

Yet Apple recently confirmed major Apple Intelligence features are significantly delayed. The lawsuit specifically cites a September ad featuring actor Bella Ramsey promoting AI capabilities that still haven't shipped.

Apple quietly pulled this ad from YouTube, the lawsuit notes. However, plaintiffs argue Apple hasn't done enough to correct ongoing consumer confusion or compensate buyers.

The lawsuit represents further pressure on Apple amid struggles to deliver promised AI advances. CEO Tim Cook reportedly "lost confidence" in Apple's AI head, John Giannandrea, according to Bloomberg.

This case comes from Clarkson Law Firm, known for suing tech giants like Google and OpenAI over AI-related practices. Unlike previous cases involving AI that underperformed, Clarkson here targets AI capabilities Apple advertised but failed to provide.

Apple has yet to publicly respond. This legal challenge adds to Apple's growing list of headaches as it attempts to regain momentum in the AI space.