OpenAI claims ChatGPT’s new default model hallucinates way less

Taille de police

OpenAI's newest default model for ChatGPT might not make stuff up as much. Hallucinations have been an ongoing problem for AI models, but OpenAI says its new GPT-5.5 Instant model has "significant improvements in factuality across the board." The company claims that, based on "internal evaluations," GPT-5.5 Instant produced "52.5% fewer hallucinated claims" than its […]

Continue reading on The Verge

This article was originally published by The Verge.

Articles liés

En développement·9 sa önce

Amazon's Mechanical Turk to Close to New Customers in July 2026

Amazon's Mechanical Turk, a crowdsourcing service launched in 2005, will close to new customers on July 30, 2026. While existing users can continue, Amazon Web Services will not introduce new features, signaling the service is on life support.

TechCrunch

Tech·13 sa önce

Sourdough Sidekick Review: A Gadget for Dedicated Bakers

The Sourdough Sidekick automates sourdough starter management, feeding it flour and water on a schedule. While it works well for frequent bakers, design quirks like mandatory large discard amounts and noisy operation limit its utility for casual users.

The Verge

En développement·14 sa önce

Vizio's Mini LED Quantum TV is the best 'dumb TV' you can buy

Vizio's new Mini LED Quantum TV, acquired by Walmart, can be used as a 'dumb TV' by bypassing its Vizio OS and Walmart account requirements, offering great performance for its budget price.

The Verge

En développement·15 sa önce

Microsoft's Xbox Division Faces Existential Crisis Amidst Costly Missteps

Microsoft's Xbox division is in crisis, with reports of layoffs, studio closures, and game cancellations following costly, unsuccessful bets on subscription services like Game Pass. Despite billions spent, revenue has declined, leading to a strategic "reset" under new leadership.

The Verge

En développement·1 g önce

Alibaba Bans Employee Use of Anthropic's Claude Code AI Tool

Alibaba will ban employees from using Anthropic's AI programming tool Claude Code starting July 10, classifying it as high-risk. This follows Anthropic's own restrictions on Chinese companies and efforts to close access loopholes, including an experimental feature to identify Chinese users.

TechCrunch

En développement·1 g önce

Mistral AI: The French AI Darling's Misunderstood Strategy

Mistral AI, a French AI company, is often misunderstood. While not aiming to be Europe's OpenAI, it follows a Palantir-like strategy of deploying AI for governments and corporations. Despite lower funding than US rivals, its revenue is rapidly growing, and it's investing in research and infrastructure.

TechCrunch