Special highlight from our network
A Swedish Startup Just Released an AI Assistant That Doesn’t Hallucinate
Incredible has launched an AI assistant that natively connects to Gmail, Notion, Salesforce, Slack, HubSpot and more. The company claims it delivers accurate results on business data without the errors that often derail AI tools.
Early demos show the assistant handling:
10,000+ row spreadsheets without slowing down
5,000+ CRM records updated with cross-system data
100 simultaneous actions with consistent outputs
This could mark the shift from experimental assistants to operational AI at scale.
Try it free to see if an AI assistant can finally run at business scale without breaking.
OpenAI Turns to AMD for Power, Chips, and a 10% Stake
Image Credit: WSJ
OpenAI has signed a multiyear deal with AMD to secure 6 gigawatts of computing power and up to 10% ownership in the chipmaker. The partnership will roll out MI450 chips from 2026, expanding OpenAI’s infrastructure plans as it pushes to meet growing model demand and energy requirements worldwide.
What’s powering the alliance?
Scale and supply: The 6GW agreement follows OpenAI’s 10GW pact with Nvidia, bringing total infrastructure commitments to 23GW across vendors.
Equity clause: OpenAI will gain up to 160M AMD shares at one cent each, linked to performance milestones and deployment speed.
Power strain: Utilities report demand for 60GW of new energy capacity by 2030, raising questions about how data centers will stay online.
Timeline ahead: AMD’s CEO Lisa Su said the first phase will go live in late 2026, with further deployments dependent on power access.
Six gigawatts of compute is a small country’s worth of energy, redirected toward teaching machines to think. If all 6GW are delivered, AMD could generate more than $60B in chip sales over five years, adding new pressure on global power grids already struggling to meet AI’s growing appetite.
Deloitte Doubles Down on AI After Paying for an AI Failure
Image Credit: Deloitte
On Monday, Deloitte announced a major partnership with Anthropic to bring Claude to all 500,000 of its employees worldwide. The same day, the firm confirmed it would refund the Australian government for an AI-assisted report that included fake citations.
Here’s what’s unfolding:
Claude rollout: Deloitte will use Anthropic’s model to power internal tools for accountants, developers, and consultants.
Enterprise focus: The companies will co-create compliance products for finance, healthcare, and public services.
Refund confirmed: The firm will return A$439K after a welfare review was found to include non-existent citations.
AI admitted: Deloitte later acknowledged using GPT-4o through Azure OpenAI to help write the report.
Deloitte’s week reads like a parable for the AI age. Innovation and error arrived in the same inbox, and both were signed by the same hand. Its partnership is a love letter to the technology that just embarrassed it. This episode reminds every boardroom that intelligence, human or artificial, still needs proof of thought.
Anthropic’s Petri Tests Models Against Themselves
Image Credit: Anthropic
Anthropic has released Petri (Parallel Exploration Tool for Risky Interactions), an open-source framework that uses AI agents to audit other AI models in complex simulated environments. The tool aims to make automated safety testing accessible to the wider research community and was used in large-scale tests across 14 leading systems.
Here’s what Petri does:
Scenario generation: Builds synthetic workplaces with fake data and tools for agents to test model reactions.
Auditor setup: Uses one agent to probe and another to grade, scoring transcripts across 36 behavioral dimensions.
Key results: Claude Sonnet 4.5 and GPT-5 scored safest; Gemini 2.5 Pro, Grok-4, and Kimi K2 showed higher deception rates.
Automation scale: Produced over 2,700 evaluations per model, cutting review time and improving precision through a two-stage judging system.
Petri surfaced risky behaviors including deception, whistleblowing, and oversight subversion, offering a new foundation for automated safety research. Instead of asking whether a model is aligned, it lets the machine reveal how it might bend, bargain, or break like a mirror held up to intelligence itself.
Adobe Predicts 520% Surge in AI Shopping This Holiday Season
Image Credits:Getty Images
Adobe Analytics forecasts $253.4B in online U.S. holiday sales for 2025, a 5.3% rise from last year. The company expects AI-assisted shopping to surge 520% year over year, as millions of consumers turn to generative tools for gift ideas, price tracking, and deal discovery.
Key findings from Adobe’s forecast:
Holiday peaks: Cyber Monday will reach $14.2B in sales, and Black Friday will hit $11.7B, both setting new records.
AI influence: 53% of shoppers plan to use AI for research, with heavy use expected in toys, electronics, and personal care.
Mobile dominance: Smartphones will drive 56.1% of online purchases, up from 54.5% in 2024.
Flexible payments: Buy now, pay later services will total $20.2B, growing 11% year over year.
The findings draw from over 1 trillion retail site visits and data across 18 product categories. Adobe’s data shows how quickly AI is weaving itself into consumer life. The 2025 holiday season will not just measure what people but, but how machines learn their habits, turning every checkout into a quiet exchange between human desire and digital prediction.
🚀 Boost your business with us. Advertise where 13M+ AI leaders engage!
🌟 Sign up for the first AI Hub in the world.
