Alibaba released its Qwen 3.5 Omni LLM with support for more than 10 hours of audio input. The omnimodal LLM supports understanding of text, images, audio and audio-visual content.

OpenAI is not only losing the buzz race against Anthropic it's increasingly seeing its missteps chronicled.

Today's episode of OpenAI is in a tailspin includes:

• WSJ's detailed account on the demise of Sora and how the Disney deal never really happened.
• Bloomberg notes that ChatGPT's app store has more than 300 app integrations, but adoption is sluggish.
• Microsoft launches a new Critique tool for its Researcher agent where Anthropic grades OpenAI.

US PC shipments grew 3% in the fourth quarter as buyers procured systems for Windows 11 commercial refreshes and stocking up ahead of higher memory and storage prices, according to Omdia. In 2026, PC shipments are expected to decline 13% due to memory and storage shortages and the higher prices that go with them.

Uber said it will acquire Blacklane in a move that expands its premium offerings to luxury and executive travel.

Blacklane operates in more than 500 cities across 60 countries. Uber recently launched Uber Elite to get into the chauffeur sector.

Palantir said it inked a new 5-year deal with Stellantis to support its AI and data initiatives. Palantir said Stellantis will broaden its use of Palantir Foundry and Palantir AI Platform (AIP). Stellantis has already established its data ontology in Foundry as the two companies have worked together since 2016.

• Palantir delivers strong Q4, sees 2026 US commercial revenue surge ahead
• Palantir and Nvidia build sovereign, on-premises AI reference architecture

Stanford published a study in Science on LLMs that found that AI affirmed users' actions 49% more often than humans on average including in cases involving deception, illegality and other harms.

The short version from researchers is here, but the paper is a better and more nuanced read.

The conclusion:

"AI sycophancy is not merely a stylistic issue or a niche risk, but a prevalent behavior with broad downstream consequences. Although affirmation may feel supportive, sycophancy can undermine users’ capacity for self-correction and responsible decision-making. Yet because it is preferred by users and drives engagement, there has been little incentive for sycophancy to diminish. Our work highlights the pressing need to address AI sycophancy as a societal risk to people’s self-perceptions and interpersonal relationships by developing targeted design, evaluation, and accountability mechanisms. Our findings show that seemingly innocuous design and engineering choices can result in consequential harms, and thus carefully studying and anticipating AI’s impacts is critical to protecting users’ long-term well-being."

A guilty pleasure of mine is reading these prompt primers that can use Anthropic Claude to turn you into Warren Buffett, the best CEO ever and the most productive person on the planet.

The latest one from Nav Toor on X may have triggered me. This one revolved around automating your entire workday with 12 prompts apparently based on consultant best practices (which by the way often lead you into a ditch).

The prompts instruct Claude to act like McKinsey, Bain, Goldman Sachs, Accenture, PwC and Deloitte with a dash of big tech culture from Amazon, Google and Salesforce. Oh and don't forget the EY "end of day shutdown routine generator."

Now I suppose if you ran these prompts you could fake like you knew what you were doing for a while with a big caveat. If you knew what the problem was and what data you needed these prompts could be handy. Most of the hard work is figuring out the problem. It's far more likely that you'd do the prompts without knowing the problem, context and data and sound something like a robot. If we all wind up sounding like a management consultant you know what happened. The new "!" and "superexcited" and " —" is going to be sounding like an executive coach.

Anyone want to give it a whirl and let me know how it goes?

Crusoe said Microsoft will lease 900 MW in data center capacity with its first building expected by mid-2027.

Crusoe outlined the deal in a press release. This expansion of the Crusoe's Abilene data center was reportedly turned down by Oracle and OpenAI. The company said:

"The new campus is located adjacent to Crusoe's existing Abilene AI factory infrastructure, bringing the total projected capacity across the full Abilene site to 2.1 gigawatts (GW). Land clearing and site preparation for the new campus are already underway, with the first building expected to be energized in mid-2027, continuing Crusoe’s record-setting pace that energized the first two buildings of the Abilene campus in under one year."

Turns out Microsoft will be the new neighbor of OpenAI and Oracle.

Google is going after OpenAI and Anthropic subscribers with an import tool that will bring all of your past conversations into Gemini. The feature, outlined in a blog, highlights how the competition for LLM subscribers is picking up.

In the big picture, it's pretty clear that Gemini smells OpenAI's blood in the water. First, Anthropic has closed the OpenAI gap, landed subscribers and has model momentum with Claude and its various agentic features rolling out. Meanwhile, OpenAI has ended the Sora app and has had a series of missteps. OpenAI is trying to refocus after spreading itself too thin.

Meanwhile, Apple, who is also in camp Gemini on the back end, will reportedly make Siri model agnostic. And there goes another exclusive for OpenAI.

UnitedHealthcare launched Avery, an AI assistant that responds in natural language with personal coverage and claims details, cost estimates and scheduling help. Avery can pull up real-time insights and case histories.

Avery will be available through the Myuhc.com site or UnitedHealthcare app. The returns for patients will be faster navigation through a morass of healthcare documentation and processes. For UnitedHealthcare, Avery can help cut costs on customer service calls. UnitedHealthcare said 90% of members who use Avery haven't required human assistance.

UnitedHealth Group, parent of UnitedHealthcare, plans to invest $1.6 billion in AI this year.

Salesforce said it has implemented agentic operating systems, powered by Agentforce and Slack, at the Veterans Health Administration (VHA) and Department of Labor.

For the VHA, which serves 16 million to 18 million vets, Salesforce said it deployed its data insights, integrated services and healthcare offerings into an AI-powered operating system in Slack.

The Department of Labor launched DOLA (Department of Labor Agent) built on Agentforce for citizen support and service and resolving cases.

Home buyers can now pledge bitcoin and other cryptocurrencies when making a down payment on a house. Don't worry, Fannie Mae says it's fine.

The Wall Street Journal reports that Fannie Mae's crypto move means you'll be able to pledge your crypto without selling. Coinbase and mortgage company Better Home & Finance are backing the new mortgage product.

This sounds swell until your crypto goes the wrong direction. This crypto-backed mortgage plan is similar to pledging stock for a down payment. It works. Until it doesn't.