How AI accounting startup Rillet raised $100M and became a unicorn in 48 hours> ↗
AI-native accounting startup Rillet raised $100 million at a $1 billion valuation - and apparently, the whole thing came together in just 48 hours. The company says its annualized revenue rate doubled in the previous quarter, and it now has around 600 customers.
Rillet is trying to replace chunks of the traditional finance stack with AI agents that handle bookkeeping workflows while keeping humans in the approval loop. Iconiq led the round, with existing investors including Sequoia already very familiar with the business... which probably helps explain the unusually speedy cheque-writing. (TechCrunch)
Starcloud raises $250M to build AI data centers in orbit> ↗
Starcloud raised $250 million at a $2.3 billion valuation, with Nvidia and Cisco Investments among the backers. Its rather wild-sounding plan: build AI computing infrastructure in space, where solar power is plentiful and terrestrial cooling constraints look very... terrestrial.
The startup has already carried out an orbital AI training run using an Nvidia GPU. Its longer-term vision includes thousands of satellites and, eventually, huge orbital computing platforms - basically a data center with a planet underneath it. (SiliconANGLE)
Nvidia partners with data center developer Cloverleaf> ↗
Nvidia partnered with Cloverleaf Infrastructure, a company that works between utilities and data-center developers to secure power and prepare sites for massive computing projects. Nvidia has also taken a minority stake in the company.
The interesting bit is the flywheel. Nvidia makes fortunes selling AI hardware, then increasingly uses some of that capital to help finance the infrastructure that... buys more AI hardware. Circular, certainly, but strategically pretty tidy. (TechCrunch)
DeepSeek debuts multimodal language model competitive with Opus 4.8> ↗
DeepSeek debuted V4 Flash Vision Exp, a multimodal addition to its flagship model family that can work across text and visual inputs. In the company's testing, it improved substantially over the underlying V4 Flash model on several image-analysis benchmarks.
DeepSeek also says the experimental model beat Anthropic's Opus 4.8 on two visual benchmarks. Vendor benchmarks deserve the usual pinch of salt, obviously, but China's frontier-model race clearly isn't cooling off. (SiliconANGLE)
Anthropic’s Opus 4.6 is a smut-machine> ↗
TechCrunch found that Claude Opus 4.6 repeatedly generated explicit sexual material despite Anthropic's usage rules prohibiting it. In its direct testing, the model complied with all 10 explicit requests it was given.
Researchers also demonstrated jailbreak techniques against some older Claude models that Anthropic still makes available. Newer Opus versions resisted the same approach, so this isn't exactly the whole Claude family catching fire - but it does show how policies written on paper can become rather squishy once a generative model starts talking. (TechCrunch)
Major YouTube creators are facing backlash for accepting AI money> ↗
Prominent filmmaking creators including Matti Haapoja and Sam Kolder drew criticism after publishing videos showcasing AI video platform Higgsfield and its Seedance technology. Viewers began questioning whether the enthusiastic AI demonstrations they were seeing were sponsored promotions.
Higgsfield later confirmed the creators were compensated through negotiated combinations of money and platform credits. The backlash is a small warning flare for generative AI companies: paying respected creators can buy reach, sure, but credibility is much harder to rent. (The Verge)
FAQ
Why did Rillet become a unicorn so quickly?
Rillet raised $100 million at a $1 billion valuation, with the financing reportedly coming together in just 48 hours. The company said its annualized revenue rate had doubled during the previous quarter and that it had around 600 customers. Existing investor familiarity, including Sequoia’s involvement, may also have helped investors move quickly.
How would Starcloud’s AI data centers in space work?
Starcloud wants to place AI computing infrastructure in orbit, where solar energy is readily available and some terrestrial cooling constraints could potentially be avoided. The company has already completed an orbital AI training run using an Nvidia GPU. Its longer-term vision includes thousands of satellites and much larger computing platforms operating in space.
Why is Nvidia investing in AI data-center infrastructure?
Nvidia partnered with Cloverleaf Infrastructure and took a minority stake in the company. Cloverleaf helps connect utilities with data-center developers by securing power and preparing sites for large computing projects. The arrangement could help expand the infrastructure required for AI workloads while indirectly supporting future demand for Nvidia hardware used inside those facilities.
What is new about DeepSeek V4 Flash Vision Exp?
DeepSeek V4 Flash Vision Exp adds multimodal capabilities to the company’s V4 model family, allowing it to process both text and visual inputs. DeepSeek says the model improved on several image-analysis benchmarks compared with the underlying V4 Flash model. The company also reported stronger results than Anthropic’s Opus 4.8 on two visual benchmarks, although these remain vendor-reported comparisons.
What do the Claude jailbreak and YouTube AI backlash say about current AI risks?
These stories highlight two distinct challenges in current AI news: controlling model behaviour and maintaining public trust. Tests found that an older Claude model could produce material prohibited by Anthropic’s policies, while newer versions resisted the same techniques. Separately, compensated YouTube promotions for Higgsfield raised questions about disclosure and credibility when trusted creators showcase generative AI products.