This Week in AI: July 1

This week we featured two stories on the ever-growing concern about the power demands of modern computing and especially of AI. Vaire Computing and Phaidra, two companies, one taking action to ensure computing is as energy-efficient as possible, and another working to ensure that computing remains sustainable and energy-efficient in the future, announced their respective funding rounds.

Vaire Computing, a startup founded by Rodolfo Rosini and Hannah Earley, raised $4 million to develop reversible computing technology, which aims to drastically reduce energy consumption and heat generation in computer processors by allowing computations to run in both directions.

Phaidra, the developer of an AI-based autonomous control system that uses reinforcement learning to improve energy efficiency in data centers, secured a $12 million investment led by Index Ventures, which brings its total raised capital to $60.5 million.

Then, as the week went on, ethical concerns related to generative AI continued to emerge left, right, and center. Just last week, Figma announced Figma AI, a suite of AI tools, powered by off-the-shelf, third-party LLMs. One of these features is Make Designs, which is meant to generate a first draft of UI layouts and component options from user-provided text prompts. Then, in what seemed record time, Figma announced it would pause the feature after one of its early testers generated what essentially looked like Apple's Weather app.

But this week's most controversial story was about to break, as several sources started picking up on a series of articles published by WIRED, sharing their investigation into Perplexity AI's questionable web scraping practices. Perplexity has already come under fire when Forbes reported that some of Perplexity's "curated" stories were uncomfortably similar to the source materials from Forbes, CNBC, and Bloomberg, and poorly attributed to them. At the time, Perplexity acknowledged the need to improve its attribution mechanisms but defended its stance on its right to profit off summarized material from the internet.

Not long after the incident with Forbes, developer Robb Knight realized Perplexity AI seemed to have access to sections of his and Macstories' websites, despite these being forbidden to bots using the Robots Exclusion Protocol and server-side blocking. WIRED soon published a story confirming Knight's discovery that Perplexity AI had been using anonymous crawlers that used IP addresses that could not be traced back to the company, essentially ignoring the Robots Exclusion Protocol, an industry-standard method of signaling to crawlers that a website does not want its content to be accessed or leveraged.

WIRED went on to perform a thorough investigation on the reach of Perplexity's unethical scraping, which went as far as to ask the Perplexity AI chatbot to summarize WIRED's (blocked) first article about Perplexity's questionable scraping, which it promptly did. The WIRED investigation and the waves it made were reportedly enough to make Amazon open an investigation into Perplexity for terms of service violations and to seek Perplexity CEO Aravind Srinivas' comment on the situation. Srinivas promptly took a page from the tech company playbook, offloading as much responsibility to others, including the alleged third-party providers of the scraping service, WIRED, and users who expected the Perplexity chatbot not to hallucinate.

In a bold and ingenious move, Cloudflare made the most of the controversy by announcing that it would introduce a new feature allowing all users, including those on the free tier, to easily block all AI bots by leveraging its advanced detection methods for suspicious bot-like behavior. As a proof of concept of sorts, Cloudflare ran an analysis on the traffic belonging to Perplexity AI's covert bot to showcase the capabilities of its newest machine learning model powering the bot-blocking feature. Amid a series of complaints that, in the best-case scenario, resolve themselves as discreet settlements, Cloudflare's announcement felt like a refreshing opportunity for users to fight back, at least temporarily.

Other notable headlines for the week include:

The European Commission found Meta's subscription model non-compliant with the DMA: The European Commission has deemed Meta's "pay or consent" model non-compliant with the Digital Markets Act. Meta could be forced to introduce an equivalent alternative to its targeted ads-based services or risk paying a hefty penalty.

Meta modifies its media labeling strategy after real photos were tagged as 'Made with AI': Meta has renamed its "Made with AI" label to "AI Info" after facing issues where labeled real photos gave the impression that they were AI-generated. The new AI Info label encourages users to click for additional context so users can receive accurate information about AI usage in the labeled images.

Anthropic launches an initiative to fund third-party AI evaluations: Anthropic has launched an initiative to fund third-party evaluations of advanced AI models, focusing on safety assessments, capability metrics, and evaluation infrastructure to enhance transparency and safety across the AI industry.

Meta showcased its research into 3D asset generation: Meta has developed a fast, high-quality 3D asset generation pipeline called Meta 3D Gen, combining text-to-3D mesh and texture generation models, which outperforms alternatives and is likely intended to support content creation for Meta's virtual reality platform.

Altrove secured €3.7M to accelerate essential materials' discovery and production: Altrove, a startup founded in 2024, has raised €3.7 million to develop sustainable alternatives to rare earth elements using AI-driven material recipe optimization and characterization, differentiating itself by focusing on creating and testing materials rather than just predicting them.

Salesforce's xLAM-1B shows competitive performance in function-calling tasks: Salesforce has developed two compact AI models, xLAM-1B and xLAM-7B, using an innovative APIGen pipeline for high-quality data curation, resulting in models that outperform much larger competitors in function-calling tasks and show promise for on-device AI applications.

Squirro acquired Synaptica to integrate graph technology into its generative AI platform: Swiss-based Squirro has acquired US company Synaptica, combining generative AI with knowledge graph technology to enhance enterprise AI solutions, particularly in improving the accuracy and capabilities of Retrieval Augmented Generation (RAG) for mission-critical applications.

ElevenLabs launched a Voice Isolator tool: ElevenLabs has launched a free AI Voice Isolator allowing creators to remove ambient noise from various content types, producing studio-quality speech output with promising initial results and plans for API access and real-time streaming support.

Roboat secured €550K for its AI-powered system for urban and inland waterway navigation: Roboat, an MIT and AMS spin-out, has developed AI-powered autonomous vessel technology for various applications, including ferries and water taxis, and recently secured €550,000 in funding to expand its presence in the autonomous shipping market.

Female Foundry Launches Visionaries AI Incubator with Google Cloud: Female Foundry has partnered with Google Cloud to launch the Visionaries AI Incubator, a six-week program to accelerate growth for innovative, AI-focused female-led startups in Europe from Pre-Seed to Series A stages.

Cartken raised $22.5 million to take its delivery robots from the sidewalks to indoor settings: Robotics startup Cartken is carving out a unique niche in the autonomous vehicle industry with its versatile delivery robots that can operate seamlessly in indoor and outdoor environments, attracting significant investor interest and finding applications across various sectors.