A start-up collaborating with Apple and backed by Nvidia's venture arm has launched what it calls the world's first "fully autonomous" platform that applies generative AI to heating and cooling systems.

Why it matters: Americans spend around $190 billion a year powering buildings — generating around one-third of total U.S. carbon emissions — and the company claims it can cut building energy costs by around one-third.

Details: PassiveLogic says its customizable "generative autonomy" systems manage building infrastructure from a few hundred sensors and controls for small buildings to 1 million for large commercial towers.

• Shifting buildings to autonomous energy systems could slash bills and make energy managers the center of gravity in the market, instead of energy providers.
• Buildings are the biggest source of wasted energy in the country, PassiveLogic founder and CEO Troy Harvey tells Axios.
• Harvey insists that most "smart" thermostats don't qualify for the label.

Between the lines: PassiveLogic claims that buildings need to become autonomous in order to achieve major energy efficiencies because "existing solutions just don't scale."

• The company's "Quantum Lens" technology can deliver these gains without the need to retrofit items like windows and insulation or install big equipment like heat pumps.
• The generative AI works alongside open standard "digital twin" technology — a digital simulation of a real life object — to predict a building's operational needs and then optimize the delivery of energy, with an efficiency humans cannot match on their own.
• Each building owner can assemble a tailored autonomous system out of pre-trained modular AI components, trained on digital twins of common building systems.
• PassiveLogic collaborated with Apple and assembled what it calls the world's fastest AI language to speed up the training of its AI.

The intrigue: Every building is different — and multiple systems and competing interests add to the complexity and cost of achieving energy efficiency.

• The systems can range from heating and cooling and ventilation (HVAC) to energy generation and storage in many different occupancy zones.
• The interests range from residents or workers who want different temperatures, to suppliers working to maximize what the building spends on energy and maintenance.

Yes, but: It will be at least another year before PassiveLogic has the capacity to expand supply from today's large customers to single family or even mid-sized residential buildings.

The other side: Incumbent players including Honeywell, Siemens, Schneider Electric and Johnson Controls already offer services such as 24/7 data monitoring and automated fault detection in HVAC systems, dominating a market worth around $75 billion each year.

• Honeywell is at an "early stage" of testing generative AI and the more dynamic systems it could support, the company's chief product officer Manish Sharma told Axios.
• Sharma said there is plenty of scope with current technology — like its Honeywell Forge system — to cut emissions by upgrading old HVAC systems.

What they're saying: Harvey compares the building world to the auto industry before Tesla: stagnant and needlessly complex.

Elon Musk has launched a new chatbot called "Grok" out of his xAI startup, claiming that AI should be "useful to people of all backgrounds and political views."

Driving the news: xAI spent four months training Grok, which it categorizes as a "frontier large language model" — which would make it subject to new AI testing requirements both from the White House and those agreed upon at last week's AI safety summit in London.

X claims Grok is "a very early beta product" with a sense of humor and "a rebellious streak" — but since hardly anyone has access to the service currently, it's hard to test the claim.

Between the lines: The xAI team says the "fundamental advantage of Grok is that it has real-time knowledge of the world via the X platform."

Context: The team that built Grok includes alumni of DeepMind, OpenAI, Google and Microsoft.

Ina's thought bubble: The characterization of Grok could add a political dimension to the AI market, with customers evaluating not just how accurate AI is, but also how much they like the politics of the answer.

Just two states — California and Oregon — have issued policy guidance for schools on artificial intelligence platforms such as ChatGPT, reports Axios' Jennifer Kingson.

Why it matters: Teachers and administrators are eager for guidelines on how to use AI — and how to quash misuse — but polling shows state governments have been loath to issue pronouncements.

Driving the news: The Center on Reinventing Public Education (CRPE), a nonpartisan research center at Arizona State University, asked each of the 50 states and the District of Columbia to share their approach to AI guidance.

• In addition to California and Oregon, 11 states are developing guidance: Arizona, Connecticut, Maine, Mississippi, Nebraska, New York, Ohio, Pennsylvania, Virginia, Vermont and Washington.
• 21 states said they didn't plan to offer guidance for the foreseeable future, and 17 states didn't respond.

Details: It's hard for educators and regulators to know what to tackle first: from teaching AI in classrooms to using AI for school operations (scheduling, ordering supplies) to cracking down on plagiarism and kids using it to do homework.

• California and Oregon's guidance was published after CRPE first polled on this topic in August.
• Bias is a big concern, given that generative AI programs are trained on models that can filter information in wrong or inappropriate ways.

The intrigue: There's a split between states that plan to issue AI guidance and those that want to let individual districts decide their own policies.

What they're saying: Superintendents have tended to tell CRPE that "they're a little overwhelmed by AI," Bree Dusseault, principal and managing director at CRPE, tells Axios.

• "They are really heads-down dealing with post-pandemic recovery, learning loss and mental health, and the introduction of AI has felt like another big question that they would like help with," she said.

Yes, but: There are guidelines out there for those who seek them out.

• Code.org, an education nonprofit, has put out a fairly comprehensive toolkit with sample guidance on the use of AI.

Context: President Biden's recent executive order on AI addresses education.

• At the same time, private sector groups like InnovateEDU and its EdSafe AI Alliance are working to advance the responsible use of AI in education.

Be smart: AI technology is advancing so quickly that any guidelines for schools must be purposely vague and flexible, lest they grow obsolete quickly.

• Teachers must emphasize that AI is "not an inherently bad tool" but one that can be used to teach critical thinking, said Noelle Ellerson Ng, associate executive director, advocacy and government, for AASA, the School Superintendents Association. Students must learn to "evaluate, is this a valid source or an invalid source?"

• As Treasury Secretary Janet Yellen prepares for two full days of meetings with her Chinese counterpart in San Francisco this week, she writes that U.S. restrictions on trade with China are aimed at China's efforts to advance "highly sensitive technologies that are critical to the next generation of military innovation" — in other words, cutting-edge AI and semiconductors. (Washington Post)
• Tech industry companies added more than 2,000 jobs in October, per CompTIA's analysis of latest federal jobs report data, with about 20% of job postings offering work from home or remote work as an option.
• WhatsApp's AI image generator displays cartoons of kids with guns when some users entered the prompt "Muslim boy Palestinian," while prompts for "Israeli boy" or "Israel army" returned drawings of children playing soccer and reading or cartoons of smiling soldiers without guns. (The Guardian)

Here's what happened when Julia Louis Dreyfus used ChatGPT to write an award acceptance speech.