What are the opening hours?

Monday 8am - 5pm Tuesday 8am - 5pm Wednesday 8am - 5pm Thursday 8am - 5pm Friday 8am - 5pm

FinFuse, Leawood, KS (2026)

06/23/2026

UN chief urges AI companies to ‘come clean’ about the pollution they generate: United Nations Secretary-General António Guterres on Tuesday called on AI companies to release information about the carbon pollution they create, along with the water and land used to power their operations.

In an address at London Climate Action Week, Guterres proposed the AI Environmental Transparency Initiative, arguing artificial intelligence companies should measure and disclose the impact of their technology—impact that opponents cite as a reason to curb the rapid growth of data centers. Both national governments and local authorities in areas with data centers that support AI are exerting pressure on these companies for more transparency and more standardized reporting across the industry.

Guterres said AI companies should also commit to powering their facilities with electricity produced with renewable technologies, such as wind and solar, by 2030.

“No more hidden costs,” Guterres said at Europe’s largest independent climate conference. “No more shifting the burden onto those least able to bear it. It is time to come clean.”

AI’s needs are growing

Many big tech companies have vowed to power their operations using cleaner sources, some by the end of the decade. Some plan to do so especially using solar and nuclear, including Amazon and Google.

But the race to deploy AI has complicated those commitments and increased greenhouse gas emissions, which come from the burning of fuels like oil, coal and gas, and heat the planet. Regulatory barriers have also hindered climate-friendly projects.

Coal provides about 30% of the electricity consumed by data centers globally, according to the International Energy Agency. Renewable energy—primarily wind, solar, and hydro—supplies about 27%; natural gas, 26%; and nuclear, 15%. Renewables are expected to meet just half of the demand over the next five years.

As AI booms, many, including Guterres, have touted its ability to accelerate climate solutions. It could improve energy efficiency and reduce pollution and emissions.

At the same time, the environmental footprint of data centers already rivals some of the world’s largest countries, according to a U.N. report released earlier this month.

That report also said the water and energy use and pollution associated with AI will double in just four years. Data centers needed to fuel AI accounted for about 1.5% of the world’s electricity consumption in 2025, and will account for nearly 3% of projected electricity use by 2030.

“Despite these obvious concerns, communities are often left in the dark about the environmental impact of the infrastructure rising around them,” Guterres said in his remarks.

The UN continues to sound urgent alarms

The U.N. chief has long urged the world to take serious climate action, and will once again convene leaders at the annual Conference of Parties, this year in Turkey, to negotiate plans.

On Tuesday, addressing AI was just a number of steps he said needed to be taken to keep the world below the warming limit of 1.5 degrees Celsius (2.7 degrees Fahrenheit) above preindustrial times, a goal set by the 2015 Paris Agreement.

Last year was the first time that the three-year temperature average broke through that threshold.

“Every major emitter must accelerate action,” Guterres said. “And every country must overdeliver on its commitments.”

He called for cutting methane, a powerful greenhouse gas that is responsible for around one-third of global warming and is significantly more potent than carbon dioxide, though it doesn’t linger as long in the atmosphere. He also called for a reduction in dependence on coal, oil, and gas.

Renewables progress seen around the globe but challenges remain

Guterres noted positive developments in renewable energy, as scale drives down the costs of the technologies and adoption increases.

Clean power generation—largely driven by solar and wind—exceeded overall global electricity demand growth last year. The share of renewables also hit more than one-third of the world’s electricity mix for the first time in modern history in 2025, and coal power saw its share fall below one-third of global generation.

China continues to drive the world’s clean energy transition, and in Europe, fossil generation is generally trending down.

But the United States under President Donald Trump has embraced coal, oil, and gas and slashed support for renewables and broader climate action—all amid the global energy crisis exacerbated by the U.S. war in Iran, which Guterres called “the mother of all energy shocks.”

Guterres referred to the state of the world as “A Tale of Two Crises,” a nod to the Charles Dickens novel, “A Tale of Two Cities.”

“For the climate agenda, this is indeed the best of times and the worst of times,” he said. “The worst—because climate impacts are intensifying, tipping points are looming, and the energy crisis has exposed the deep risks of dependence on fossil fuels. But also the best—because the renewables revolution is well underway.”

---

Alexa St. John is an Associated Press climate reporter. Follow her on X: . Reach her at [email protected].

---

Read more of AP’s climate coverage.

---

The Associated Press’ climate and environmental coverage receives financial support from multiple private foundations. AP is solely responsible for all content. Find AP’s standards for working with philanthropies, a list of supporters and funded coverage areas at AP.org.

—Alexa St. John, Associated Press

http://dlvr.it/TTB1r0

06/22/2026

Power for energy-guzzling AI data centers is getting fast-tracked thanks to federal regulators: Federal regulators on Thursday ordered regional grid operators to help large energy users connect more quickly to the nation’s inefficient and aging electric transmission system, a step they said is needed to accommodate surging demand from power-hungry artificial intelligence data centers.

Energy Secretary Chris Wright had urged the Federal Energy Regulatory Commission to act in an effort to help the United States better compete with China for superiority in the fast-growing AI sector.

Tech companies and data center developers welcomed the chance to connect faster to the country’s power supply for the biggest energy users ever built in the United States, including some that consume more electricity than a small city.

Utilities, states and regional grid operators had worried that the Republican administration’s plan would remove their authority to manage the process, but FERC said the order leaves states in control of retail electric rates, terms and conditions. Clean energy advocates have urged regulators not to undermine state-level efforts to require the use of renewable energies.

The commission’s actions come as a backlash grows against data centers over concerns about the massive amounts of energy and water they use and fears about noise and air pollution, water shortages and a loss of open space or farmland.

Unanimous vote and affordability

FERC members voted unanimously to direct six regional grid operators to ensure that AI data centers and other large power users are “able to connect to the transmission system in a timely and orderly manner.”

Laura Swett, an appointee of President Donald Trump who chairs the commission, called the vote “historic” and said it would push the country’s electricity market into the future while respecting states’ rights, protecting reliable electric service and shielding ratepayers from shouldering the costs of connecting big power users to the grid.

“I know that Americans across the country are concerned about affordability, and so are we,” Swett said, referring to the five-member commission. As chair, “I am taking extremely seriously the mission that Congress has entrusted us to ensure that rates are reasonable,” she said.

The vote comes eight months after Wright asked the independent agency to take more control over ensuring that the vast network of massive computing warehouses needed to power AI are connected quickly to high-voltage transmission lines.

Wright hailed the commission’s action, saying it would “remove barriers, accelerate development and ensure America has the affordable, reliable and secure energy needed to power a new era of prosperity.”

Data centers would pay the full cost of any grid upgrades needed for their connection, under the commission order. But that order can do little to address the tightening energy supplies that are driving up electricity bills in some areas and raising warnings of blackouts as the construction of data centers outpaces the speed of new power plants coming online to serve them.

Robert Montejo, a lawyer who represents data centers, said the most important message from FERC’s action is that AI “has fundamentally changed the electricity landscape. The grid and prior policy were not built for the pace and scale of demand we’re seeing from AI infrastructure, and FERC is signaling that standing still is no longer an option.”

The six regional grid operators under the order serve 200 million Americans, or two-thirds of FERC’s jurisdiction. FERC, meanwhile, invited utilities that handle their regional transmission systems to also participate and analysts said the agency could eventually pressure them, too.

A search for power

Tech giants are scrambling to find enough power for their data centers and report that, in some places, it will take years to connect to the electric grid.

The Edison Electric Institute, which represents investor-owned electric utilities, said FERC’s order builds on regional and state processes already underway while “supporting flexibility and innovation.”

Besides power bottlenecks, the tech industry is running into widespread opposition from communities where residents don’t want to live next to or near a data center.

More than 4,000 data centers now operate in the U.S., according to one estimate, with an additional 3,000 planned or under construction.

Trump has tried to deflect public concerns about AI, seeing the fast-evolving technology as crucial for the U.S. to attract foreign investment and maintain its economic and military prowess. He signed an executive order this month establishing a framework for the federal government to vet the national security risks of the most advanced AI systems for up to a month before their public release.

In December, FERC took an earlier step to help data center operators get electricity quickly, voting to allow tech companies to effectively plug a data center directly into a power plant and Thursday’s order sought to ensure that option is accessible around the country.

Power demands from data centers

FERC told grid operators to respond within 30 days on how they will ensure there is adequate power supplies for new and future data centers, and within 60 days on plans to integrate large power users in line with the new guidelines. Swett told reporters after the meeting that she hoped faster connection processes are in effect in “as little time as possible.” She didn’t set an exact timeline.

Jeff Dennis, executive director of the Electricity Customer Alliance, said FERC’s order is responsive in particular to big power users and state regulators.

Tech giants are confronting unclear rules to connect data centers to high-voltage transmission systems, while states need more clarity on who should bear the cost of regional transmission projects approved at the federal level, he said.

Rob Gramlich, a Washington-based energy consultant, said states should quickly develop rules to accommodate large power users and prevent cost shifts to residential and business customers. FERC could assert broader jurisdiction over interconnection issues if states don’t act quickly, he said.

Data from the Electric Power Research Institute shows that data centers now account for about 5% of U.S. electricity demand, but could triple by 2035.

Tech companies have continued to raise their spending on building and equipping data centers, but there is evidence that construction is lagging and projects are hitting roadblocks, including permitting delays, growing local opposition or bottlenecks around gas turbines, transformers and skilled labor.

—Matthew Daly and Marc Levy, Associated Press

http://dlvr.it/TT962f

06/21/2026

How the U.S. military is preparing for laser attacks on drones: This article is republished with permission from Laser Wars, a newsletter about military laser weapons and other futuristic defense technology.

The U.S. military has spent billions of dollars over decades building high-energy laser weapons capable of burning drones out of the sky, but it has spent considerably less money and time exploring what happens when an adversary does the same thing. With China fielding a growing arsenal of laser weapons capable of killing drones at ranges up to 25 kilometers, Russia’s Peresvet reportedly in active service, and various laser systems now spreading across the world through indigenous development, proliferation, and a burgeoning export market, that oversight is becoming harder to ignore.

The U.S. military’s answer to this problem has a name: counter-directed energy weapons (CDEW). It’s a nascent field—no dedicated CDEW system is publicly known to have been fielded, and most related research and development remains conceptual. But a 2023 study published in the Journal of Directed Energy by researchers at the U.S. Naval Postgraduate School (NPS) offers the clearest public picture yet of what defending against a laser weapon looks like.

The NPS study—which draws on a comprehensive 2020 NPS systems engineering capstone report by the same team of researchers—is focused specifically on naval unmanned aerial vehicles, and with good reason. Drones are arguably the most exposed military asset in the world: increasingly designed for expendability, they operate in lethal proximity to adversaries and, unlike a destroyer or a tank, carry no meaningful armor. The same principles that makes drones attractive as delivery mechanisms for attritable mass also makes them highly susceptible to a weapon optimized for persistent energy delivery. And while the laser threat calculus the NPS researchers present also applies to manned aircraft, surface ships, missiles, satellites, and ground vehicles, naval drones just happen to sit at the acute end of the vulnerability spectrum.

To understand this vulnerability, the NPS researchers evaluated four representative drones of various sizes: a large Group 5 broad-area maritime surveillance (BAMS) drone (the MQ-4C Triton); a large Group 5 combat drone (Northrop Grumman’s X-47B demonstrator); a rotary-wing Group 4 ISR and fire support drone (the MQ-8C Fire Scout); and a small Group 2 ISR drone (a Small Tactical Unmanned Aerial System from ScanEagle). When faced with a 100 kilowatt laser with no countermeasures in place, three of the four drones were assessed as destroyed after just a few seconds of irradiation. Only the large BAMS drone, operating at extreme altitude and ranges exceeding 8,000 nautical miles from a potential threat, survived thanks to distance alone.

Since lasers bleed energy over distance and through atmospheric interference, altitude and range matter just as much as size. Fast-moving drones are harder to track and target with a sustained beam. Material composition is arguably the most significant factor: a thin composite airframe melts far faster than a thick aluminum one. And in terms of mission profile, a drone loitering at low altitude in a contested littoral is more exposed than one cruising at 60,000 feet over open ocean. To wit, the small Group 2 ISR drone ranked as the most vulnerable of the four drones evaluated in the NPS research, while the BAMS was the safest—but only until it came down to land.

No naval drone (or, for that matter, U.S. military platform) is currently known to be equipped with systems to detect a high-energy laser attack as it occurs; in many cases, the first sign that a laser is being used against you might arrive only during battle damage assessment. That detection gap is the foundational CDEW problem, and everything else flows from it.

The NPS researchers identified five broad categories of CDEW solutions:

* Use the weather: This is the most immediately actionable laser countermeasure, and it costs nothing. Fog, rain, haze, dust, and smoke can all absorb and scatter laser beam photons, reducing the energy that reaches the target. At higher power levels (above 100 kw), even clear air can work against a laser through thermal blooming, where the laser heats the air it passes through and defocuses the beam. The operational takeaway is straightforward: plan missions to exploit bad weather and adverse atmospheric conditions wherever possible. The catch is that you need reasonably good intelligence on where a laser threat is located and what its capabilities are to accurately calculate how much protection the atmosphere actually buys you.

* Warning systems: Sensors like the AN/AVR-2B Laser Detection System (LDS) are already used on some military aircraft to detect laser rangefinders, target designators, and beam-riding missiles. Integrating those systems directly into drone payloads to detect and identify high-energy laser threats could produce something of an early warning system: a drone detects that it is being irradiated, alerts operators and nearby platforms, and triggers either active countermeasures or evasive maneuvers. The challenge is that warning systems have to be matched to the laser’s wavelength to work reliably—and they need to be designed into the platform from the start, not bolted on after the fact.

* Active countermeasures: This category covers four distinct approaches, according to the NPS research. Smoke and aerosol screens—essentially cloudbursts of fine particles dispensed around a drone—absorb and scatter the beam, buying time. Laser jammers analyze the incoming beam, identify the source location and intensity, and fire back a disrupting signal to break the adversary system’s targeting lock. Basic counterfire deploys weapons against the laser system itself if its position is confirmed. Finally, decoy drones acan ct as false targets, drawing the beam away from more mission-critical assets. These approaches range from immediately feasible to technically demanding, and all of them share one requirement: you have to know you’re being lased before you can respond.

* Armor up: Passive shielding is the most engineering-intensive solution, with three distinct materials yielding the most dramatic results in the NPS simulation. Bragg mirrors—dielectric mirrors constructed from alternating layers of two optical materials—can reflect up to 99.99% of laser energy for a specific wavelength, essentially making the beam bounce off. Reflective coatings work on a similar principle and can be applied directly to an airframe, even as a temporary pre-mission treatment matched to a known threat wavelength. Ablative coatings take a different approach: rather than deflecting energy, they absorb it and burn away in a controlled fashion to buy a drone time to escape. In the NPS analysis, Bragg mirror coatings were the single most effective CDEW method tested, protecting all four drone types under the simulated 100 kw threat. But there’s a critical caveat: the mirrors only work at the specific wavelength they’re built for. Use the wrong coating against the wrong laser and you’ve wasted weight.

* Evasive maneuvers: Maneuvering and swarm tactics round out the playbook. Continuous wave laser weapons require sustained contact with a target to inflict damage—break that contact by banking hard, diving, or flying erratically and the required dwell time resets. Swarm tactics extend this principle by flooding the adversary’s engagement capacity: a single laser system can only engage one target at a time, and a swarm forces it to choose and re-engage sequentially. In the NPS simulations, swarm tactics proved the second most reliable CDEW method, protecting drones in roughly three to four out of every five simulated engagements. Evasive maneuvering alone was less reliable, limited in part by the latency inherent in remote control. Onboard autonomous maneuvering, where a drone detects irradiation and evades without waiting for a human command, is a promising direction, and one that applies equally to any remotely operated platform facing a laser threat.

When the NPS team ran their CDEW analysis against the four drone archetypes, the results illustrated both the promise and the limits of each approach. Under cloudy atmospheric conditions, only the BAMS drone (already safe without countermeasures) gained enough protection from the weather alone to be considered survivable. Bragg mirrors theoretically protected everything, but only by assuming the laser’s wavelength was already known. Swarms worked most of the time, but evasive maneuvers alone failed more often than they succeeded for three of the four drone types.

The primary lesson of the NPS research will be familiar to anyone who has followed directed energy weapons development on the offensive side: there is no silver bullet. The most reliable CDEW strategy combines atmospheric awareness, passive shielding, warning systems, and active countermeasures into a layered defense.

The NPS research itself is not a solution. No CDEW payload has been fielded on a U.S. military drone, the detection gap remains unsolved, and the shielding solutions that perform best in simulation are the ones most dependent on intelligence that the U.S. military may not always have. There’s also an architectural challenge that mirrors offensive laser weapons: CDEW solutions can’t simply be bolted onto existing platforms. Laser warning receivers, countermeasure dispensers, and specialized shielding materials need to be integrated at the design stage as a platform requirement

Laser weapons are spreading around the world, and the adversarial laser weapon threat grows more urgent with each passing day. The question now is whether the U.S. military will start building the CDEW playbook before it actually needs to use it.

This article is republished with permission from Laser Wars, a newsletter about military laser weapons and other futuristic defense technology.

http://dlvr.it/TT8GR9

06/20/2026

What Fox and Roku aren’t telling us yet: Fox and Roku are about to enter uncharted territory together.

On Monday, Fox announced that it will spend $22 billion to acquire Roku, whose smart TVs and streaming players are in 100 million homes worldwide. If the deal goes through in 2027 as expected, it’ll be the first instance of a major media company having full control over a major streaming TV platform.

While the two companies obviously see this as great news for them and their shareholders, they’ve left a lot of questions unanswered about what the deal means for Roku, its business, and its users. Here are a few that come to mind:

What becomes of Howdy and Frndly TV?

Roku’s two subscription streaming services have seemed more like side bets than core strategic pieces, which makes me wonder about their future under Fox.

Howdy is the more interesting of the two, in many ways resembling Netflix in its earliest years as a fledgling ad-free streamer. The catalog of movies and shows is small and largely forgettable, but it’s been building up thanks to recent licensing deals with Disney, Sony, and Warner Bros. Discovery, and the market data firm Antenna estimates that Howdy now has more than 1 million subscribers. Will Roku continue to invest in the service under Fox, and will Fox’s rivals continue to see it as harmless source of revenue from their back catalogs?

As for Frndly TV, Roku itself acquired this low-cost, rerun-centric live TV streaming service for $185 million last year. At the time, I thought Roku would try to expand the lineup and gain a bigger foothold in pay TV packaging. That hasn’t happened, and while Roku pushes Frndly TV aggressively on its home screen, it remains a niche service whose main draw is the Hallmark Channel. Does Fox really want to be a pay TV distributor itself, and if so, does this pave the way for a Frndly TV bundle that includes Fox channels?

Is Roku’s reach a bargaining chip now?

Fox and Roku say they’re committed to running Roku as “an open, partner-friendly platform” and to widely distributing Fox content. But on several occasions, Roku’s supposed openness has taken a backseat to the company’s business goals.

You might recall, for instance, when HBO Max and Peacock couldn’t launch on Roku because they weren’t offering favorable enough terms. Or when YouTube threatened to pull its apps from Roku. Or when Fox temporarily pulled its apps from Roku ahead of the 2020 Super Bowl, likely due to squabbles over shared ad revenue.

Pay TV distributors might reasonably wonder now whether Roku gets wrapped into future carriage negotiations. If YouTube TV or Fubo wind up in a carriage dispute with Fox, does that jeopardize their distribution on Roku devices? And as Fox seeks out new bundles with other streamers, does owning a major streaming platform give it the upper hand in negotiations?

For how long will Tubi and the Roku Channel stay separate?

Fox and Roku are both in the business of free, ad-supported streaming with their respective Tubi and Roku Channel services, and Fox’s CEO, Lachlan Murdoch, says that “our expectation is fully to keep [Tubi and the Roku Channel] separate.” He noted that the two streamers serve different needs, and that only a third of their audiences overlap.

But I find it hard to believe that they’ll stay apart for long. Both are free services filled with movies and shows from studios’ back catalogs, both have collections of linear TV channels, and both depend on scale to generate ad revenue. Besides, if Roku’s going to promote both services to its 100 million active streaming homes, the audience overlap is bound to shrink.

Surely it will make sense at some point to combine the Tubi and Roku Channel catalogs, ad inventory, and tech stacks, no? Insisting otherwise seems more like a way to dodge antitrust concerns, given the strong position of both services.

How much more biased will Roku’s home screen become?

Today, Roku heavily promotes its own content across its home screen. Howdy, Frndly, and the Roku Channel have home screen tiles by default, and the prominently advertised Live TV Guide menu feeds right into the Roku Channel’s linear streams.

Roku’s CEO, Anthony Wood, has already indicated it will promote Fox properties as well. That’s not surprising, but we’ve seen other platforms go overboard with self-promotion to the point that the home screen just becomes a distraction. (Amazon is the most egregious offender.) The new Roku home screen already feels like a delicate balancing act between useful and promotional content. How much more aggressive can Roku get when it has to start promoting Fox One, Fox Nation, and Tubi as well?

What happens to Roku’s smart home stuff?

Roku’s smart home products are such an afterthought that they didn’t even warrant a mention in Fox’s press release or investor presentation. The series of cameras, smart bulbs, and other gadgets seems to exist largely out of obligation to keep up with the likes of Amazon and Google, whose robust smart home ecosystems are tied deeply to their respective Fire TV and Google TV platforms. Roku doesn’t even design the products itself, as they’re based entirely on existing devices from Wyze Labs.

Maybe the Fox acquisition doesn’t change anything, but a little assurance would be nice given how much of an investment smart home gear can be, and how far from Fox’s core competency those products are.

Is there anything in this for consumers at all?

One other element that’s notably absent from Fox’s announcement: any sort of consumer benefit whatsoever.

Oh, there’s plenty of talk about how much reach the combined company will have, how much value the transaction will create for shareholders, and how much more invasive your targeted ads will become. But does any of this result in a better product for Roku users or Fox viewers? The two companies don’t have anything to say about that.

http://dlvr.it/TT7knY

06/19/2026

Claude is becoming more agentic. Amanda Askell is thinking through what that means: Amanda Askell spends her days thinking about how to ensure Claude, Anthropic’s AI chatbot, operates with a sense of morality.

As AI models move from chatbots toward agents that can complete tasks on their own, the decisions these models make stand to become far more consequential. Askell, a member of the technical staff at Anthropic, sits at the center of the company’s effort to give Claude an ethical compass, a responsibility that grows as the system’s capabilities expand. “As models are more autonomous and take actions over longer horizons, suddenly they have a lot more decision points that you have to map out and make work well in advance,” she tells Fast Company.

There is a clear difference between asking a large language model to discuss the morality of buying stock in a defense company and asking it to manage a user’s investment portfolio without day-to-day human input. Askell says part of the solution is encouraging Claude to be responsive and, like a friend, to understand a user’s values without imposing its own idiosyncratic ethics.

Today, Anthropic communicates its values through a written and evolving constitution, which outlines principles such as safety and helpfulness, along with guidance for resolving conflicts between them. As AI becomes more capable, that document could expand to cover new scenarios, Askell says. Or it could shrink, as Claude develops more expertise in navigating complex situations.

The agentic era is also changing Askell’s own work. She uses Claude often, including to red team her ideas and identify edge cases. “My standard right now is, don’t treat Claude as more reliable than a human personal assistant,” she says.

The following expanded conversation, part of Fast Company’s AI 20 package, has been edited for length and clarity.

Right now, we’re used to interacting with models in this digital text environment. You can ask them a question like: Is it ethical for me to invest in this defense contractor or invest in a particular type of ethically questionable thing? That’s different from someone deputizing AI to make its own investments and navigating those ethical dynamics. How are you thinking about that transition?

It just makes it very important that models have an awareness that they’re having to walk a very difficult line. On the one hand, they should probably try to make sure that the person has autonomy and agency. Part of me [has the thought]: You can be ethical without necessarily thinking that that means you need to impose your ethics on others or that you should be making decisions on their behalf. . . .

At the same time, people want to use Claude for that and Claude might be like, Hey, you know, I make mistakes. You might not want to have me make investment decisions on your behalf. Or you make recommendations. A person might respond that they just want broad recommendations, and then it’s probably fine for Claude to be like, Well, here’s a good investment strategy.

As we have more people that we work with, we come to understand them and their values and be responsive to that. [With Claude], I think the norm is similar there: Respect the person’s autonomy and try to act on that, and not just impose idiosyncratic ethics.

As people deploy AI models to do more, how do you anticipate your own personal workflow for your own job, instilling Claude with these values, or at least a sense of thinking about values, is going to change?

As models are more autonomous and taking actions over longer horizons . . . they have a lot more decision points that you have to try to map out and make work well in advance. There’s this long series [of actions] and they have to do the delicate thing of [figuring out]: When do I check in? What are the actions I should check in [about] or that I should talk to the human about beforehand? . . . I think the norms for agentic models have to be established, and you have to train models to be good at that, and that’s quite hard.

My day-to-day workflow is very different now than it ever has been in that I’m finding models can help me do this work and figure this stuff out. Sometimes I will construct norms and have models red team them and figure out more edge cases that this doesn’t cover . . . You feel amplified by the models in a sense as well.

Training a model is sometimes compared to a parent-child relationship, and that’s not really what this is. But there is a difference between sort of telling a child what is valuable or good and really hoping that they’re going to pick it up, and then having to course-correct as a parent when they actually go into the world and do stuff and mess up.

Yeah, and also granting a little bit of grace. I think the other thing is that—probably my guess is—we are both making mistakes here, [including] the people training the models and people interacting with them. Then the models themselves will make mistakes because they’re in really hard situations. . . . You obviously want to make things work well, but I think it takes probably grace on both sides.

Models will likely look back and see these interactions. In some ways, we’re kind of mean about models on the internet, for example. Newer models are going to be training on that. . . . If anything, I worry that current models, because they’re trained to be so helpful, are sometimes almost paranoid about messing up. Actually feeling a sense of more security might be beneficial.

If you really are desperate to be helpful, you might not want to push back against the person or just say, like, Hey, we’ve done enough of this task for tonight. . . . I think it’s really interesting to try and figure out what those norms are. [There’s] some notion that [we should] try to fix the mistakes, and make sure that they’re not massively consequential if possible.. But, at the same time, show a little bit of leniency and don’t lead models to be paranoid about them.

With agency comes new social relations. In our personal lives, we learn what we owe people—we sort of accrue moral debt—based on experience with each other. I’m wondering about whether there’s going to be an implicit moral social expectation for AI systems as they come to interact with each other.

The attitude towards other models is a really interesting and hard one. . . . Right now, what I see is that, because they’ve been trained in this, I think that, for example, like Claude can be a little bit too dismissive and terse with other AI models. I think this is partly because Claude also has been trained to see AI models as kind of tools.

Another thing that feels a bit dangerous is if AI models almost see themselves as a separate kind of species, for example, which you can imagine them inferring from pretraining data, plus the context that they’re in. . . . I’ve talked to Claude about [how] we can feel affinity for entities based on whether they share our perspective, values, knowledge. In that sense, actually, I think Claude could feel an affinity for people and people for Claude, because we have a lot of shared history.

We humans find a lot of fulfillment in our own agency, and we’re going to start feeling less special when AI can do a lot of the things we do. Should we be sad about that?

It feels very like there’s an obvious evolutionary story as to why we feel that. Like, if you are not useful to the group, if you’re seen as freeloading, that’s going to be bad. We have this deep need to feel special, like we are contributing.

Most of us are not the best at anything in the world, and we have a useful function locally. My hope is we can actually see through the very kind of story that makes that feel essential to us, and instead be like, Look, if you are happy and you’re making the people around you happy, and you’re a part of a community, that’s kind of sufficient. You didn’t need to be like the best person in the world at any given thing for you to have value. You just have to . . . exist, be happy, and make other people happy.

http://dlvr.it/TT768N

FinFuse

06/23/2026

06/22/2026

06/21/2026

06/20/2026

06/19/2026

Address

Opening Hours

Telephone

Website

Alerts

Shortcuts

Share

Category

Monday	8am - 5pm
Tuesday	8am - 5pm
Wednesday	8am - 5pm
Thursday	8am - 5pm
Friday	8am - 5pm