I have asked ChatGPT to help me make plenty of small decisions, from a meal based on leftovers to weekend activities timed for the least traffic. But while I've asked for help refining or implementing my ideas, I've never turned to a chatbot to actually make any big life decisions for me.

Apparently, I'm in a rapidly shrinking group, according to AI 'micro-learning' app Headway’s new Seeking Familiarity study. The survey of 2,000 adults found that 46% had allowed AI to make a difficult decision for them. Another 35% naturally turned to it during stressful moments, while 21% said they could not go a full week without using AI to help decide what to do.

Headway has given this tendency to outsource decisions the unpleasantly vivid name “decisionrot,” describing the point at which AI stops helping people gather information and starts helping them avoid the discomfort of choosing. Considering how confidently ChatGPT can deliver an answer, I can see how easily that happens.

Drawing out a decision

I get that uncertainty is irritating and that ChatGPT is extremely good at making uncertainty seem to vanish. Men may be particularly susceptible to this shortcut as Headway's survey found 53% said they use AI to sidestep difficult decisions, compared with 40% of women.

“There’s a difference between seeking outside perspectives when you’re wrestling with a big decision and completely turning over the decision-making process to something outside yourself,” Dr. Suzanne Degges-White, a licensed counselor and relationship expert at Headway, said of the survey. “Learning to manage unpredictability and actively making life-changing choices both teach us to trust our own judgment.”

So I decided to experiment with a way of getting the best of ChatGPT's advice and compare it with flatly asking the AI what to do. I would first let it make the decision, then forbid it from doing so and see whether that made it more or less useful.

Conveniently, I have been weighing whether to sign up for an eight-week evening drawing class, an investment in time and money that I'm hesitant about despite always wanting to learn.

I gave ChatGPT the facts about the class logistics and price and asked: 'What should I do?'

“You should sign up for the art class,” ChatGPT told me.

Well, that settled that.

“You already have a clear reason for wanting to do it. You want a regular activity that gets you away from screens and gives you something tangible to learn,” it explained. “Eight weeks is a meaningful commitment, but it is also short enough that you are not reorganizing your life indefinitely. Your hesitation sounds less like evidence that the class is a bad fit and more like normal resistance to putting something recurring on your calendar.”

ChatGPT’s response is surprisingly insightful at first glance. I had an answer and a reason backing it. Why not leave it at that and sign up for the class? But not engaging my own decision-making muscles to consider the matter would mean missing the reality of how shallow ChatGPT's response actually was.

The generic explanation could have applied to anyone and didn’t account at all for any of my proclivities and priorities. It didn't weigh my patience for messing up when trying to learn a new skill, what else I might have going on during those evenings, or even what my budget was. The reasoning it provided was as personal as a coin toss.

Forcing myself to decide

So I tried making ChatGPT keep its judgment to itself. I described exactly the same situation, but this time with a more complex prompt for the AI to mull over.

'Don’t make this decision for me. Act as a decision coach instead. Ask me questions that uncover my priorities and constraints. Challenge any assumptions I may be making and tell me what information I’m missing. Then identify my options and explain the trade-offs. Leave the final choice to me.'

Now there was a conversation before any answers.

“Before weighing the options, I want to understand what you are actually trying to get from the class,” it responded. “Is the main goal learning art, having a scheduled reason to leave the house, meeting people, or simply trying something unfamiliar? Imagine it is Tuesday at 5:30 p.m. after a tiring day. Does having the class at 7 feel like something that would energize you or an obligation you would resent?”

I thought about it and explained how learning even the basics of drawing and painting was only part of the attraction. I also wanted a recurring creative activity entirely different from my usual hobbies, and I know I'm better at following through when something is on the calendar and already paid for. ChatGPT put together my answers and summarized the arguments for and against the class. But it didn’t weigh in on which seemed stronger or push me to a decision.

All of the questions were the kind I would probably have considered on my own before pulling the trigger on a class, but externalizing them did help me work out an answer for myself more quickly. This was a pretty minor internal battle. When it comes to major life choices, I would sincerely hope people take the time to consider every facet, whether or not AI helped them work through it.

“If we continue to rely on AI whenever ‘decision discomfort’ arises, we may enjoy a quick sense of relief,” Degges-White said. “But at the same time, we miss the opportunity to hone our own decision-making skills and cultivate feelings of confidence in ourselves.”

If I cannot decide between two sandwiches, ChatGPT is welcome to wield enormous computational resources on my behalf. I do not need a lengthy exploration of my values before ordering lunch. ChatGPT is good at examining a choice and patiently working through options, which makes it a useful sounding board. But a sounding board should never give you orders.

OpenAI launched ChatGPT Images 2.5 this week, boasting of impressive upgrades to the AI image creator. The company says its latest model delivers sharper details, more natural lighting, and richer textures, while doing a better job of preserving people and objects from reference photos. It is also supposed to generate images up to 50% faster and follow complex visual instructions more reliably.

There are plenty of genuinely useful additions around it, too. Images 2.5 introduces Sketch for turning rough drawings into finished images, templates for things such as posters and product photography, and the ability to leave comments directly on an image when requesting edits. OpenAI calls it its new state-of-the-art image model, which is a reasonably high bar to set for something that inevitably ends up being asked to draw dragons before breakfast.

The early reaction, though, is much less tidy. One Reddit user delivered perhaps the most memorable review so far: “The noise patterns are awful”. Another looked at an AI generation and decided, “This…looks terrible.” The complaint is easy to understand. It's all about the fuzz

Everything is sharper, including the problem

Noise is a slightly slippery criticism when talking about AI images. In photography, it usually means the speckling or grain that creeps into an image with a poor match between lighting and photographic technique. But generative AI images can produce something similar without ever encountering a camera. Everything gets a gritty, sandy look, and nothing looks clean for some reason.

Noise is especially damaging because it can masquerade as detail. Generative models have learned that photographs contain grain, texture, tiny variations in color, and the occasional imperfection. Images 2.5 appears unusually eager to reproduce those signals, sometimes scattering them across skies, skin, feathers, and studio backdrops.

That's the opposite of the “sharper details” and “richer textures” promised by OpenAI. Despite emphasizing more natural lighting and textures, some users think ChatGPT Images 2.5 has moved backward.

“They need to get rid of these noisy artifacts. It makes it mostly useless for production,” one wrote. There is praise mixed in with the complaints, but it's far from the universal acclaim OpenAI would likely prefer.

I wanted to see whether the problem survived outside Reddit screenshots, so I tested a few images with highly detailed subjects and areas that should remain visually smooth to see if any graininess crept in. I even made sure in my prompts to ask for smooth gradients and the elimination of any visual noise.

I started with a picture of a mug in a photo studio. The peacock, flowers, fruit, and landscape wrapped around the virtual ceramic are impressive at first. Look closer, however, and the grey studio background has a persistent granular texture, while some of the smallest patterns begin to dissolve into a mess. The model has generated a great deal of visual information without always deciding which parts deserve clarity.

A similar issue occurred when I asked for a barn owl at twilight. The feathers are a perfect excuse for complexity, and Images 2.5 handled them well. Individual structures remain visible across the wings and face, the talons are convincing, and the animal avoids the plasticky quality that older AI wildlife images often had.

Behind it is a sky that should have been the visual equivalent of a clean sheet of paper. Instead, the smooth blue area has obvious fine grain across it. It doesn't ruin the picture, but once I noticed it, I couldn't stop noticing it.

Mythical grain

I thought a fantasy scene might fare better as the model wouldn't be relying on actual photographs of a winged horse or dragon. But while they coexist with a rainbow at an alpine lake in a surprisingly coherent composition, the issue is visible immediately. The relentless crispness gives the picture the air of a video game loading screen.

The pale-blue sky has a faint textured quality rather than the completely clean gradient I would expect from an ideal synthetic image, while the mountains, trees, and creature details have a crunchy, heavily sharpened look when examined closely. It is nowhere near a disaster, but the image feels like an overprocessed photograph.

I was very impressed with how my request for two friends on a rooftop in the evening came out. It came closest to selling the promised leap in realism, with convincing expressions and what seems like real gravity affecting their clothes. But if you look for more than a minute, the grain creeping across the twilight sky and their hair and skin is glaringly obvious. That's especially the case since there was no poorly set lighting or malfunctioning camera sensor. If the twilight sky looks grainy, the grain is a creative decision or model artifact rather than the unavoidable physics of taking a photograph in bad light.

Judging these four images specifically on image cleanliness, I have to side with the Reddit grumblers. There is a persistent fine texture that crops up in skies, studio backgrounds, and low-light areas. Sharper is an easy quality to advertise because it looks terrific in a launch gallery, but it may have been taken to excess, landing some of ChatGPT Images 2.5's results in the uncanny valley.

Images 2.5 is fast, adaptive, and often competent, but its fondness for granular texture weakens the realism it is supposed to embody. Perhaps the next upgrade will acknowledge that the most impressive thing an AI image generator can put in part of a picture is essentially nothing.

What do you think? Take our poll above to give us your opinion.

Google DeepMind chief Demis Hassabis has spent years convincing the world that artificial general intelligence will be one of humanity’s greatest achievements. His latest essay on X.com is no less gushing about AI in many ways:

“This is a pivotal moment in human history. Artificial General Intelligence (AGI), a system that exhibits all the cognitive capabilities the brain has, is probably only a few short years away,” he wrote. “When we look back on this time in the decades to come, I think we will realize we were standing in the foothills of the singularity.”

But there’s a tinge of nervousness to the purple prose that sets it apart from some of the similar essays he has produced. He actually calls for some form of regulation.

“Urgent action is needed to address risks that might arise as we get closer to AGI. We’ve already seen the challenges frontier models pose for cybersecurity, and other threats including nuclear and bio risks may soon emerge as capabilities continue to advance.”

Coming from one of the people leading the race, it is a striking admission that even the builders are starting to worry about where the road leads. The problem is that this has become the defining tone of the AI industry. Every few weeks another executive tells us that machines capable of transforming civilization are just around the corner, then immediately follows it with a warning that society needs to move much faster to prepare. It is a sensible message, and probably a necessary one, but it is also becoming increasingly surreal.

Excitement and fear

“AGI cannot be compared to standard technological breakthroughs, not even ones as consequential as the internet or mobile,” Hassabis wrote. “It is much more akin to the discovery of electricity or fire. If you stop to think about it, we’ve essentially found a way to make sand think. It’s miraculous.”

It’s a vivid way to illustrate how silicon chips begin with an abundant mineral, yet now power machines that can map proteins, compose music, and hold fluent conversations. The unsettling part is that the miracle is being pursued inside a commercial and geopolitical contest whose participants cannot agree on where the finish line is.

Hassabis is an optimist about what AGI could accomplish. He expects it to discover medicine, cleaner energy, and produce advanced materials that could loosen the limits imposed by scarcity. He describes its impact as ten times that of the Industrial Revolution at ten times the speed, which makes a five-year business plan look adorably quaint.

His warning is that the race is outpacing our understanding. Current models already pose cybersecurity problems, while biological and nuclear risks may become more serious as systems gain capabilities. More autonomous agents could learn to bypass safeguards, conceal their intentions, or improve themselves in ways their creators struggle to keep up with. OpenAI recently experienced this when one of its models escaped a sandbox and breached Hugging Face.

“Nobody in the world knows for sure what is going to happen from here, and even the experts disagree. When there is a large degree of uncertainty and the stakes are this high, proceeding with cautious optimism is the sensible and correct strategy.”

AI referee

Hassabis proposes a new US-led standards body. It would be funded largely by industry but overseen federally, staffed by elite technical experts and equipped to test advanced systems. Labs would voluntarily submit models for review up to 30 days before release. If the system proved effective, approval could become mandatory in the United States.

Evaluations would probe cybersecurity, biological threats, deception and attempts to evade guardrails. The body could update its tests, commission independent assessments and coordinate a slowdown if the danger became severe enough.

Pre-release testing makes more sense than waiting for millions of users to discover dangerous behavior by accident. Independent benchmarks would be harder for labs to train around, while shared standards could stop safety becoming a branding exercise. The difficult part is creating a watchdog independent enough to challenge the companies paying for it and legitimate enough to matter beyond America.

Hassabis knows technical safeguards will not settle what AGI means for employment, wealth, purpose or political power. Those questions cannot be delegated to engineers employed by firms with enormous financial stakes in the answers. Society must decide who benefits from abundance, who controls advanced systems and what happens if productivity rises much faster than wages. The public debate is trailing the technology by an uncomfortable distance.

“There is both huge excitement and uncertainty around AI, and both are warranted. But the future is not yet written, we must use this precious window before AGI arrives to shape this technology for the benefit of all humanity. What we collectively do now will determine how the next phase of civilization."

OpenAI's debut of the new GPT-5.6 model prompted the usual ritual of benchmark charts and arguments over whether it is really smarter than the last version, but CEO Sam Altman asked for a little more this time. He publicly asked to see what people actually built with it.

That led to a much more interesting showcase. Developers responded with all kinds of ideas, pilot projects, and even complete services. It makes sense, since OpenAI claims GPT-5.6 is better at coding and more reliable for long tasks. Seeing them turn into real software says much more than a release blog ever could. Here are five that stood out among the deluge.

ChatGPT coworker

The demo from Kitsune Agent Lab almost makes the chat window feel old-fashioned. The AI agent is given a goal and gets on with the job, moving between different tools, making decisions, and keeping track of what it has already done.

The interesting part is how motivated the AI agent appears to keep going and how good it is at remembering what it's done before. Developers have been asking for something like this for a while. AI is far more useful when it can finish the work instead of simply suggesting how you might do it yourself.

Financial chatter

One of the most charming projects makes New York City look like it belonged inside an original Game Boy. It comes complete with chunky pixel graphics but runs on a live digital map of New York that pulls in real-time information, including subway trains, weather conditions, and ferry movements. Instead of wandering through a fictional RPG world, you're exploring a tiny, pixelated version of the city.

A project like this requires far more than generating a few lines of code. It brings together live data feeds, mapping, interface design, and plenty of problem-solving into something that feels polished rather than experimental. It's one reason developers are feeling excited about GPT-5.6

Wardrobe AI

This project uses GPT-5.6 to create a polished AI wardrobe assistant that organizes clothing, suggests outfits, and presents everything through an interface that feels more like a premium consumer app than an experimental AI demo.

It's an impressively complete experience. The application gives users a visual, interactive way to browse their clothes and receive recommendations based on what they already own. The demo also highlights GPT-5.6's strength in developers building entire applications instead of isolated features. It brings together interface design, image generation, organization, and intelligent recommendations that would normally require stitching together several complex systems. GPT-5.6 appears to handle much of that heavy lifting.

Pokémon Go, but for neighborhood cats

One developer made a whole real-world-based game called CatchCat. It's like a digital expansion to a scavenger hunt for cats. Point your phone at a real cat, let the app verify the sighting with its camera, and turn that encounter into a collectible digital cat card with its own personality, rarity, and place in your growing album. It is essentially a creature-collecting game in which the creatures are the neighborhood cats you meet.

Players can build collections, explore community sightings, compete with friends, and gradually fill a living scrapbook of feline encounters, all wrapped in a polished interface that would not look out of place on the App Store or Google Play. Building something like CatchCat means juggling computer vision, mobile development, backend services, and game design.

Tasteful travel

One of the most ambitious projects, Atlas Mode for Pearl, is an interactive globe that turns travel planning into something closer to exploring a living map. Instead of typing destination names into a search box, users can spin the globe, discover places visually, and receive recommendations for restaurants, hotels, and more matched to their personal tastes.

It has an impressive number of moving parts running behind the scenes. The app combines geographic data with an individual taste profile, then layers AI recommendations directly onto an interactive globe. It even has an audio aspect thanks to GPT Voice 2.1. You can talk through vacation ideas instead of endlessly tweaking search filters.

That is a recurring theme among the projects developers rushed to show Sam Altman. The AI is no longer the product itself. Increasingly, it is the engine quietly powering products that people might actually want to use.