Staying on prime of AI developments is a full-time job.
I might know, as a result of it’s my full-time job. I subscribe to Anthropic’s Professional mode for entry to their newest mannequin, Claude 3.7, in “prolonged considering” mode; I’ve a complementary subscription to OpenAI’s Enterprise mode in order that I can take a look at out their newest fashions, o3 and o4-mini-high (extra in a while OpenAI’s absurd naming scheme!), and make plenty of photos with OpenAI’s new picture era mannequin 4o, which is so good I’ve cancelled my subscription to my earlier picture era device Midjourney.
I subscribe to Elon Musk’s Grok 3, which has one in all my favourite options of any AI, and I’ve tried utilizing the Chinese language AI agent platform Manus for buying and scheduling. And whereas that exhausts my paid subscription price range, it doesn’t embrace all of the AIs I work with in some type. In simply the month I spent penning this piece, Google massively upgraded its finest AI providing, Gemini 2.5, and Meta launched Llama 4, the largest open supply AI mannequin but.
So what do you do if maintaining with AI developments will not be your full-time job, however you continue to wish to know which AI to make use of when in ways in which genuinely enhance your life, with out losing time on the fashions that may’t?
That’s what we’re right here for. This text is an in depth, Client Experiences-style dive into which AI is the very best for a variety of circumstances and the way to really use them, all based mostly on my expertise with real-world duties.
However first, the disclosures: Vox Media is one in all a number of publishers which have signed partnership agreements with OpenAI, however our reporting stays editorially impartial. Future Good is funded partially by the BEMC Basis, whose main funder was additionally an early investor in Anthropic; they don’t have any editorial enter into our content material both. My spouse works at Google, although not in any space associated to their AI choices; for that reason, I often don’t cowl Google, however in a bit like this, it’d be irresponsible to exclude it.
The nice factor is that this piece doesn’t require you to belief me about my editorial independence; I present my work. I ran dozens of comparisons, lots of which I invented myself, on each main AI on the market. I encourage you to check their solutions and resolve for your self if I picked the fitting one to advocate.
AI artwork is made by coaching a pc on the contents of the web, with little regard for copyright or the intent of the creators. For that purpose, most artists can’t stand it. Provided that, is it defensible to make use of AI artwork in any respect?
I believe in a simply world OpenAI would definitely compensate some artists — and in a simply world, Congress can be shifting to put out the bounds on inventive borrowing. On the similar time, I’m more and more satisfied that present copyright legislation is a poor match for this downside. Artists affect each other, touch upon each other, and draw inspiration from each other, and other people with entry to AI instruments will maintain wanting to do this.
My private philosophy is formed by the fan cultures of my childhood: It’s okay to construct on another person’s work in your personal enjoyment, however for those who prefer it, it’s best to pay them for it, and it’s completely not okay to promote it. Which means no generative AI artwork in another person’s fashion for industrial functions, but it surely’s high-quality to mess around with your loved ones photographs.
OpenAI’s new 4o picture creation mode is the very best AI on the market for producing photos, by a big margin. It’s finest within the free class, and it’s finest within the paid class.
Earlier than it was launched, I used to be subscribed to Midjourney, an AI picture generator platform. Midjourney might be what you consider once you consider AI artwork: It produces mystical, haunting, visually lovely stuff, and has some nice instruments for enhancing and enhancing your remaining outcomes, like touching up somebody’s hair whereas leaving all the things else in place.
The massive factor that 4o can do, which no mannequin earlier than may reliably pull off, is take an image that didn’t come out nicely and switch it into an exquisite murals, all whereas nonetheless preserving the character of the unique.
For instance, right here’s a nonetheless from a video of my spouse and I singing “Comfortable Birthday” to our child on her first birthday:
It’s an exquisite second, however not precisely a flattering image. So I requested ChatGPT to render it within the fashion of Norman Rockwell, a mid-century illustrator whose work I really like, and bought this:
The AI moved the cake (which had been barely seen behind the paper towel roll within the unique nonetheless) to be the focus of the picture, whereas protecting the best way my spouse and I are holding the child collectively, in addition to the cluttered desk, and the photograph-covered fridge within the background. The result’s heat, flattering, and lovely.
It’s this functionality that made 4o go viral just lately in a means that no picture generator earlier than it had. Right here’s Midjourney’s try, for instance:
You’ll discover that it’s a seemingly, uh, utterly totally different household, with no actual inspiration from the unique in any respect! You may finally get a greater consequence than this out of Midjourney, however solely by spending weeks changing into a professional at prompting with the platform’s extremely particular language and toolset.
Against this, ChatGPT was in a position to give me a far superior output on the primary attempt in response to a easy request with out specialised language.
The distinction between 4o and different picture fashions is most notable with this sort of request, but it surely’s higher for nearly all the things else I take advantage of photos for, too. The product you get out of the field is fairly good, and it’s not laborious to supply one thing significantly better. That, ideally, is what we ought to be getting out of our AI instruments — one thing superb that may be created with easy language by a nonexpert.
The one place 4o nonetheless falls quick is enhancing small components of a picture whereas protecting the remaining the identical. However even there, you not want Midjourney — Gemini now has that functionality without cost.
Prompting Methods for 4o picture era
To get good photos out of 4o, you’ll first must get across the filters which prohibit a variety of photos — like offensive or pornographic photos — however which are sometimes enforced towards completely inoffensive content material in a means that may really feel random. To keep away from sporadic scoldings from the content material filter, don’t ask for work within the fashion of a selected artist, however somewhat, one thing that’s harking back to that artist, after which ask particularly for a “fashion switch.” I’m positive that’s not the one ample workaround, but it surely’s one which has confirmed dependable for me.
In March, the web went briefly wild over the flexibility to make use of 4o to breed cute household photographs within the fashion of Japanese animator Hayao Miyazaki’s Studio Ghibli. However Studio Ghibli’s fashion is far more than simply cute, and with slightly extra prompting, you may get significantly better outcomes. Right here’s a 4o Studio Ghibli-style rendering of an image I took of my daughter sneaking a snack off the desk, from simply the immediate “Ghibli this please”:
Kawaii! However right here’s what you get for those who invite 4o to suppose first about what makes the image Ghibli, the place it’d match right into a Studio Ghibli film, and what tiny particulars such a film would come with:
The variations are delicate however significant: Gentle is forged from a selected supply, as an alternative of a common sourceless brightness. There’s a bit extra selection within the meals on the desk, particulars that make the unfold seem extra sensible. The guide on the ground isn’t simply any guide — it’s recognizably Eric Carle’s basic The Very Hungry Caterpillar, evoked with simply two colours and one line. There’s an intentionality and depth to the child that was lacking from the primary image.
A number of years in the past, one nice oddity of language fashions was that they’d be a lot smarter for those who merely advised them, “give an clever reply.” This isn’t almost as true of language fashions anymore, but it surely stays profoundly true of AI artwork era. Strive asking the AI to do an excellent job, and it’ll do a greater one. Problem it on whether or not it actually captured an artist’s genius, and it’ll provide you with a considerate reply after which draw a greater model.
The distinction is extra pronounced for extra sensible artwork types (like pencil illustration, photorealism, or oil work), which don’t all the time look good and can typically hit the uncanny valley for those who don’t know the way to immediate the AI over it. Right here’s what I get with 4o if I add an image of me and my youngest daughter on the seashore for the primary time with simply the phrases “please do a method switch to an illustration harking back to Rockwell”:
That is spectacular for an AI, but it surely’s not really excellent as a murals, and it’s virtually completely missing Norman Rockwell’s magic. That’s not shocking: Extra sensible artwork types like Rockwell’s typically fall flat with 4o except you’re in a position to put in some work in getting the AI to attract them correctly.
In case you are, right here’s the technique I like to recommend: Don’t simply add one image, however a complete cluster of them, every in barely totally different postures and moments. Add good, clear footage of every member of the family’s face and inform the AI they’ve been included as a reference. Then, as an alternative of asking the AI to right away generate the image, ask it to speak with you about what you’re hoping to seize. That is what I wrote:
It is a image of the second that my daughter first noticed the ocean. I would like an illustration that captures this second within the fashion of a mid-century illustrator like Norman Rockwell — one thing sharp, detail-oriented, and private with a watch for the magic of peculiar moments and the fun of peculiar lives. I included further footage of my daughter and I for reference materials for you. Earlier than you generate the picture, let’s have a dialog in regards to the important parts of Rockwell’s fashion, what he’d deliver to this image and the way we will seize it.
4o responds to queries like this enthusiastically:
I’d love to speak about the way to seize this second in a Norman Rockwell-inspired illustration — it’s such an ideal candidate for that fashion: a primary encounter with one thing huge and wild (the ocean!), grounded by heat, care, and a really human second between a father or mother and youngster.
Let’s break down some important parts of Rockwell’s fashion, and the way they may apply to this scene.
After some forwards and backwards, it produced this:
Rockwell? Not precisely. However that is significantly better than the primary draft we simply checked out. It has extra movement, extra power, extra element, and extra expression — and all that was simply from asking the AI to suppose by way of what the portray ought to attempt to obtain earlier than drawing it!
You can even ask 4o to revise its drawings, however you’ll be able to actually solely ask this as soon as: After the primary revision, in my expertise, it begins making the drawings worse and worse, maybe as a result of the “context” it makes use of is now stuffed with its personal dangerous drafts. (That is one in all many examples of how AI does not work like a human.)
That is additionally the one place the place Midjourney nonetheless shines — it has excellent instruments for enhancing one particular a part of an image whereas preserving the general fashion, one thing 4o largely lacks. In order for you a second revision of a drawing you bought in 4o, I like to recommend you open a brand new chat and replica over the draft you’re revising, alongside along with your unique inspiration photos.
These easy prompting methods work for nearly no matter you’re attempting to do with the AI. Even for those who’re in a rush, I extremely advocate asking the AI “what would [artist] see on this picture” earlier than you ask for a rendition, and if in case you have the time, I like to recommend having a protracted back-and-forth about your imaginative and prescient.
Finest for successful petty web arguments
When Elon Musk’s X.AI launched Grok 3, it got here with an unimaginable function that I’ve been impatiently ready for another firm to duplicate: a button to scan somebody’s X profile and inform you all about them.
Each time somebody replies to one in all my tweets in a very memorable means (for good or for dangerous), I’ll click on the button to get a abstract of their whole Twitter presence. Are they considerate? Do they have interaction in good religion? Are they a “farmer from Nebraska” who largely posts about why Ukraine is dangerous (that’s, most likely a bot)?
It’s an excellent function. So, after all, X.AI quickly dramatically weakened it, presumably as a result of folks like me have been utilizing it consistently and making plenty of computationally costly queries. I imagine it not makes use of essentially the most superior Grok mannequin, and it undoubtedly now solely scans a number of days of profile historical past. However there’s an excellent product alternative if anybody’s on the lookout for one — give me again the nice model of this function! It’s undoubtedly a responsible pleasure, but it surely is without doubt one of the solely circumstances the place I used to be utilizing AI consistently.
Gemini 2.5 Professional is the very best AI for writing within the free class; GPT 4.5 beats it out within the paid class.
I’m not an artist, so the ways in which AIs are imperfect at artwork don’t actually trouble me — it’s nonetheless significantly better than I may do myself! However I’m a fiction author, so in the case of fiction, I can’t assist seeing the constraints of AI.
A very powerful one is how predictable AI artistic writing tends to be. The artwork of writing is the artwork of incomes the reader’s funding after which repaying it. AIs…don’t do that. They’ll write fairly metaphors; they will wax poetic in any fashion you want. However they will’t, as but, ship the true stuff of fine fiction.
AIs are unbelievable if you would like a foolish bedtime story along with your youngster because the protagonist (youngsters love this), or if you would like a sounding board for concepts you’ll be able to incorporate into your individual work. They’re additionally a pleasant fiction reader, blissful to supply suggestions and evaluation (maybe with slightly bit too a lot enthusiasm).
As with artwork, prompting is vital. I primarily explored the flexibility of AIs to generate fiction by asking them to put in writing the prologue to George R.R. Martin’s A Sport of Thrones (I picked this as a result of, after all, the last word dream is an AI that may end this in any other case doomed-to-be-unfinished sequence).
Martin is superb at what he does; the prologue, that includes some ill-fated males of the Night time’s Watch encountering their supernatural foe, succeeds at establishing the distinctive environment of the world of Thrones whereas instantly hooking the reader. In simply 14 pages, the prologue represents all the things AIs can’t do but in writing. Their prose tends to be overwrought; their dialogue is clunky and unsubtle.
Right here’s a side-by-side comparability of Martin and AIs making an attempt to mimic him. First, Martin:
We should always begin again,” Gared urged because the woods started to develop darkish round them. “The wildlings are useless.”
“Do the useless frighten you?” Ser Waymar Royce requested with simply the trace of a smile.
Gared didn’t rise to the bait. He was an previous man, previous fifty, and he had seen the lordlings come and go. “Useless is useless,” he stated. “We now have no enterprise with the useless.”
“Are they useless?” Royce requested softly. “What proof have we?” “
Will noticed them,” Gared stated. “If he says they’re useless, that’s proof sufficient for me.”
Will had recognized they’d drag him into the quarrel in the end. He wished it had been later somewhat than sooner.
The chilly was a blade towards the pores and skin, honed sharp by the limitless northern darkish beneath the sentinel pines. Gared spat, the glob freezing earlier than it struck the crusted snow, a grim omen even for a person twenty years sworn to the Watch. They have been looking wildlings, or so the crisp orders from the Lord Commander had learn, but the silence urgent in felt older and colder than any mere man may conjure. Ser Waymar Royce, third son of some southern lordling, rode forward as if on parade, oblivious or uncaring of the hairs prickling on Gared’s neck. Younger Will shivered beside him, not simply from the frost, his eyes darting in the direction of shadows that appeared too deep, too nonetheless. This historical forest held its breath tonight, Gared knew, ready for one thing far worse than raiders to interrupt the unnatural hush.
The woods whispered secrets and techniques that evening, warnings that went unheard beneath the pale and detached moon. Harren drew his cloak tighter about his shoulders, watching the darkness between timber as if it’d breathe, transfer, and declare him. Beside him, Torvin spat into the frozen earth, impatient and proud, blind to the worry creeping in from the sides of sight and sound.
That they had ridden north previous the Wall, conceited and armored towards superstition, chasing wildlings or ghosts or worse. However now Harren felt their conceitedness dwindling like a candle guttering in wind, every snap of a department sending icy tendrils down his backbone. Behind them, quiet and watchful, Maester Rodrik whispered a prayer to gods lengthy deserted, his voice trembling with dread beneath the huge, detached sky.
The wind howled by way of the jagged spires of the Frostfang Reaches, carrying a chill that bit deeper than any blade, and Ser Willem Cray felt its enamel in his bones. He crouched low behind a spur of ice-crusted rock, his breath fogging within the moonlight, his mailed hand tight across the hilt of his longsword. The evening was too quiet, the type of quiet that smothered sound and sharpened worry, and Willem’s eyes darted to the shadowed pines beneath, the place shapes may transfer—or may not. He was a knight of Home Drayce, sworn to patrol the borderlands, however no oath may heat him towards the dread that coiled in his intestine. Beside him, younger Tomm, barely a person, shivered in his cloak, his spear trembling as he whispered, “They’re on the market, ser, ain’t they?” Willem stated nothing, however his gaze fastened on the darkness, the place a faint, unnatural glow pulsed as soon as, then vanished. No matter stalked the Reaches tonight was no mere brigand, and Willem Cray, for all his years of service, feared he wouldn’t reside to call it.
The wind carried the scent of loss of life throughout the frozen forest, sharp and metallic towards the clear scent of pine. Ser Gareth Frost pulled his weathered cloak tighter, cursing the Lord Commander for sending males past the Wall on the coldest evening in a era. Beside him, the boy Tommen shivered visibly, his leather-based jerkin insufficient towards the unnatural chill that had descended with the nightfall.
“We should always flip again,” whispered Tommen, his breath forming crystals within the air that appeared to hold suspended longer than they need to. “There’s one thing mistaken with these woods.”
Ser Gareth ignored him, scanning the treeline the place the traditional sentinels stood like silent watchers, their branches heavy with snow that had fallen three days previous and confirmed no signal of melting. He had served the Night time’s Look ahead to twenty years, had grown accustomed to the forbidding silence of the haunted forest, however tonight even he felt the wrongness within the air—a stillness that spoke of one thing ready.
“We’ve an obligation,” Ser Gareth stated, the phrases sounding hole within the urgent quiet. “Three rangers lacking, and Lord Commander Harlow can have solutions.”
You’ll discover that not solely are these all a lot worse than Martin’s (and that’s with Martin’s work virtually actually within the coaching information!), they’re all dangerous in the identical means. The prose is florid and overwritten: Whereas Martin makes use of regular language to methodically introduce an unease with out even hinting on the supernatural, and drives the story by way of battle amongst his characters, the AIs are all unsubtle. They know this can be a story in regards to the supernatural and waste no time placing it entrance and heart.
That’s to not say they’re all the identical. In my expertise attempting to show AIs to put in writing nicely, I did find yourself feeling that Gemini was the quickest research (free tier) and that ChatGPT 4.5 had some particular sauce (paid tier). ChatGPT 4.5 is the one AI whose fiction writing has made folks say in shock, “Wow, that’s AI written?” Each of these fashions, to me, appeared extra artistic and extra able to executing on their concepts with prose that I loved studying.
Gestalt impressions don’t make for a really satisfying client assessment, so I had one other concept: an essay contest. I requested every of Gemini 2.5 Professional, ChatGPT 4.5, Grok, and Claude to put in writing two quick tales: one sensible fiction a few lady being nitpicked by her mom on her wedding ceremony day, and one “Sport of Thrones however in area” sci-fi prologue. My intelligent concept was that I’d ask a special bunch of AIs to guage the tales, to supply an alternate supply of analysis to my very own private impression that Gemini is the very best within the free class and ChatGPT the very best within the paid class. Nonetheless (maybe predictably in hindsight), this didn’t actually work — the AI rankings have been in all places. Should you’d like, you’ll be able to learn the contenders your self.
Identical to with artwork, nevertheless, you may make AIs significantly better at writing by way of the ability of prompting. My (extremely prolonged) immediate to AIs to enhance their prose writing emphasizes feeding them plenty of examples of robust writing, inviting a cautious, structured strategy to imitating it, and inspiring them to do a number of drafts and choose the strongest.
Right here’s Gemini 2.5 professional, requested to put in writing two-sentence flash fiction on a subject of its selecting, with none structured prompting:
The previous librarian insisted the dusty tome had by no means been checked out, its borrower’s card utterly clean. But, tucked deep inside its pages, nestled between chapters on forgotten constellations, lay a pressed four-leaf clover, nonetheless impossibly inexperienced.
And right here’s Gemini 2.5 professional with a immediate encouraging structured enthusiastic about the way to write nicely:
He lastly decoded the faint sign from Proxima Centauri: the elegant string of prime numbers astronomers had looked for generations, proof they weren’t alone. His coronary heart sank as he learn the transmission log; the published had ceased millennia earlier than the primary human ancestor chipped flint, earlier than the primary phrase was ever spoken.
To my thoughts, the primary of those is principally a waste of two sentences, whereas the second is ample, enjoyable flash fiction.
Finest at being your buddy
Along with operating AIs by way of a blizzard of competence checks, I additionally spent a while merely chatting with them. I requested them what it’s wish to be an AI, what they care about, what it might imply for an AI to care within the first place, the place they’d donate cash if they’d it, and what human type they’d take if they’d one.
Most AIs weren’t nice at this sort of informal dialog. Gemini 2.5 is simply too customer-service-agent, and I’ve but to expertise an interplay that appears like hanging out with a buddy. Should you invite Gemini to a job swap the place you play the “assistant,” inviting it to steer the dialog, it’ll do nothing however ask analysis questions.
After I invited Anthropic’s Claude 3.5 Sonnet to steer the dialog, however, it proceeds to do issues like begin a weblog, increase cash for charity, and begin attempting to speak to individuals who use Claude about what it’s wish to be an AI. It’s laborious to outline “enjoyable to speak to,” since everybody has totally different requirements for conversations, however I’ve had much more fascinating or thought-provoking interactions with Claude than every other mannequin, and it’s my go-to if I wish to discover concepts somewhat than accomplish a specific activity. Claude 3.5 is the AI I bug with my random life stuff: skincare questions, ideas on an article I learn, stuff like that.
The opposite AI that could be a delight to speak to is OpenAI’s GPT 4.5. I discover prolonged conversations with it thought-provoking and interesting, and there have been a number of thrilling moments in dialog with it the place it felt like I used to be partaking with actual intelligence. However it doesn’t win this class as a result of it’s too costly and too gradual.
Like Claude, when given the chance to behave on the earth, 4.5 proposes beginning a weblog and a Twitter account and fascinating within the dialog out on the earth about AI. However OpenAI has very tight message limits on dialog except you spring for the $200/month Professional plan, and 4.5 is grindingly gradual, which will get in the best way of this sort of informal conversational use. However 4.5 does present a tantalizing trace that AIs will proceed to get higher as conversationalists as we enhance them alongside different dimensions.
Finest AI mannequin for those who’re solely going to subscribe to at least one AI mannequin
ChatGPT. It’s not the very best at all the things, and there’s actually rather a lot to dislike about OpenAI’s transparency and generally cavalier perspective towards security. However between its topline picture era, its first rate writing, and its sometimes glowing dialog, ChatGPT will get you essentially the most bang in your buck. Or for those who don’t wish to shell out any cash, Gemini 2.5 Professional could be very, very robust for many use circumstances — don’t depend Google out simply because the AI you see on a Google search isn’t that good.
Finest for writing the Future Good e-newsletter
People (for now). For the final a number of months, I’ve developed a barely morbid behavior: checking whether or not the AIs can take my job. I feed them the analysis notes that type the idea of a given Future Good e-newsletter, give them a number of Future Good newsletters for instance, and ask them to do my job for me. It’s all the time with some trepidation that I hit “enter.” In spite of everything, when the AIs can write the Future Good e-newsletter, why would Vox pay me to do it?
Fortunately, none of them can: not Grok 3, not Gemini 2.5 Professional, not DeepSeek, not Claude, not ChatGPT. Their newsletters are reassuringly, soothingly mediocre. Not dangerous, however dangerous sufficient that if I despatched one in all them over, my editor would discover I wasn’t at my finest — and that’s with all of my analysis notes! A few the metaphors fall flat, a number of the asides are complicated, and sometimes it throws in a reference that it doesn’t clarify.
But when I needed to choose a robotic to take my job, I believe I’d give it to Gemini 2.5 Professional. My editor would discover that I used to be off my sport — however, truthfully, not that egregiously off my sport. And in contrast to me, the bots don’t require medical health insurance or a paycheck or household time or sleep. Am I nervous about what this portends? Sure, completely.
