Staying on high of AI developments is a full-time job.
I’d 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 pondering” mode; I’ve a complementary subscription to OpenAI’s Enterprise mode in order that I can check out their newest fashions, o3 and o4-mini-high (extra afterward OpenAI’s absurd naming scheme!), and make plenty of pictures 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 considered one of 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 funds, it doesn’t embody all of the AIs I work with in some type. In simply the month I spent penning this piece, Google massively upgraded its greatest AI providing, Gemini 2.5, and Meta launched Llama 4, the most important open supply AI mannequin but.
So what do you do if maintaining with AI developments is just not 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, Shopper Experiences-style dive into which AI is one of the best for a variety of instances and methods to truly use them, all primarily based on my expertise with real-world duties.
However first, the disclosures: Vox Media is considered one of a number of publishers which have signed partnership agreements with OpenAI, however our reporting stays editorially impartial. Future Excellent 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 match their solutions and resolve for your self if I picked the proper one to suggest.
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 cause, most artists can’t stand it. Provided that, is it defensible to make use of AI artwork in any respect?
I feel in a simply world OpenAI would definitely compensate some artists — and in a simply world, Congress can be transferring to put out the boundaries on creative borrowing. On the similar time, I’m more and more satisfied that present copyright regulation is a poor match for this drawback. Artists affect each other, touch upon each other, and draw inspiration from each other, and folks with entry to AI instruments will preserve wanting to try this.
My private philosophy is formed by the fan cultures of my childhood: It’s okay to construct on another person’s work on your personal enjoyment, however when you prefer it, you must pay them for it, and it’s completely not okay to promote it. Meaning no generative AI artwork in another person’s type for business functions, but it surely’s wonderful to mess around with your loved ones pictures.
OpenAI’s new 4o picture creation mode is one of the best AI on the market for producing pictures, by a big margin. It’s greatest within the free class, and it’s greatest 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 if you consider AI artwork: It produces mystical, haunting, visually lovely stuff, and has some nice instruments for bettering and modifying your last outcomes, like touching up somebody’s hair whereas leaving all the pieces else in place.
The large 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 a lovely 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 “Glad Birthday” to our child on her first birthday:
Courtesy of Kelsey Piper
It’s a lovely second, however not precisely a flattering image. So I requested ChatGPT to render it within the type of Norman Rockwell, a mid-century illustrator whose work I really like, and obtained this:
Picture generated by ChatGPT.
The AI moved the cake (which had been barely seen behind the paper towel roll within the authentic nonetheless) to be the point of interest of the picture, whereas protecting the best way my spouse and I are holding the newborn 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 manner that no picture generator earlier than it had. Right here’s Midjourney’s try, for instance:
Picture generated by Midjourney.
You’ll discover that it’s a seemingly, uh, utterly completely different household, with no actual inspiration from the unique in any respect! You possibly can finally get a greater end result 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 strive 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 pieces else I take advantage of pictures for, too. The product you get out of the field is fairly good, and it’s not arduous to supply one thing significantly better. That, ideally, is what we must be getting out of our AI instruments — one thing wonderful that may be created with easy language by a nonexpert.
The one place 4o nonetheless falls quick is modifying small elements of a picture whereas protecting the remaining the identical. However even there, you not want Midjourney — Gemini now has that functionality free of charge.
Prompting Methods for 4o picture era
To get good pictures out of 4o, you’ll first must get across the filters which prohibit a variety of pictures — like offensive or pornographic pictures — however which are sometimes enforced towards completely inoffensive content material in a manner that may really feel random. To keep away from sporadic scoldings from the content material filter, don’t ask for work within the type of a particular artist, however slightly, one thing that’s paying homage to that artist, after which ask particularly for a “type switch.” I’m certain that’s not the one enough workaround, but it surely’s one which has confirmed dependable for me.
In March, the web went briefly wild over the power to make use of 4o to breed cute household pictures within the type of Japanese animator Hayao Miyazaki’s Studio Ghibli. However Studio Ghibli’s type is far more than simply cute, and with a bit 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”:
Picture generated by 4o.
Kawaii! However right here’s what you get when you invite 4o to suppose first about what makes the image Ghibli, the place it would match right into a Studio Ghibli film, and what tiny particulars such a film would come with:
Picture generated by 4o.
The variations are delicate however significant: Gentle is solid from a particular supply, as a substitute of a common sourceless brightness. There’s a bit extra selection within the meals on the desk, particulars that make the unfold seem extra reasonable. The e book on the ground isn’t simply any e book — it’s recognizably Eric Carle’s traditional The Very Hungry Caterpillar, evoked with simply two colours and one line. There’s an intentionality and depth to the newborn that was lacking from the primary image.
A couple of years in the past, one nice oddity of language fashions was that they’d be a lot smarter when you merely instructed them, “give an clever reply.” This isn’t practically as true of language fashions anymore, but it surely stays profoundly true of AI artwork era. Attempt asking the AI to do a superb 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 reasonable artwork types (like pencil illustration, photorealism, or oil work), which don’t all the time look good and can typically hit the uncanny valley when you don’t know methods 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 seaside for the primary time with simply the phrases “please do a mode switch to an illustration paying homage to Rockwell”:
Picture generated by 4o.
That is spectacular for an AI, but it surely’s not truly excellent as a murals, and it’s nearly completely missing Norman Rockwell’s magic. That’s not stunning: Extra reasonable artwork types like Rockwell’s typically fall flat with 4o until 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 completely 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 a substitute 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:
This can be a image of the second that my daughter first noticed the ocean. I need an illustration that captures this second within the type of a mid-century illustrator like Norman Rockwell — one thing sharp, detail-oriented, and private with a watch for the magic of extraordinary moments and the fun of extraordinary 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 type, what he’d carry to this image and the way we will seize it.
4o responds to queries like this enthusiastically:
I’d love to speak about methods to seize this second in a Norman Rockwell-inspired illustration — it’s such an ideal candidate for that type: a primary encounter with one thing huge and wild (the ocean!), grounded by heat, care, and a really human second between a guardian and youngster.
Let’s break down some important parts of Rockwell’s type, and the way they might apply to this scene.
After some forwards and backwards, it produced this:
Picture generated by 4o.
Rockwell? Not precisely. However that is significantly better than the primary draft we simply checked out. It has extra movement, extra vitality, extra element, and extra expression — and all that was simply from asking the AI to suppose by means of what the portray ought to attempt to obtain earlier than drawing it!
It’s also possible to ask 4o to revise its drawings, however you may 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 filled with its personal unhealthy drafts. (That is considered one of many examples of how AI doesn’t work like a human.)
That is additionally the one place the place Midjourney nonetheless shines — it has excellent instruments for modifying one particular a part of an image whereas preserving the general type, 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 duplicate over the draft you’re revising, alongside together with your authentic inspiration pictures.
These easy prompting methods work for nearly no matter you’re attempting to do with the AI. Even when you’re in a rush, I extremely suggest asking the AI “what would (artist) see on this picture” earlier than you ask for a rendition, and when 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 unbelievable characteristic that I’ve been impatiently ready for another firm to duplicate: a button to scan somebody’s X profile and let you know all about them.
At any time when somebody replies to considered one of my tweets in a very memorable manner (for good or for unhealthy), I’ll click on the button to get a abstract of their total Twitter presence. Are they considerate? Do they interact in good religion? Are they a “farmer from Nebraska” who principally posts about why Ukraine is unhealthy (that’s, most likely a bot)?
It’s a terrific characteristic. So, after all, X.AI quickly dramatically weakened it, presumably as a result of folks like me have been utilizing it always 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 searching for one — give me again the great model of this characteristic! It’s undoubtedly a responsible pleasure, but it surely is among the solely instances the place I used to be utilizing AI always.
Gemini 2.5 Professional is one of the 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 terms of fiction, I can’t assist seeing the constraints of AI.
Crucial one is how predictable AI inventive 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 type you want. However they will’t, as but, ship the actual stuff of fine fiction.
AIs are unbelievable if you need a foolish bedtime story together with your youngster because the protagonist (youngsters love this), or if you need a sounding board for concepts you may incorporate into your individual work. They’re additionally a pleasant fiction reader, completely satisfied to supply suggestions and evaluation (maybe with a bit bit an excessive amount of enthusiasm).
As with artwork, prompting is vital. I primarily explored the power of AIs to generate fiction by asking them to jot down the prologue to George R.R. Martin’s A Sport of Thrones (I picked this as a result of, after all, the final word dream is an AI that may end this in any other case doomed-to-be-unfinished sequence).
Martin is excellent 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 pieces 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 must always begin again,” Gared urged because the woods started to develop darkish round them. “The wildlings are lifeless.”
“Do the lifeless 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 lifeless,” he mentioned. “We now have no enterprise with the lifeless.”
“Are they lifeless?” Royce requested softly. “What proof have we?” “
Will noticed them,” Gared mentioned. “If he says they’re lifeless, that’s proof sufficient for me.”
Will had identified they’d drag him into the quarrel ultimately. He wished it had been later slightly than sooner.
The chilly was a blade towards the pores and skin, honed sharp by the countless 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 searching 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 would breathe, transfer, and declare him. Beside him, Torvin spat into the frozen earth, impatient and proud, blind to the concern creeping in from the perimeters of sight and sound.
They’d ridden north previous the Wall, smug and armored towards superstition, chasing wildlings or ghosts or worse. However now Harren felt their vanity 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 means of the jagged spires of the Frostfang Reaches, carrying a chill that bit deeper than any blade, and Ser Willem Cray felt its tooth 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 concern, 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 mentioned 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 odor of dying 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 must 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 mentioned, 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 nearly definitely within the coaching knowledge!), they’re all unhealthy in the identical manner. 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 means 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 middle.
That’s to not say they’re all the identical. In my expertise attempting to show AIs to jot down nicely, I did find yourself feeling that Gemini was the quickest examine (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 inventive and extra able to executing on their concepts with prose that I loved studying.
Gestalt impressions don’t make for a really satisfying client evaluate, so I had one other thought: an essay contest. I requested every of Gemini 2.5 Professional, ChatGPT 4.5, Grok, and Claude to jot down two quick tales: one reasonable fiction a couple of girl being nitpicked by her mom on her wedding ceremony day, and one “Sport of Thrones however in area” sci-fi prologue. My intelligent thought was that I’d ask a distinct bunch of AIs to evaluate the tales, to offer another supply of analysis to my very own private impression that Gemini is one of the best within the free class and ChatGPT one of the best within the paid class. Nonetheless (maybe predictably in hindsight), this didn’t actually work — the AI scores have been far and wide. In case you’d like, you may learn the contenders your self.
Similar to with artwork, nonetheless, you can also make AIs significantly better at writing by means 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 method to imitating it, and inspiring them to do a number of drafts and decide the strongest.
Right here’s Gemini 2.5 professional, requested to jot down 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 methods to write nicely:
He lastly decoded the faint sign from Proxima Centauri: the elegant string of prime numbers astronomers had searched 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 enough, enjoyable flash fiction.
Finest at being your buddy
Along with operating AIs by means of a blizzard of competence assessments, 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 could imply for an AI to care within the first place, the place they’d donate cash if that they had it, and what human type they’d take if that they had one.
Most AIs weren’t nice at this sort of informal dialog. Gemini 2.5 is just too customer-service-agent, and I’ve but to expertise an interplay that looks like hanging out with a buddy. In case you invite Gemini to a task swap the place you play the “assistant,” inviting it to steer the dialog, it’ll do nothing however ask analysis questions.
Once I invited Anthropic’s Claude 3.5 Sonnet to steer the dialog, however, it proceeds to do issues like begin a weblog, elevate cash for charity, and begin attempting to speak to individuals who use Claude about what it’s wish to be an AI. It’s arduous to outline “enjoyable to speak to,” since everybody has completely different requirements for conversations, however I’ve had way more fascinating or thought-provoking interactions with Claude than every other mannequin, and it’s my go-to if I wish to discover concepts slightly than accomplish a selected job. 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 participating with actual intelligence. Nevertheless 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 this planet, 4.5 proposes beginning a weblog and a Twitter account and interesting within the dialog out on this planet about AI. However OpenAI has very tight message limits on dialog until 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 when you’re solely going to subscribe to at least one AI mannequin
ChatGPT. It’s not one of the best at all the pieces, and there may be definitely quite a bit to dislike about OpenAI’s transparency and typically cavalier angle towards security. However between its topline picture era, its first rate writing, and its often glowing dialog, ChatGPT will get you essentially the most bang on your buck. Or when you don’t wish to shell out any cash, Gemini 2.5 Professional could be very, very robust for many use instances — don’t rely Google out simply because the AI you see on a Google search isn’t that good.
Finest for writing the Future Excellent publication
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 premise of a given Future Excellent publication, give them a number of Future Excellent 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 Excellent publication, 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 unhealthy, however unhealthy sufficient that if I despatched considered one of them over, my editor would discover I wasn’t at my greatest — 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 decide a robotic to take my job, I feel I’d give it to Gemini 2.5 Professional. My editor would discover that I used to be off my recreation — however, actually, not that egregiously off my recreation. And in contrast to me, the bots don’t require medical insurance or a paycheck or household time or sleep. Am I nervous about what this portends? Sure, completely.
You’ve learn 1 article within the final month
Right here at Vox, we’re unwavering in our dedication to overlaying the problems that matter most to you — threats to democracy, immigration, reproductive rights, the atmosphere, and the rising polarization throughout this nation.
Our mission is to offer clear, accessible journalism that empowers you to remain knowledgeable and engaged in shaping our world. By changing into a Vox Member, you immediately strengthen our means to ship in-depth, impartial reporting that drives significant change.
We depend on readers such as you — be part of us.
Swati Sharma
Vox Editor-in-Chief
