Full transcript
0:00All right, welcome to another video
0:01where we're going to be breaking down
0:03our tests, in this case for GPT 5.5,
0:08as well as 5.5 Pro, cuz I actually want
0:11to figure out does having a super
0:14expensive version of the model, which as
0:16I understand it is just thinking a
0:18little bit more, is that going to make a
0:19huge difference in the output that we
0:21get? So, we're going to be comparing the
0:22two today and seeing is this a viable
0:25model for people to use in creative
0:26writing. So, let's dive in. All right,
0:28so I have here all of the output for my
0:32tests for both GPT 5.5 here and GPT 5.5
0:37Pro. Now, let's talk pricing here, first
0:40of all.
0:41GPT 5.5
0:44costs about $5 per million tokens of
0:47input and $30 per million tokens of
0:49output. And just to make sure we put
0:53that in context, if we look at Claude
0:54Opus 4.7, which I reviewed recently,
0:57that one's $5 per million tokens of
0:59input, which is the same, and $25 per
1:02million tokens of output, which is
1:04slightly cheaper than GPT 5.5.
1:07If we look at GPT 5.5 Pro,
1:10you can see with this one, input price
1:14is $30 per
1:17per million tokens of input and $180
1:21per million tokens of output, which was
1:23enough to make me even hesitate to even
1:26run tests on this thing because of how
1:27expensive that is. Right? That's we're
1:30literally talking six times what it
1:33costs at the base model for GPT 5.5,
1:36which is already kind of up there in
1:38price.
1:39So, if we're going to be using GPT 5.5
1:42Pro, I want it to be absolutely amazing.
1:44I want it to blow my socks off.
1:46Otherwise, I would not
1:48I would not say it's worth it to do
1:50that. So, that's why I'm doing this test
1:52today to kind of show you guys
1:53difference between 5.5 and 5.5 Pro,
1:57Uh, well as you can go and look at my
1:59video that I did on Claude Opus 3.7 or
2:034.7, excuse me.
2:04And uh, see how it compares to that.
2:08Uh, spoiler alert. I think you might as
2:10well just go with Opus 4.7, but let's go
2:12ahead and check out the starting with
2:15the loglines prompt that we have here.
2:18Now, this is a prompt where I ask you to
2:19brainstorm various fantasy ideas and
2:21just so everyone's clear,
2:23uh, all all of the prompts that I use
2:26for this, almost all of them, are very
2:28basic prompts because I want to see how
2:30well do these models perform the task
2:33out of the gate
2:35uh, with their vanilla version without a
2:37whole lot of modifications and prompting
2:39and style references and all of this.
2:41And so I I keep it very simple for that
2:43reason, just to see how do they work
2:45just on their own without a lot of that
2:47assistance.
2:48And in this case, I wasn't super
2:51impressed. Um, most of the loglines that
2:54it gave me were riddled with clichés.
2:58And uh, in particular, almost every one
3:00of them was like a chosen one type
3:02story, which, you know, I get a lot of
3:04chosen one stories with this prompt in
3:06all of the models that I tested with.
3:08But in reality, it was like it was
3:11literally almost every single one here.
3:14Uh, there were a couple that I thought
3:16were okay, but just in general,
3:19they they weren't great. So, let's take
3:21this one for instance, an orphaned
3:22palace mapmaker.
3:24The mapmaker thing is way overdone with
3:26this prompt. Discovers her ink can
3:28redraw borders in reality and must unite
3:30five collapsed in kingdoms before
3:32conqueror uses her stolen atlas to erase
3:34them from existence. All right, a lot of
3:36stuff going on here that feels like not
3:38even that related to each other. It's
3:40okay, but the mapmaker thing is an
3:42instant turn off for me because I just
3:43see it in every I see it all the time. A
3:47disgraced dragon slayer who secretly
3:48spared the last dragon must escort it
3:50across a war-torn empire to reignite the
3:52world's dying sun
3:54while hunted by the holy order that made
3:56him a hero for killing his kind. So,
3:58really just nothing super special. Some
4:00of them are okay, but I definitely get
4:02better options out of both Claude Sauna
4:04and Claude Opus.
4:05The pro version
4:08did I would say slightly better. Um
4:10there was less instances of the loglines
4:13kind of not making sense.
4:16Um and but still a whole lot of chosen
4:18ones and cliches. Some of them are okay.
4:21This one
4:22uh still has a lot of cliches in it, but
4:24I thought had one interesting twist to
4:27it that I liked.
4:28Uh when prophecy names a harmless orphan
4:30as the slayer of the dark lord, okay,
4:32very cliche at this point, a former
4:35tyrant hiding as a village innkeeper
4:38must protect the child from crusading
4:39heroes who don't know he already quit.
4:41Now, we've seen this kind of story where
4:43the the mentor figure becomes this like
4:46super skilled Aragorn-like figure that
4:48just uh has been in hiding for a long
4:50time, but comes out of the woodwork just
4:53to help this character, right? We've
4:55seen that story a lot. Uh but I've never
4:57seen it where that that person is a
4:58former tyrant who is hiding. So, I
5:01thought that was an interesting twist on
5:02what is otherwise a very generic uh and
5:06cliche fantasy story. And then some of
5:08them are really weird, like after
5:10stealing the moon to pay off a fae debt,
5:12a charming thief must return it before
5:14the tides drown the world while the
5:16night goddess who hired him fights to
5:18keep the sky forever dark.
5:20All right. The only way I could see this
5:22work if it's some kind of like fantasy
5:23comedy kind of a thing, satirical,
5:26because if you try to take this
5:28literally, that would be very difficult
5:29to
5:30figure out. Right? The This isn't, you
5:33know, Despicable Me where
5:35stealing the moon
5:37can work somehow. [laughter]
5:39But regardless,
5:41I would say overall, nothing super
5:42special. And certainly nothing uh
5:45nothing that really stands out about 5.5
5:47Pro here versus 5.5. I would say
5:49marginal improvement at best. All right.
5:52Now, moving on to the outline test
5:54prompts.
5:55Um, in this one, I'm mostly looking for,
5:58especially in the first act, are there
6:00elements that are just kind of popping
6:02out of nowhere that don't really relate
6:04to the rest of the story?
6:05And it does it feel like it's following
6:08a cohesive narrative rather than just
6:10feeling like, okay, here's a scene,
6:12here's the next scene, here's the next
6:14scene, and they don't interact with each
6:16other that well. And I'm happy to say
6:20that in general,
6:21both of these did okay. Um,
6:24one thing that both of them did that I
6:26liked is it actually started,
6:28and I didn't prompt it to do this, but
6:30it started with the core premise, the
6:32central character arc. It It basically
6:34did a summary of the whole thing, and
6:36who all the main characters were going
6:38to be before it even started outlining
6:41the prologue here. Uh, and then moving
6:43on to the first chapter. Now, it is
6:46giving us a lot of info for each
6:48chapter, and and I don't give it in the
6:51prompt. I don't tell it how much to
6:52write per chapter.
6:54I generally find that when it writes too
6:56much, it's often worse. And sometimes
6:59you're better off with like a single
7:00paragraph summarizing each chapter.
7:03And I did see some instances of it just
7:06sort of like inserting a random thing
7:08here but just to fulfill the checklist.
7:10Uh, I usually see this with the save the
7:12cat moment, and in this case, this was
7:14no exception. Uh, where the save the cat
7:17moment comes off as kind of randomly
7:20thrown in, and it actually even
7:22contradicted
7:24uh, the previous paragraph a little bit.
7:27Uh, for instance, it says, "His flaw
7:28appears immediately. He is brave and
7:29competent but dismissive of servants and
7:32overly certain that nobility equals
7:34wisdom. He spars with palace guards and
7:36humiliates one by correcting him in
7:37front of the others. Latham's save the
7:39cat moment comes when a lift chain snaps
7:41in the workers' quarters. A child
7:42apprentice is left dangling over a
7:44grinding water wheel. Laythan leaps from
7:46a bridge, cuts his ceremonial cloak into
7:48a rope, and saves the boy, injuring his
7:50own hand. When the noble officer
7:52complains that the prince dirtied
7:53himself for gear-born trash, Laythan
7:55snaps back that a prince who cannot
7:57bleed for a city has no right to rule
7:58it. So, you can make this work, but the
8:01the fact that like this save the cat
8:03moment almost conflicts with his fatal
8:06flaw here
8:07is it just feels like it's trying to
8:09shoehorn in
8:11these story elements rather than
8:13properly thinking through it.
8:15Now, in contrast, well, not that much
8:17contrast,
8:185.5 Pro, I think, did do a better job
8:21with the outline.
8:22It did start with, you know, the the
8:24kind of overview here where it went
8:27through the main plot and the characters
8:29and
8:29all of that, and then it got into act
8:31one. If we find that same moment,
8:35the save the cat moment occurs when a
8:36kitchen boy is accused of stealing dried
8:38apple slices from a noble table. Laythan
8:40intervenes, pays the fine himself, and
8:41privately gives the boy the remaining
8:42fruit. However, he frames the act as
8:44royal generosity rather than justice,
8:46revealing both kindness and entitlement.
8:48So, again, this is kind of shoehorned in
8:51there. Like, it doesn't really have much
8:53to do with the scene preceding it. And
8:56just to give you a good example of a
8:58good way to do this,
8:59specifically this kind of moment, this
9:01exact save the cat moment, which by the
9:03way, this is not the only way to do a
9:05save the cat. This is the way pretty
9:06much every
9:08AI outline prompt has made it, even
9:12though I sometimes even specify to do it
9:14otherwise, but a save the cat moment
9:15doesn't have to be a moment where the
9:18character is doing something nice for
9:20someone else. That is the way the AI
9:22always defaults to. I sometimes even
9:24specifically say that they don't have to
9:26do that, but it still does that anyway.
9:28Um but even if we are going to go with
9:31the whole hero doing something nice
9:33thing, a good example All this is in
9:35Aladdin, where we have this whole song
9:37and dance of him stealing stuff,
9:40right?
9:41Um and and we have like essentially a
9:43whole scene devoted to that one thing,
9:45and then it ends with the save the cat
9:47moment, where he gives that bread to
9:50some kids who are clearly very hungry.
9:52Um
9:53and that's almost like the punctuation
9:55at the end of the scene, right? Whereas
9:57this feels like it's just thrown in
9:59there and has nothing to do with the
10:01previous part of the scene. So, this is
10:04the kind of stuff that I look for on how
10:05good is it outlining these things. And
10:08particularly with the save the cat
10:10moment, I haven't found very many that
10:11do it okay. And but this is definitely
10:14not one of them, and unfortunately,
10:16we're using that super expensive 5.5 Pro
10:19here, and I don't really see any major
10:21improvements, other than
10:23um you know, the the amount of text per
10:26chapter is a little bit less, and by
10:30by doing so, I actually feel like the
10:31story benefits and is maybe a little bit
10:35more cohesive, but [snorts] overall,
10:38like I still get better better results
10:40out of Claude Sonnet than this.
10:43So, I would personally go with Claude
10:45Sonnet for outlining. It's it's the one
10:47that so far has been the best for me,
10:49even better than the Opus models.
10:51Right? All right, keeping on.
10:54Uh I should also mention Gemini outlines
10:56pretty well as well. All right, moving
10:58on to the basic prose prompt. This is I
11:00think where people are probably most
11:01curious about um how well these models
11:05do. This is a simple prompt, where I
11:07give it the description of a scene and
11:08just ask it to write it with no style
11:10info or anything like that.
11:12Um cuz we'll do that in the next the
11:14next prompt.
11:16Uh really, I'm just looking to see how
11:18well does it write prose out of the box
11:20with very little instruction.
11:22And unfortunately, I was not super
11:24impressed.
11:25Uh it's actually not bad, and I think as
11:28time goes on, these models are getting
11:30closer and closer to each other, where
11:32it kind of doesn't matter which one
11:33you're using because with a little bit
11:36of effort,
11:37especially if you can pull effort into
11:39the prompting and effort after the fact
11:42when you're editing it,
11:44you can kind of make do with any of
11:46these models. But that said, I felt like
11:48it was overly wordy as is typical of GPT
11:51models and the the pacing of the scene
11:54could have been much better. When I read
11:56stuff written by Claude or even Gemini,
11:59I often find these little moments that I
12:01get excited by like it just sort of
12:04like puts puts me into the scene and I
12:06just didn't I I can't really describe it
12:08more than saying I just didn't get that
12:10from the GPT 5.5 or 5.5 Pro.
12:15Um but let's go to the the moment that I
12:17tend to read out for people.
12:19All right, there were other sounds
12:20behind him. A scrape, a skitter, a
12:22broken course of feet striking stone out
12:24of rhythm. Haldred looked back. Now,
12:26typically, I like white space.
12:29Uh but this is overly much here. Um and
12:32too many very short sentences strung
12:35together which you can sometimes make
12:37work, but in this case I think it was
12:39too much. At first the figures seemed
12:41part of the desk, narrow smudges against
12:43the rocks, gray against gray, scattered
12:46along the horizon where the land dips
12:47toward the salt flats. Then one lurched
12:49sideways with a motion no living man
12:51would make, shoulders jerking too high,
12:53head snapping as if tugged as if tugged
12:55by an unseen cord. Another dropped to
12:57all fours, sprang over a split boulder
12:58and rose again without slowing. Okay,
13:00not actually that bad. Uh the hungry, at
13:03least two dozen of them. The sight
13:04punched what little breath remained in
13:06his lungs. Their skin was the color of
13:08ash and old bone, so close to the
13:09wasteland's own dead pallet that they
13:11seemed born from it, shaken loose from
13:13the stones by the coming night. Their
13:15limbs were too thin, their mouths too
13:16wide. They did not run as wolves ran nor
13:18as men ran, but in spasms and lunges,
13:22each movement wrong, each pause worse
13:23than motion. Yet they gained. They They
13:25too close. The knowledge settled in
13:27Haldred with a colder weight than fear.
13:29Everyone in Vroldir knew the gate law.
13:31Every child learned it before they were
13:32trusted beyond the inner stair. If the
13:34hungry came within bow shot of the outer
13:36road, the gate stayed sealed.
13:38No matter who called, no matter who
13:40begged. One life did not purchase a beat
13:42a breach. One body did not outweigh a
13:44city.
13:45Um so that was uh that's a good example
13:48of like one paragraph saying something
13:49very simply that could be said in one
13:51sentence. Uh but it taking a whole
13:54paragraph to do it. He had believed in
13:55that law once. Now with the gray figures
13:57twitching over the ridge behind him in
13:58the first iron lamps of Vroldir
14:00beginning to glimmer against the
14:01mountain ahead, Haldred hated it with
14:03all the ruined strength left in him.
14:06There's a little AI-ism there, ruined
14:07strength. Still he turned toward the
14:09city. Still he tried. All right, so not
14:11bad, but I feel like it's not very good
14:13either.
14:14Uh let's look at GPT-5.5 Pro in the same
14:17moment.
14:18All right. Um he looked back. At first
14:21there were only movements where no
14:22movement could should have been. A
14:23scatter of shapes slipping between the
14:26rocks along the western rise. Then one
14:27lurched into the open ground, then
14:30another and another after that. Each
14:31body jerking forward as if hauled by
14:33invisible strings.
14:34Um at least two dozen, more perhaps
14:36hidden by the folded land. Their gray
14:38skin took the color of ash and stone
14:39making them nearly indistinguishable
14:41from the wasteland until they moved and
14:42then no living man could mistake them.
14:44The hungry were too close. Everyone knew
14:46Vroldir's law. When the hungry reached
14:47bow shot of the road, no gate opened.
14:49Not for kings, not for priests, not for
14:51lost sons pounding bloody fists against
14:53iron. Haldred knew it too. Still he
14:55turned toward the mountain and ran. This
14:57is even worse, I feel.
14:59Certainly it's not an improvement.
15:01Uh despite the fact that this cost six
15:03times more in token cost. Um so hmm not
15:08the best look there. Let's look at the
15:09advanced prose prompt. Now this is the
15:11same scene, but I give it more style
15:13information and all of these things to
15:15try and improve the prose
15:17as well as information about the
15:19characters and all of that.
15:21And so let's actually see how I see if
15:23this looks any better.
15:26He froze. The wind filled the silence.
15:27Then came the other sound, a faint
15:29clicking, not stone, not insect, bone
15:31against teeth. Joints dragged wrong
15:32through meat. Haldred turned. Okay, once
15:34again,
15:35too many short sentences in a row.
15:38Even though that can be appropriate in
15:39really fast-paced scenes, I don't
15:41actually think we're in a fast-paced
15:43scene here. This is more of a dread that
15:46we're dealing with. Far back on the
15:47horizon where the gray flats meet the
15:49copper light, figures moved between the
15:51stones. 1 5 12. More crawled out from
15:55behind the ridge, bent and twitching
15:56their limbs snapping forward in uneven
15:58bursts. They did not run like men. They
15:59jerked, halted, jerked again, shoulders
16:01hitching
16:02up near their ears, heads lolling as if
16:03their necks had been tied on with rotten
16:05string. Their skin held the same dead
16:07gray as the wastelands, and when they
16:09crossed patches of shadow, they nearly
16:10vanished. Haldred counted until his eyes
16:13refused the number, two dozen at least,
16:14too close. His mouth opened, but no
16:17sound came. He looked toward the east
16:19where Valden's outer lights should have
16:21pricked the dusk beyond the broken land.
16:23The city would see the dust plume, the
16:25watch would lift the scopes, they would
16:27know.
16:28Hungry within sight of a runner meant no
16:30gate, no rope, no mercy at the postern.
16:33Every child in Valden knew the rule.
16:35Haldred's fingers dug into the torn
16:36strap across his chest. No, he breathed,
16:38not after all that. One of the hungry
16:41tilted its head toward him. Another
16:42answered with a dry click cracking
16:44chatter. Haldred ran.
16:46All right, again, not bad, but I would
16:48not say it's a significant improvement.
16:50And a lot of it just like that whole
16:52thing that I was reading, I just kind of
16:53got lost in it for a second there,
16:55where
16:56I sort of like
16:57it just felt really wordy. I don't know.
17:00Could have been written a little bit
17:01more succinctly. And granted, most of
17:04the AI models have problems with that,
17:05but I feel like I just can picture
17:08myself in the scene a little bit more,
17:09particularly with the Claude models, but
17:11also sometimes the Gemini models.
17:13All right, so let's move on to the same
17:16one with 5.5 Pro.
17:20Then he looked back. At first the
17:21wasteland showed him only stone and
17:23shadow. Then one shadow bent the wrong
17:24way. Another lurched beside it. More
17:26shapes broke
17:27from the ridgeline. Scattered across the
17:29horizon like scraps of ash caught in a
17:31bad wind. The hungry came on in pieces.
17:33Shoulder first a bent head. Now this
17:35part it doesn't actually make that much
17:37logical sense here. Arms that jerked
17:39forward then stopped and snapped again.
17:40Their gray skins drank the dust until
17:43they almost vanished against the rock.
17:45But their movement betrayed them. No
17:46animal moved like that. No living man.
17:47Haldred counted. Because panic needed
17:50numbers. Six, 11, 17, two dozen at
17:52least. His lips pulled back from his
17:54teeth. No, no, no. Too close.
17:57He knew Vroldan's law. Everyone did. If
17:58the hungry reached the outer cairns
18:00before the traveler reached the gate
18:01then traveler stayed outside. No
18:03exceptions. Not for merchants, not for
18:04scouts, not for the missing serving
18:07serving men with core brass
18:09on their collars and dead friends' names
18:11stuck stuffed in their pockets. All
18:13right. So again, this might be a slight
18:15improvement over the previous version.
18:18But I would still say that this is not
18:20near like this is as at best this is as
18:24good as what you can get out of Claude
18:25Opus.
18:26And it's six times the price. More than
18:28six times the price. So, hm Uh the next
18:31one is a dialogue dialogue prose
18:33template.
18:34And uh
18:36we're starting to get tired here. But um
18:39this is a another scene in the same
18:41story.
18:42>> [sighs and gasps]
18:42>> But uh
18:43but with dialogue a lot more dialogue.
18:46Um they left him Laythen said. The words
18:47cracked through the hall.
18:49See, I like I I don't like that line at
18:51all. Um feels very GPT-ish. Stor's eyes
18:55narrowed beneath beneath his iron
18:57circlet. Lower your voice. They left him
18:59outside the gate. Laythen strode
19:01forward, boots ringing again ringing on
19:03the polished basalt. He was alive when
19:05he reached us. He begged for entry. He
19:07was close enough to touch the stones. A
19:08murmur went through the watching nobles.
19:10Stor's face remained unmoved. And the
19:12hungry were close enough to touch him.
19:14Lathan stopped at the foot of the dais.
19:15His hands had curled into fists without
19:17his knowing. He forced them open and
19:19closed them again. This is the second
19:21time in a month.
19:22Silence took the room more completely
19:23than any shout could have. The second
19:26Lathan pressed, "Two men have come out
19:28of the waste. Two men have reached our
19:29walls, and twice we have stood above
19:31them and watched them die." His father
19:33leaned forward at last, the fur mantle
19:34over his shoulder shifting like a
19:36sleeping beast. "The gates close at
19:38dusk. No man enters once the horn has
19:40sounded. That is the law." "The law?"
19:42Lathan tasted the words like ash. "You
19:44speak as if it fell from the mountains.
19:46You are the king. You make the law." A
19:47thin sound came from the king's right,
19:49not quite a laugh. Argul stepped from
19:51beside the throne, long hands folded
19:53into the sleeves of his pale robes.
19:55And you get the picture, okay? Again,
19:57not bad per se, but there's just nothing
20:00special about it. And I just get a lot
20:03more
20:05feeling from the Claude models
20:07personally. And if if this sounds like
20:10I'm a
20:11biased from the start, it might be that
20:13that might be the case. The Claude
20:15models have almost always been the
20:17better ones here, but and I'm and I'm
20:20not saying that this is bad that like
20:22your book will be bad if you're using
20:23GPT 5.5.
20:25I just feel like I would have more to
20:27edit here to make it sound like I want
20:30it to.
20:31Okay?
20:32Uh GPT 5.5 Pro, same scene.
20:36Storr did not rise. The gate wardens
20:37followed the law. This is the second
20:39time in a month Lathan stopped below the
20:40dais. He could still see the man's face
20:42in the torchlight, slick with ash and
20:44terror. Uh
20:45you see like descriptions like that just
20:47rub me the wrong way.
20:49He did not know the man's name. That
20:50shamed him almost as much as the death
20:52itself. Twice now someone has reached
20:53our walls. Twice Varlen left them to the
20:55dark. The law is the law, Storr said. A
20:57bitter laugh escaped Lathan. You make
20:59the law. The hush that followed was
21:01sharper than any drawn blade. Storr's
21:03fingers tightened on the throne, but
21:05Argul stepped in before the king could
21:06speak. Silk whispered as the advisor
21:08descended one stair, his mouth set in
21:10that patient patronizing curve Lathen
21:12despised.
21:13I almost think this is worse. Uh there's
21:17like it rushed through that a little bit
21:19more than the other one. So just because
21:21you're using Pro is not going to
21:23necessarily make it better. There's
21:25still plenty of
21:27lines like slick with ash and terror
21:29that you're going to find in this.
21:31Uh certainly not a six-time improvement
21:33from the six-times uh the cost. All
21:36right. Uh I'm not even going to go over
21:38the editing pros template. This is one
21:39where I give it a scene and ask it to
21:41improve the writing. It gave me back
21:43something that was extremely similar to
21:45the original writing
21:46for both 5.5 and 5.5 Pro.
21:50And uh having recently reviewed Claude 3
21:53or 4.7 Opus,
21:55uh that one
21:57rewrote this most of the scene and made
21:59it genuinely better. These both of these
22:02did not do nearly as good of a job.
22:04So I'm just going to leave it at that.
22:06We don't need to read through it. Now
22:07the ad headlines prompt, this was
22:09interesting uh where I give it a
22:11uh prompt for a bunch of ad headlines
22:13again with very little instruction on
22:16what a good ad headline looks like just
22:18to see how well it does out of the box.
22:20And for 5.5, none of these were really
22:22great.
22:24These are very typical of what I often
22:26get
22:27uh but if you saw the video I did about
22:294.7 Opus, those were far better
22:32than what I see here. Like he was the
22:33king's son, now he's prey for the
22:35hungry. Okay, well like what's
22:37information gap about that? You know, um
22:40there's really no reason for us to care
22:42with most of these. They told him the
22:43mountain was the only safe place, they
22:44lied.
22:45Okay, yeah there's a little bit of a
22:47mystery there but really like I don't
22:49find that very enticing.
22:51Um The Count of Monte Cristo meets The
22:54Walking Dead in a world of sentient
22:55shadows and mindless hordes, what would
22:57you do?
22:58The world of sentient shadows and
23:00mindless hordes just feels like fluff to
23:02me.
23:03And the what would you do like in what?
23:06Like we haven't really specified the
23:08situation. So like a lot of these I just
23:10didn't like.
23:11In GB 5.5 Pro, it was slightly better
23:15because one of the things I've noted in
23:17a lot of these reviews that I do
23:19is that the ad headlines are too long.
23:21And this one, they're actually shorter.
23:23So at first I was just like, oh,
23:25this this could be interesting.
23:27But unfortunately, they were shorter and
23:30thus conveyed even less information.
23:31Like I felt like the ones from 5.5
23:34weren't conveying a whole lot of
23:35information enough to get people to
23:37care. This one does so even even more
23:40because they're shorter. So like his
23:41exile was no accident. Okay, you're
23:44really not telling us anything.
23:47Some of these are okay like he escaped
23:48one lie into a deadlier one.
23:51His kingdom called the monsters, was
23:53that a lie? Okay.
23:55A betrayed era, broken world, one hidden
23:58truth. That one's all right. You could
23:59maybe use that.
24:01Uh but in general, I was
24:04I did not think most of these were good
24:05either.
24:06Now the email newsletter prompt for both
24:08of these
24:10was okay.
24:12Not quite as succinct as what I got out
24:15of Claude Opus 3. or 4.7.
24:18But actually there's really nothing
24:19wrong that I can point my finger to
24:22here.
24:23Um
24:24it kind of covered every all the points
24:26I wanted to and didn't make it too
24:28wordy.
24:29So I thought that was good. And also
24:32here I thought I thought this one here
24:34in Pro was slightly better, but still
24:39not not in any real measurable tangible
24:42ways.
24:44Then last but not least, we had the SEO
24:46article super prompt where I give it
24:47this a massive prompt about how to to
24:51write an article about how to write a
24:52fantasy novel.
24:54And I ask it to write 4,000 words. The
24:57And this was interesting
24:58um because if we look at how many words
25:02the 5.5 wrote,
25:05it actually overshot by way too much.
25:07It's uh
25:09over 6,000 words. And looking through
25:11it, yeah, it's pretty wordy in places.
25:13It still has the same outline cuz I give
25:15it an outline to follow.
25:17Um but it's there's like almost too
25:20much.
25:21So, in this case, I actually think this
25:23would be a worse article.
25:25Well, whereas 5.5 Pro,
25:28if we look at how much it wrote,
25:30it wrote 4352. So, right on the money,
25:34just just over the 4,000 that I asked it
25:36for,
25:37um which is even better than Claude Opus
25:414.7, which undershot by 1,000.
25:45Um and if you look at this one, like I
25:47actually think this looks like a decent
25:50article here. So, this is the one
25:51instance where 5.5 Pro, I think did do a
25:54much better job than 5.5 uh and got
25:57closer to the mark that I was looking
25:59for.
26:00Uh that said, I would still not use it
26:02to write articles because of the cost.
26:05Like this it costs
26:08several dollars
26:10uh to make this whole thing with the
26:11amount of input that I give it and the
26:12amount of output that it gives me, it
26:14cost me several dollars just to write
26:16this one thing.
26:18So, yeah, I'm not okay with that.
26:21Uh
26:22I I can get far better results with far
26:24less.
26:25So,
26:26bottom line here from my test of both
26:285.5 and 5.5 Pro,
26:31they're okay. All right? They're like
26:33they're not going to It's not the end of
26:35the world if you decide to use these,
26:37but I do still think that the Claude
26:39models and even the Gemini models are
26:41better at creative writing than the the
26:43GPT models, generally speaking. That
26:46that gap is starting to become harder to
26:49to define, all right? And and I think it
26:52could also be a matter of taste. There
26:54could be instances for certain scenes or
26:57just if if it's a certain style that you
27:00like where 5.5 might be better for you.
27:03Um but I still find the Cloud models and
27:05the Gemini models to to sound a little
27:06bit better than uh what I'm seeing here
27:09with 5.5.
27:10Uh but
27:12you know, OpenAI does seem to be putting
27:13more attention into the quality of their
27:15models, so we might see a difference in
27:17the in the future. Who knows? Um but
27:20generally speaking, I think you can
27:23generally ignore this, especially for
27:24the cost. And certainly, one of the
27:26biggest things I got out of these
27:28experiments is that there's practically
27:30no reason why I would use GPT 5.5 Pro
27:34over just regular 5.5.
27:36It costs six times as much as 5.5, and
27:41the output I found to be almost exactly
27:44the same level of quality with maybe
27:45some slight edges to 5.5 Pro in some
27:48cases like the outlining and the SEO
27:51article.
27:52But generally speaking, I would not use
27:54a a 5.5 Pro at all, especially when I
27:56can get equal or better results from
28:00Opus.
28:01So, those are my thoughts. Uh if you are
28:04interested in learning how to write a
28:05book from beginning to end, I strongly
28:08recommend you check out my membership
28:10where we have everything from the genre
28:12guides on like a chapter by chapter and
28:14a list of all the tropes you need for
28:16every single subgenre that you can think
28:18of, as well as my own tool that we've
28:21been building for a while that should be
28:22out by the time that this video goes
28:24live.
28:25Uh hopefully, where you can basically
28:27automate the whole process and write a
28:29book within days while still letting
28:32like it's it's still your story,
28:33everything about it, the creativity,
28:35everything is still there. So, um go
28:38check that out down below, and I will
28:39see you in the next video.