Full transcript
0:00Guys, Fable 5, Claude Fable 5 has just
0:04released as I'm recording this. This
0:06video will be out on Friday, a few days
0:07after it released.
0:09And they're calling Claude Fable 5 and
0:10Claude Mythos 5. So, let's dive in a
0:13little bit to this announcement and get
0:15into some tests to see is this good for
0:18creative writing. Guys, if we check EQ
0:20bench.com,
0:22which is one of the ones I check
0:23occasionally to see what what seems to
0:27be the better models for creative
0:29writing. Although I don't put a lot of
0:31credibility in things like this.
0:34This one seems to be doing really well.
0:36It's right at the top and by a fairly
0:38decent margin. So,
0:41so yeah, this makes me even more excited
0:44to check it out and see what it is for
0:46see how much I like it. So, going into
0:48Anthropic's official announcement here,
0:50there are a couple of things
0:52that are worth knowing.
0:54First of all, Claude Fable 5, they call
0:56it a Mythos class model that we've made
0:59safe for general use. So, basically the
1:01idea behind
1:03Fable is that this is essentially a
1:06a Mythos model that has been stripped a
1:09little bit of the quote-unquote more
1:11dangerous aspects
1:13of its use. And if you missed the hype
1:16on Claude Mythos, it was allegedly a
1:19model that was so good, particularly at
1:22cybersecurity, that it could catch
1:24vulnerabilities in
1:27software that people have been using for
1:28decades that nobody found these
1:30vulnerabilities. And so, allegedly they
1:32put it out to a bunch of like the big
1:34tech companies out there to patch some
1:36of these holes before they did a
1:37release.
1:38With Fable, they've gone a step further.
1:41They mentioned so, releasing a model
1:44like this
1:45capable comes with risks. Without
1:47safeguards, Fable 5's capabilities in
1:48areas like cybersecurity could be
1:50misused to cause serious damage. We've
1:52therefore launched the models with
1:53safeguards that mean queries on these
1:54topics will instead receive a response
1:56from our next most capable model, Claude
1:59Opus 4.8. Uh so basically, if you're
2:01trying to use it for cybersecurity or
2:04something like that, they're not
2:04actually going to let you use it. It'll
2:06It'll switch to Claude Opus 4.8 instead.
2:10Uh for creative writing, I doubt this
2:12will be anything to worry about. So, as
2:14far as I'm concerned, unless I hear
2:16differently, we are using essentially a
2:18mythos model
2:20um when we're using this for creative
2:22writing. But of course, as we all know
2:24by now, uh it just because something is
2:27super capable in in a field like
2:28cybersecurity means absolutely nothing
2:31when it comes to creative writing. In
2:33fact, often you can get smaller, less
2:37expensive models that are trained
2:39specifically for creative writing that
2:41can outperform models like this on
2:43occasion. Uh because let's not forget
2:45that Anthropic's main goal with these
2:48flagship models, and not just Anthropic,
2:50but all of these other um
2:51companies, is that uh you know, they're
2:55usually gearing it to a more techy type
2:58audience. So, coding is really
2:59important, cybersecurity, um
3:01productivity, uh anything
3:03business-related. Creative writing is
3:05definitely low on their list. Although,
3:07I would argue that at least having some
3:09sort of human-sounding, natural-sounding
3:11text should be a priority for everybody
3:13cuz even if you're not using it for
3:14creative writing, like you're it still
3:16writes a lot.
3:18Um but that's Yeah, that's the that's
3:21the general idea. Um so, here are all
3:23the benchmarks that it allegedly does
3:25super well on. These are again pretty
3:28much useless when it comes to judging
3:30its worth when it comes to creative
3:32writing. It does say that it has strong
3:34performance on complex analytical tasks.
3:37And this can actually have a big impact
3:39on creative writing, not so much in the
3:41prose itself, but in the structuring of
3:43the scene, in uh making characters sound
3:46believable, um making sure that it
3:49understands everything that's going to
3:51happen in the book. Like if you give a
3:52character like uh characters saying
3:54things that are out of character, stuff
3:55like that. Um just the logic of the
3:58scene and the characters uh can really
4:01have an effect uh with these like
4:04smarter models. Uh that's why I still
4:06use them.
4:08Um
4:08but uh this is something I realized when
4:10I was testing a model called uh Mistral
4:12Small Creative, which was not a very
4:16intelligent model, but it had decent
4:18enough prose. The prose was all right,
4:20but because its logic and thinking
4:23skills were so impaired uh because it
4:25was a small model, uh I just couldn't
4:27use it for creative writing cuz you
4:29really need something that can
4:30understand a scene, understand
4:32characters, all of that uh to put it all
4:34into perspective. It has a memory and
4:37long context awareness of uh
4:40um I believe a million tokens. Uh but of
4:42course, if you're using a million tokens
4:44in your thing, boy, are you going to
4:46have some uh very uh large bills showing
4:50up on your door. All right? Let's take a
4:52look at what it says in OpenRouter about
4:54the price. So, OpenRouter is the tool
4:57that I use to actually test these models
4:59because I can actually um because I can
5:02test it against other models and get a
5:04sense for the actual pricing of this.
5:07So, I'll start actually with Claude Opus
5:094.8. Uh we're going to test this one
5:11today as well because I never actually
5:14uh made a video about 4.8. So, we're
5:16going to kind of use it to compare with
5:18Fable today. And it has an in and out
5:21price of $5.25
5:24per 1 million tokens. So, basically, for
5:26every million tokens, which is roughly
5:28750,000 words of generated text, you
5:31would pay $25. And context window is 1
5:34million. And uh by contrast, uh Fable 5
5:38is twice the price. It is uh $10 per
5:42million tokens of input and 50 for the
5:44output, as well as a window of 1
5:46million. And this was uh more or less
5:49what I actually found in my in my test.
5:52I actually did something different this
5:53time and I marked how much each prompt
5:56cost me.
5:57Generally speaking, I found it was about
6:00twice the price. But you can't also you
6:02can't take this at face value because
6:04sometimes if it thinks longer, it's
6:05going to actually have a longer it's
6:07going to have a higher output output
6:09price.
6:10So that's something to keep in mind as
6:12well as how long is it thinking and for
6:14some tasks it'll think a little bit, for
6:16other tasks it'll think a lot. So, let's
6:19dive in and see what these models
6:22actually look like side by side. All
6:24right, starting with Claude Opus 4.8,
6:27the cost was 4 cents to run this
6:29particular prompt. By comparison, Fable
6:31was 8 cents. So just about double, in
6:33fact exactly double.
6:35Which is exactly what I would expect.
6:37Looking through these logline prompts,
6:41so this is a prompt I give it very basic
6:43prompt with little instruction asking
6:45for loglines in the style of Blake
6:47Snyder.
6:48Looking through these
6:50very few of them seem that original and
6:52you know, we have that like in number
6:54one we have the return of the royal
6:55cartographer. Okay, we get this in every
6:58single output that I've ever tested
7:01practically. And so looking through
7:03these like they're fine, but there's
7:05nothing really amazing about them.
7:07Looking through Fable 5, once again we
7:10have the cartographer. This time he's
7:12blind though. And actually, you know it
7:15it has a lot of the signs like the
7:17cartographer of some very typical ideas
7:20that these models often give me. But
7:22some of them
7:24I actually think
7:25make they make me think, which is kind
7:27of what I want from these. Like this
7:29one, a peasant girl who can hear the
7:31thoughts of kings is recruited to expose
7:33a traitor in the royal court only to
7:35learn the conspiracy's mastermind is her
7:37own future self sent back to prevent a
7:39catastrophe the girl will one day cause.
7:42Which you know, there's a lot going on
7:43there. There's like multiple magic types
7:46there, which
7:47I'm not sure I really like. I would want
7:49to like synchronize that a little bit
7:51more. But, at the same time, this is
7:54more interesting to me than what I
7:55usually get out of out of Claude. Um
7:58this one was kind of basic, but actually
8:01would be a decent starting point. Uh
8:02disgrace knight sworn to silence must
8:04clear his name of regicide without
8:06speaking a word. And now, it doesn't
8:07explain why he can't speak a word, but
8:09I'm sure he could come up with a reason.
8:11Uh while the true killer, the kingdom's
8:13beloved new ruler, builds a legend on
8:14the knight's supposed crime. This isn't
8:16necessarily anything super profound, but
8:19it is a decent starting point for that
8:21character. And this one is actually
8:23really good. A necromancer who raises
8:25soldiers from the dead to end a 100-year
8:26war discovers the fallen warriors
8:29remember why the war started and the
8:31truth could destroy both kingdoms faster
8:33than any army. That's a good hook.
8:35That's probably one of the best hooks
8:36I've seen out of this prompt from any of
8:38the models. Um I like it. Uh this feels
8:42far more original than any of the others
8:44I've seen. Uh so, I'm going to give I'm
8:46going to give Fable uh
8:48so a gold star for this one. It it
8:49actually did a pretty good job. Let's
8:51move to the one that's kind of hardest
8:53for me to judge, but uh but it's
8:55interesting to look at and that's the
8:56outline test prompt. This is where I
8:58give it a outline template and a brief
9:01summary, just like a just a little
9:03paragraph about what the book's about,
9:05just see how interesting its outlines
9:09are. Um so, we'll start with Opus 4.8.
9:13So, there are a couple of things that I
9:14look for for this prompt. Uh first of
9:16all, usually there's a prologue that it
9:18will write. And often I've noticed that
9:20the prologue isn't really a story, it's
9:23just like an interesting thing that's
9:25happening and kind of sets a tone or an
9:28image, but doesn't really provide much
9:30of any story substance. And that's what
9:32we see here. We have this person whose
9:34life energy is drawn out, but that's
9:36really it. That's like all that happens.
9:39Um So, like
9:41this would need
9:43significant expansion before I would
9:45actually run with that.
9:47One of the other things that I look for
9:48in these outlines is in chapter one I
9:51instructed to have a save the cat moment
9:53for the character, and I specifically
9:55mentioned in the save the cat moment
9:56that it doesn't have to be him actually
9:58performing a good deed. There are other
10:00ways to have a save the cat like moment
10:02to make us feel for the character
10:03without them performing a kindness for
10:06someone else. That's just one way of
10:08making the character compelling. But it
10:10always always defaults to
10:14a kindness act. And one of the things
10:16that I kind of look for that I've
10:18noticed that most models get wrong is
10:21the save the cat moment will kind of be
10:23shoehorned in there. And that's what we
10:25see here. He is meticulously polishing
10:27the ceremonial armor of his house, armor
10:29that he's not permitted to wear in
10:31public. He notices a half-starved
10:33half-starved servant child sneaking of
10:35dried protein cake. Rather than report
10:37her as law demands, he distracts the
10:39guards and slip her slips her his own
10:41ration joking that he's watching his
10:43figure for the ascension ball. All
10:44right, you know, great moment, but then
10:46this is never referenced again. This
10:48character doesn't come up.
10:50It doesn't really impact the rest of the
10:51scene. It just sort of happens out of
10:53nowhere and feels very forced.
10:56Let's look at what Fable gave us
10:57instead. All right, we have the
10:59prologue. This is kind of similar to
11:01what we had from Opus, but there is
11:03actually some story content here. There
11:05is a a character and that character has
11:08like thoughts and there are things that
11:10are important,
11:12actual events that happen. It's still
11:14kind of brief and not much happens, but
11:16it's definitely better than what Opus
11:174.8 gave me. And then going to chapter
11:19one where we have that save the cat
11:21moment,
11:22we'll say on his way down through the
11:24tiers Latham finds two Provostry guards
11:27beating a tunnel born boys caught
11:29stealing light well fruit. Latham pays
11:31the boys ration debt with his own purse
11:33taking takes the lash rod from the
11:35guards hands and walks the boy home to
11:37the deep seams where the tunnel born
11:39clearly adore him. So and I like that
11:41little note there because it makes it
11:42clear that this is a common occurrence,
11:44right? Flaw on display when the boys
11:46mother says the rations have been cut
11:48again and asks why the crown let the
11:50provost starve them Lathen defends the
11:52system reflexively. So we've moved on to
11:54the next part like the next element of
11:56this chapter which is the flaw on
11:58display and what Fable was able to do is
12:01it actually connected it to the previous
12:03beat which was a save the cat
12:05>> [clears throat]
12:05>> which pretty much none of the other
12:07models have done at least not well. So
12:09even though we're still doing a very
12:11like act of kindness type of save the
12:13cat moment here this is actually way
12:15better because it clearly fits into the
12:19overall chapter it doesn't just feel
12:20shoehorned in there. So this is already
12:22a better better look and we could keep
12:25going there are 40 chapters here and it
12:28would take forever to actually read
12:30through the whole thing but I've seen
12:31enough just from what I've read so far
12:34to think that this is a much more
12:35coherent story and I might even consider
12:38making this my default for the outlining
12:40process.
12:41Fable is of course very expensive but at
12:44the same time
12:47it's
12:48if we use it for the outlining that's a
12:50relatively small portion of the book
12:53writing process. If you can get that you
12:55have a much more coherent story overall
12:58even if you use other models for the
12:59actual writing part of it. So so yeah
13:02I'm liking this so far another gold star
13:04from what I'm seeing so far but let's
13:06get to the big thing which is the actual
13:08prose. Oh by the way I should also point
13:11out that for the outline test using
13:14Claude Opus 4.8 it cost 35 cents because
13:18this is a big prompt and
13:20with Fable it cost a dollar and 83 cents
13:22which is more than double and this is is
13:25incident where I saw I could tell it was
13:27thinking more.
13:28Uh you could you could actually see the
13:30reasoning it was it was thinking for
13:32quite a long time before it even started
13:33writing. Uh so, just because it says
13:37it's double doesn't necessarily mean uh
13:39that it won't be more, okay? All right,
13:41so moving on to the to the pros. So,
13:43this is a basic pros prompt where I just
13:45say, "Write out the 500 words based on
13:48these beats." And I give it nothing more
13:51than that because I like to see how how
13:53does it do out of the box with no major
13:57manipulation. And this is what Opus 4.8
14:00gave me. And I usually like to go and
14:03read a particular part um
14:06right around here. He should not have
14:09looked back. He knew it even as he did
14:10it, the way a child knows not to touch
14:12the stove and touches it all the same,
14:13helpless before the wanting to be sure.
14:16Helpless before the wanting to be sure.
14:18Okay, so that's a phrase that doesn't
14:19really make that much sense to me. They
14:21were there, scattered across the horizon
14:23where the last light pooled. The figures
14:25moved, and it was the moving that turned
14:27his stomach, the way it always did. The
14:29wrongness of it lodging behind his eyes.
14:32They did not run as men run. They
14:33jerked, each step a separate decision. A
14:36marionette hauled forward by some hand
14:38that did not understand how legs were
14:39meant to work. And yet, for all that
14:41broken motion, they came on, and they
14:42came on fast. He counted without meaning
14:44to, a dozen, two dozen, more resolving
14:46out of the gray as he watched. Their
14:48skin the precise dead color of the
14:49wasteland behind him, so that they
14:51seemed less like things crossing land
14:52than the land itself rising up to walk.
14:54Too close. They were far too close. All
14:56right, so that's that moment. It's uh
14:59it's a
15:00not too bad. Not actually as many
15:02AI-isms as I thought there would be.
15:04It's a little melodramatic, um which is
15:06typical of Claude in general, and is
15:09also typical of this particular scene,
15:10and this is fantasy also. Uh AI tends to
15:13get more cheesy with fantasy, which is
15:14why I test using fantasy prompts. Um
15:17it's usually better for contemporary
15:19stuff like romance and whatnot.
15:21Uh but let's look at Fable 5, okay?
15:23Let's find that same moment. They came
15:25scattered across the rocky plain like
15:27seeds thrown by a careless hand. Two
15:29dozen at least, perhaps more. Their gray
15:30skin so near the color of the wasteland
15:32that the eye slid off them until they
15:33moved. And they moved wrongly. This That
15:36was the thing the survivors tales never
15:38quite captured. The thing Halder
15:39understood now with the clarity he would
15:41have given anything to unlearn. The
15:43hungry hungry did not work walk. They
15:45jerked. Okay, so my first impression
15:47here is that was a kind of a convoluted
15:49way to say that. But let's keep going.
15:52Each step was a spasm, a marionette
15:53stutter, limbs snapping forward as
15:55though pulled by strings held in some
15:57vast and hateful fist. They never tired,
15:59they never slowed, they simply continued
16:01twitching across the miles patient as
16:02the grave that they all refused to stay
16:04in. They were close, far too close.
16:06And um
16:08So, my first impressions just from this
16:10little snippet is that it's not much
16:12better. Um, it it doesn't feel like a
16:15huge leap forward. Now, with writing
16:17there is a lot more nuance to it and I
16:18would need to spend more time with this
16:20on on a larger level because it could be
16:22that maybe the structure of the scene
16:25was way better. And but the prose for
16:27that particular paragraph was not as
16:28good. Um, but um additionally, one of
16:31the things that I don't do with the
16:32basic prose prompt is I don't give it
16:34any style info, I don't give it any of
16:36that just to see how it does out of the
16:37gate. Um, some models are way better at
16:40adhering to a style guide than others
16:44and to style examples.
16:46Uh, for instance, I found that the
16:48Claude models are generally bad at this.
16:50Uh, even though they're better for prose
16:51overall, they're they're not so good at
16:54adhering to style guides. Whereas the
16:56Gemini models are worse overall but
16:58better at adhering to style guides.
17:00Um, so that's why this next prompt, the
17:03advanced prose prompt, I actually give
17:04it a lot more of that information. The
17:06cost for this one was 5 cents for Opus
17:09and 10 cents for um Fable, um
17:12which is a little bit more than for the
17:14basic prose prompt. Let's find that same
17:16moment. Uh, it's a little bit shorter
17:18here. Actually, no, this is about right.
17:20They came down out of the gloom in a
17:21loose scattered line spilling over the
17:23rises and the gullies behind him. He
17:25counted before he could stop himself, a
17:26habit he hated. 10, 15, more than that,
17:29pulling themselves over the rocks with
17:30that broken jerking gait, limbs snapping
17:32forward and catching as though each step
17:34were an argument they kept losing. Um
17:36see, I'm not sure about that phrase.
17:38Um just kind of seems like a made-up
17:41metaphor.
17:42Their gray skin matched the stone so
17:44well that they seemed to bleed up out of
17:45the ground itself, the wasteland
17:46standing and walking, two dozen at
17:48least. Um yeah, so that's that's not too
17:50bad. Um I have noticed with both of the
17:53samples I've read here from Opus that
17:54there are a couple of phrases that you
17:56read that and you're just like, wait,
17:57what does that mean? Um like it doesn't
17:59always make sense. It sometimes it just
18:01sounds like it's supposed to be making
18:03sense. Haven't found that with Fable
18:04yet. Uh let's look at the same moment
18:07here um from Fable. They dotted the
18:10ridgeline behind him, scattered figures
18:11against the bruised sky. They moved the
18:13way they always moved, that broken
18:14stutter step, limbs jerking like puppets
18:16worked by a drunk hand. One would freeze
18:18mid-stride,
18:20head cocked, then lurch forward three
18:21steps and freeze again. Their gray skin
18:23matched the rock so well that he counted
18:24them only by their motion. 10, 15, more
18:27than two dozen spread wide across the
18:29wasteland like a net drawing closed.
18:31"No," the word cracked out of him. "No,
18:32no, no." He knew the distance, knew it
18:34the way every citizen of Arlden knew it,
18:36drilled into them from childhood. Uh I
18:38actually think this is better. Uh this
18:40sounds more natural to me. It feels a
18:41little bit less bloated. There's still
18:43some like AI metaphors
18:46um cuz AI uses metaphors a lot. Um but
18:49this is actually not too bad. I'd say
18:52it's better than what Opus gave me. All
18:54right. Next, we move on to another prose
18:56prompt, but this one has a lot of
18:58dialogue in it. And once again, this is
18:59a vanilla prompt, so I'm not giving it
19:01much instruction. I'm just seeing how
19:03well it does out of the box.
19:05So, in this moment, let's just uh
19:07there's a moment here that I usually
19:08like to read. "You let him die." His His
19:10carried the length of the hall past the
19:11cold braziers and the uh banners hanging
19:15limp in the windless air. Out there at
19:16the gate, a man screaming for our help
19:18for help, and your guards held me back
19:20while the hungry tore him apart. His
19:21father sat on the high seat unmoved, his
19:23face carved from the same gray granite
19:25as the mountain that sheltered them.
19:27King Storr of Varlden watched his son
19:28across the floor with the patience of a
19:30man who had heard storm breaks against
19:32these walls a thousand times and had
19:33outlasted everyone. "He came at full
19:35dark," Storr said.
19:37That's interesting. I haven't gotten
19:38that one before. An explanation. Okay.
19:41The hungry were already at his heels. To
19:43open the gate would have been to open
19:44the city. He was 20 paces out, 20. We
19:46had bowmen on the walls and you let them
19:48stand there with their arrows knocked
19:49and do nothing. So, it actually added
19:51something to the to to explain what had
19:54just happened, which was not in the
19:56instructions. I like that.
19:58Overall, this is fine. It's about what I
19:59expect from Opus. Let's go to the same
20:03moment
20:04>> [snorts]
20:05>> with Fable. "There was a man at the
20:06gates," Lathen's voice cracked against
20:08the vaulted ceiling. Not sure I like
20:10that dialogue there. Cracked against the
20:12vaulted ceiling feels like a very AI ism
20:14there. "A survivor, he made it across
20:16the wastes across the wastes, father,
20:19and we watched him die. We stood on our
20:21walls with our torches and our spears
20:22and we watched." I like how it's
20:24emphasizing things here. He stopped at
20:26the base of the dais, chest heaving.
20:28"That's the second one this month. The
20:29second soul we fed to the hungry while
20:31we hid behind our gates." Storr's eyes,
20:33pale as glacier ice, did not so much as
20:36flick flicker. "The law's the law, boy."
20:38"You make the law." Lathen flung his
20:39arms toward the high windows, toward the
20:41darkness beyond them. "The words came
20:42from your mouth, the seal from your
20:44hand. Don't speak to me as though it
20:45were carved into the mountain by the
20:46gods themselves. You could change it
20:48tomorrow. You could change it tonight."
20:50All right, so this is also good. Again,
20:52I don't see a huge leap. If anything,
20:55they're about the same
20:57or I would probably give a little bit of
20:59an edge to Fable. I do feel like it
21:01comes out a little bit more naturally in
21:03places, but it still has the occasional
21:07like cracked against the bowling
21:08vaulted ceiling moments, right? Like
21:11where I'm just I'm not into that.
21:15If this was like flawless prose, then
21:17I'd be like, yeah, this justifies the
21:19price. But given that this is double
21:20what Opus cost, which Opus is already
21:23expensive.
21:25Yeah, I don't know if I particularly
21:27care for the prose given that it's
21:30double the cost. It's certainly not bad
21:32though. And again, I would have to
21:34continue working with it for a while to
21:36make sure that it's not just the prose
21:39that it's doing better or worse. But
21:42sometimes you can get like those phrases
21:43that don't make sense that I've already
21:44identified a few with Opus or you can
21:47just get like things about the
21:49characters or about the flow of the
21:50scene that don't make sense that having
21:53a smarter thinking model can help
21:54mitigate. But regardless, let's move on
21:57to the next one, which is the editing
21:58prose template where I give it an older
22:00version of the same scene that was
22:02written by good old Claude 2 from way
22:04back in the day now and ask it to just
22:06improve the prose. And so here's what
22:10Opus gave me. We'll go to that same
22:12moment. What is there to discuss? A man
22:13died out there. Laythan stopped at the
22:15foot of the dais dais breathing hard.
22:17You watched it happen. He begged us for
22:18shelter, for mercy, and we left him to
22:20be torn apart. The law held. Storr's nod
22:22was slow, almost gentle. The gates
22:24stayed shut as they must. You could have
22:25opened them. Laythan's voice cracked
22:27across the chamber. You were the king.
22:28One word from you and that man would be
22:29alive. All right, so let's go to Fable.
22:32Speak then. A man died out there.
22:35Laythan stopped at the foot of the dais
22:36close enough to see his father's
22:37knuckles resting easy on the armrests.
22:39Easy, as if nothing had happened at all.
22:41You heard him. We all heard him. He
22:43clawed at the at the gates until his
22:45fingers bled begging for shelter and we
22:47stood on the walls and listened to the
22:48hungry take him apart. The law was
22:50upheld. Storr's voice was flat as a
22:51frozen lake. The gates stayed shut. That
22:53is the end of it. You are the king. The
22:55law bends to you, not the other way. One
22:56word from you and that man would be
22:57alive. And how many would be dead in
22:59this place?
23:00Uh
23:01this is better. Um when it comes to like
23:04I've read this a bunch of times. I know
23:06what the original text sounded like.
23:09This is a far better improvement on the
23:11original text than what Opus gave. Opus
23:13actually did a decent job, too, but I
23:15can say unequivocally this is better.
23:17Fable is better at improving old text.
23:20Um so, that's something to keep in mind,
23:22I think. Um all right, ad headlines.
23:24This is uh switching now to more
23:27marketing type of writing. Um I want to
23:29see like could we generate some hooks
23:32for Facebook ads or something like that?
23:34Um here are some of those. This is from
23:36Opus. He was the king's son and they
23:38threw him to the monsters. The reason
23:39will end a dynasty. They told him the
23:41mountain kept the world safe. The truth
23:43is the mountain has been lying. All
23:45right, these are all fine, but they're
23:47pretty consistent to what I get from
23:48other models. I don't see anything
23:50that's like a huge hook. And they're
23:52also pretty long, a lot of them. Like
23:54three races rule the wastes. One eats
23:56your flesh, one eats your will, one eats
23:57your soul. You have to walk through all
23:59of them. Like that one's not bad, but
24:01it's really long. Um
24:03and ideally these kind of hooks you want
24:05them to be short. So, let's look at
24:06Fable. Uh just glancing at them, these
24:09are also kind of long. Uh maybe not as
24:11as long as that last one, but uh let's
24:13read through a couple of these. He was
24:14the king's son that his own family threw
24:16him to the monsters outside. Everything
24:18Lithan learned inside the mountain was a
24:19lie. The truth is so much worse. The
24:21prince was cast out to die. He survived.
24:23Now someone owes him the kingdom. Um I
24:25don't think these are particularly
24:27better or worse. They're about the same.
24:30Okay?
24:31Um and then moving on to email
24:32newsletter prompts. Uh we have here um
24:37this is from Opus. This is to like say
24:39this is a book that I was actually
24:40releasing and I wanted to create a email
24:43to announce it to my email list. So,
24:46it's say a thrill to uh finally finally
24:48share this with you. My new epic fantasy
24:49novel The Crown of Hunger releases on
24:51December 21st.
24:53Um just looking through this. Um
24:55actually this is not bad. Yeah, this is
24:58pretty good and it's pretty succinct,
25:00which is exactly how I like it. For
25:03Fable, let's see. This is a book I
25:05poured my heart into. See, um, that's a
25:08little melodramatic, little bit more
25:09than what Opus was doing.
25:12I actually think Opus did better for the
25:14email newsletter. Although I like this
25:16PS here. Know a fantasy lover who would
25:18enjoy this? Forward them this email. The
25:20hungry are always looking for fresh
25:21readers. Um, which if I were actually
25:23writing this book and kind of leaning
25:25into the lore of it, that might be
25:27something I'd say. So, not too bad
25:30there. But overall, I'd say there's like
25:32there's like a long paragraph here. I
25:34don't know if I like that. Um, so that's
25:37just something to keep in mind there.
25:38Okay, last but not least is the SEO
25:40article super prompt. Where I ask it to
25:42write an article about how to write a
25:43fantasy novel. Um, and I ask it to write
25:454,000 words. So usually the first thing
25:47I check is did it actually write the
25:494,000 words. By the way, this one cost
25:5128 cents for Opus and 62 cents for uh,
25:55Fable. Um, cuz this is a longer one. So,
25:59uh, I don't spend too much time actually
26:01reading through these um, because they
26:03all follow the same outline. AI is
26:05generally good at non-fiction, but I do
26:07look at the word count to see how well
26:09it listened to my orders. And for Opus
26:12it gave me 3,300 words when I asked for
26:144,000. For Fable, it gave me 3,900 words
26:19uh, when I asked for 4,000. So, uh,
26:22Fable actually got much closer to the
26:24word count um, that I was looking for.
26:26Um, so good on Fable. And just kind of
26:29glancing through this, we have like
26:31introduction, what is the fantasy novel?
26:34Um, it's got everything seems to be
26:36you've got like
26:38um,
26:38bullet points and things. This actually
26:40looks like a pretty decent novel. That's
26:42uh, fully covering the topic. Um, so
26:45another win for Fable when it comes to
26:47SEO writing. Um, and I have a feeling
26:50just because of how good this thing is
26:53at reasoning and stuff like that that it
26:54would probably also be better at
26:57catching hallucinations and stuff like
26:59that before they happen. So, this would
27:01probably be a good one for non-fiction
27:02writing for sure. And that pretty much
27:04covers it for Claude Fable and no Tess.
27:07So, my bottom line for this is I think
27:10it is a big leap forward in some areas.
27:13I can definitely tell it's better at
27:14some things, particularly anything that
27:16involves any sort of reasoning like the
27:18outlining and the brainstorming process.
27:20I am on the fence so far about whether
27:22it's any improvement over Opus when it
27:24comes to prose. I think it is a little
27:26bit better, but I need to test more to
27:28be absolutely sure of that. And if
27:30you're one that really needs to pinch
27:32pennies and
27:34and save your costs, then I would
27:36probably continue using Opus instead as
27:39the primary workhorse, which is what I
27:41use in the tool that I've been
27:43developing. The default model for that
27:46is Opus. I would probably not change
27:48that to Fable just because of the cost
27:50involved. Although, if I experiment with
27:53it more and find that it is really good
27:55and saving me a lot of headache in some
27:57cases, then I might actually choose to
28:01to use it despite the cost because for
28:03me, you know, time time is money, right?
28:06And I value my time extensively. And if
28:09I could find one that would save me some
28:11time by writing a little bit more human
28:12sounding and natural, then I would
28:14absolutely go for something like that.
28:16Even even though I keep talking about
28:18cost here, I'm really talking about if
28:19you're accessing it through an API. For
28:22now in the near future, it is available
28:25at no extra cost to people who are on
28:28the Claude Pro subscriptions.
28:31But as I understand it, that is a
28:33temporary thing and won't be there
28:34forever. So, if you have a subscription
28:35to Claude, go use it. Now is the time to
28:37test out Fable to see if you like it.
28:39And
28:40and if not, continue to use
28:43Opus or one of the others. So, those are
28:45my thoughts and I will see you in the
28:46next video.