|
Heine vs Claude - The preview battle
|
| Heine |
Posted on 21-08-2026 20:42
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
Starting with a little teaser, the front page might lie a bit
Edited by Heine on 21-08-2026 20:49
|
| |
|
|
| Heine |
Posted on 21-08-2026 20:43
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
If you want to download the CT newspaper
Edited by Heine on 22-08-2026 16:55
|
| |
|
|
| Ad Bot |
Posted on 23-08-2026 23:56
|
Bot Agent
Posts: Countless
Joined: 23.11.09
|
|
| IP: None |
|
|
| Heine |
Posted on 21-08-2026 20:43
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
PDF of the entire PCT preview
Edited by Heine on 23-08-2026 13:10
|
| |
|
|
| Heine |
Posted on 21-08-2026 20:43
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
reserved
|
| |
|
|
| SotD |
Posted on 21-08-2026 21:18
|
World Champion
Posts: 13060
Joined: 29-11-2006
PCM$: 2980.00
|
Pretty cool and fun concept. 
|
| |
|
|
| knockout |
Posted on 21-08-2026 21:26
|
Grand Tour Champion
Posts: 8702
Joined: 21-12-2010
PCM$: 500.00
|
So first you are exploiting AI for content and now your own kids? 
The more previews the better but in the end your personal take is still the one im most interested in 
Just to be sure: The expected points ranking is the AI generated formula from last season and the claude ranking is a different AI with a different data set?
A Big Thank You To All MG Reporters! |
| |
|
|
| Heine |
Posted on 21-08-2026 21:35
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
knockout wrote:
So first you are exploiting AI for content and now your own kids?
The more previews the better but in the end your personal take is still the one im most interested in
Just to be sure: The expected points ranking is the AI generated formula from last season and the claude ranking is a different AI with a different data set?
2 models yeah.
1 made before the season that is just based on expected points made by a formula from last season. Ive had this formula available during transfers but not really used it. I can give a spoiler that it dosen't really like my own team 
The other is a formula that takes into consideration how different riders scored last season but put quite a lot more emphasis on about 200-250 test races I have uploaded. Considering the reason for these test races has been to test riders I was interested in and considered it will be "based" upon PCT results and PCT fields, so there are some considerable weaknesses both for PT and CT (though there are some C2 and quite a lot of PTHC data in it). Startlists has mostly been last years (though with updated stats) with some changes
The main part is as I wrote me squinting in the database and totally objectively consider different riders and teams. I've also written on it a few teams at a time since the db got uploaded, so riders quite a like can be quite differently ranked just based on when I wrote it (thought the ranking was made before I started writing)
I really hope my own opinions beat pure formulas, and even more important I hope it beats a random ranking from my kids (which I have to get them to do....)
|
| |
|
|
| kandesbunzler26 |
Posted on 22-08-2026 12:18
|
Small Tour Specialist
Posts: 2543
Joined: 07-07-2013
PCM$: 400.00
|
I really look forward to you getting beaten by your kids!
|
| |
|
|
| TheManxMissile |
Posted on 22-08-2026 16:20
|
Tour de France Champion
Posts: 18059
Joined: 12-05-2012
PCM$: 0.00
|
*two mid-sized towns were demolished to create the data centres to generate this preview
Better make it worthwhile for us to read! (I know it will be) I'm putting my money on the kids.
|
| |
|
|
| Heine |
Posted on 22-08-2026 16:50
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
TheManxMissile wrote:
*two mid-sized towns were demolished to create the data centres to generate this preview
Better make it worthwhile for us to read! (I know it will be) I'm putting my money on the kids.
At least I just made a newpaper generation program rather than making every page itself (after the first ones...)
|
| |
|
|
| Heine |
Posted on 22-08-2026 16:57
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
I considered doing CT a few teams at a time, but just dropped them all Imageheavy, so if you want you can just download the newspaper here
|
| |
|
|
| Dane Bramage |
Posted on 22-08-2026 18:43
|
Stagiare
Posts: 243
Joined: 16-06-2025
PCM$: 200.00
|
Heine's kids > everyone....
|
| |
|
|
| quadsas |
Posted on 22-08-2026 19:32
|
Classics Specialist
Posts: 2961
Joined: 18-01-2013
PCM$: 300.00
|
Lutsenko the top dawg you say..
|
| |
|
|
| TheManxMissile |
Posted on 22-08-2026 20:29
|
Tour de France Champion
Posts: 18059
Joined: 12-05-2012
PCM$: 0.00
|
OK, so the Kids are really just going on vibes. Claude is going on, something, idk but it's both logical and insane - which is a perfect summation of LLM AI. The actual Heine written parts are top notch, nice thoughts on all the teams very easy to read and understand.
Definitely a general CT hierarchy shared across previews so far, but as always proof the battle for 5th will be intense, and some surprise is going to be the difference between going up and not.
|
| |
|
|
| seancoll |
Posted on 22-08-2026 22:19
|
Protected Rider
Posts: 1055
Joined: 20-12-2022
PCM$: 500.00
|
Beyond the obligatory thank you's for some high quality content (they're very nice of course), I am simply going to relish in being the only team with Top 5s in all 4 rankings! A top 3 with the kids is a big win for our management, who typically struggles to bring as much whimsy as we try to with the Turtles!
On a statistical note, what were the basis for 1,2,3 stars?
|
| |
|
|
| Heine |
Posted on 22-08-2026 22:27
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
No statistics at all, pure subjective opinion right there and then with pure human randomness
|
| |
|
|
| knockout |
Posted on 22-08-2026 22:40
|
Grand Tour Champion
Posts: 8702
Joined: 21-12-2010
PCM$: 500.00
|
Lovely read!
How were the startlists created for the test runs? The Claude parts felt quite weird to me, particularly when they listed a rider not being used enough as negatives. Also the terrain scores in the Claude part (stuff like "X adds 0,73 GC and positive on two other terrains") feel very unexplained.
On to the positive part: I think the stars overview is really intuitive and very good and i like the presentation of the roster. The Heine analysis is also really nice and a great read! And the ranking prediction itself also looks pretty good - seems like there is a bit of a consensus despite the general opinion that the division is super close..
A Big Thank You To All MG Reporters! |
| |
|
|
| jackgibbs |
Posted on 22-08-2026 23:49
|
Under 23
Posts: 94
Joined: 25-07-2021
PCM$: 300.00
|
Obviously a bit more positive on my own position then what you have here, and definitely see more disagreement then what I have with seancoll.
Will be very interesting to see how the accuracy plays out. Clearly some agreement on the top and the bottom of the rankings, but the middle is just so tight and could break so many different ways… |
| |
|
|
| cunego59 |
Posted on 23-08-2026 00:02
|
Team Manager
Posts: 6707
Joined: 14-09-2008
PCM$: 1290.00
|
Some oddities in there with Claude calling Galta the most raced rider (in the tests?) while asserting he has never scored - feels like either simply an error or there were no mountain races in the test data. And then Gregory Daniel has the poorest return of any rider despite apparently one more Top 10 than Galta in fewer race days (and presumably less salary). There will be mistakes in any LLM analysis, but I would be curious about the underlying dataset with these ones.
That said, still a cool presentation, love the effort you put in, and especially your own analysis
|
| |
|
|
| Heine |
Posted on 23-08-2026 07:03
|
Grand Tour Specialist
Posts: 5075
Joined: 08-04-2007
PCM$: 200.00
|
There has ben some different startlists used, I made one «mayor» startlist early that was 20 teams with riders I considered and riders I had and used that for a lot of races (and several times trough all my hc and pthc races). That means riders like Galta has ridden countless flat classics etc. So he is raced alot 
Ive also had a startlist with all too pct riders before transfers and many fas
The lastly I’ve raced quite a lot with last years startlists for all my races with my current team
That gives a bias in the dataset that will effect both ct and pt specially. I told the model when it made it this and it shouldve made adjustments for it, but I didnt interfer too much and refused to edit any text even when I felt it was very strange
I will post a download link to the entire claude analysis later as it also includes some data I dont copy in
|
| |
|
