Njal

@cryptonjal

442 Following

310 Followers

Njal pfp

My tipping in perspective: 🪀Until around early Aug. '25 I will tip $YOYO, every week, mostly on Sat./ Sun. to everyone on my list with /rewardr (add @dosir's mini app to receive notifications!) CA: 0xE31876C6A62A813f57b815d8d2d0F5c8Aa06f49B (Base layer) 💰 At that time, I expect that we are realizing that we are in a bear market. So I'm planning to tip $USDC at that time, and I'm saving my Farcaster rewards for that. If you're curious, the wallet address with my tipping treasury is: 0x9974e8d237e5fc71C051800d66b85C5B2B238f6A

3 replies

1 recast

5 reactions

Njal pfp

Sounds a little bit France: le farclet(s)

0 reply

0 recast

0 reaction

mvr 🐹 pfp

The new spam labels dataset (which was still a draft) was just removed from github Something cooking? cc @pichi @cryptonjal @compez.eth

3 replies

2 recasts

9 reactions

Njal pfp

That's a good question, and good for speculation. My guess is that former label 1s should have resulted in more 0s. Viewed from a distance: 96.7% from label 1 to 2 isn't credible. We can look for big differences next time.

0 reply

0 recast

0 reaction

Njal pfp

Pro Farcaster tip 😂 I love creativity https://warpcast.com/maretus/0x620d8830

2 replies

0 recast

5 reactions

Njal pfp

Interesting: - 1/2 no label, 1/4 label 0, and 1/4 label 2. Imo the unitial ratios where and are to neat https://warpcast.com/cryptonjal/0x4338e36e - Total FID label changes are in favour of label 2, but this change is bigger in active FIDs then in all FIDs in the dataset. 39.4% (was 17.2%) of all active FIDs are label 2, and only 25.8% of the dataset has label 2. Could be the same ratio active/total label 2 FIDs in the old system, but I'm not behind my computer atm (if I remember well it was 12-14%). If that ratio apply it would be an incentive to get label 2 numbers as high as reasonable accountable because the most important figure, active FIDs, changes 2x more in favour of label 2.

1 reply

1 recast

1 reaction

Njal pfp

Framedl 2025-05-23 3/6*

0 reply

0 recast

2 reactions

Njal pfp

Yes, there is a huge tension there, because it is also interesting for spammers to create a new FID without a spam label. Incidentally, some (new) low-quality FIDs seem to show remarkably honest behavior lately instead of spammy, and shamelessly ask for tips in their own timeline without sounding spammy. They also innovate 😁.

0 reply

0 recast

1 reaction

Njal pfp

Separate testnet from the real nets with a switch option.

0 reply

0 recast

0 reaction

Njal pfp

They are totally Based 😁

0 reply

0 recast

0 reaction

Njal pfp

Check the new dataset figures here: https://dune.com/nhejyht/farcaster-dashboard https://dune.com/nhejyht/farcaster-monthly-dashboard

0 reply

0 recast

1 reaction

Njal pfp

So, is it better? Not necessary. If you're building a system where all outcomes pass a validation test, it doesn't mean that the judgement of the system is better. Besides that, there could be management motives to keep the number of spam label 2s as high as possible: - investors would value the company/project higher (happy investors are important!) - more happy casters happy with a label 2 of course, there needs to be a balance with people getting annoyed with label 2 casters who they consider as spam.

2 replies

0 recast

2 reactions

Njal pfp

Is the new spam labelling system then the old one? Let's first dive into the figures of active FIDS in the last 21 days: Label | old system | new system 0 | 87,945 | 82,379 1 | 22,508 | 0 2 | 26,002 | 59,695 no label | 14,912 | 9,293 Total | 151,367 | 147,897 I don't know where the difference in total active FIDs, between the two datasets came from, tbh. Interestingly: - with the new dataset there are 3,470 active FIDs missing(?) I dive into that when I have time. - the change was only in favour of spam label 2. - Disappearance of label 1 suggest that all outcomes of the system are (considered) valid. - Less no labels suggest that they need less data, so the outcome of the LLM seems to have stronger signals.

3 replies

1 recast

8 reactions

Pichi pfp

HUGE changes to Spam. Spam is now binary. The algorithm is now set to likely to be spammy or unlikely to be spammy. The weird limbo level 1 level is gone and good riddance! Half of accounts on the network don’t have a label (probably inactive). Resources below on helping friends who are mislabeled.

6 replies

5 recasts

41 reactions

mvr 🐹 pfp

Add the Tipopolis mini app if you'd like to get notifications like these so you won't forget to extend your lock Tipopolis mini app by @mvr

7 replies

10 recasts

59 reactions

Njal pfp

Interesting that a lot of 1s went to 2. I assume that the old LLM outcome was used to make a correlation calculation. In such test you can have positive or negative correlation (spam / no spam), you need enough data (explains why they can't label everyone), and you need a significance test (that was label 1 I assume). It interest me, why they haven't a non significant group anymore. Can't believe all outcomes are significant. The things that I came up with is: - the new LLM polarise the outcome more - or maybe the run a multiple regression analysis now (maybe they already did, but more simplistic) to determine if a caster is a certain type of 'spam' or 'no spam'. In that way all non significant correlations are ruled out, redo the analysis with the significant correlations and if the mayority of outcome correlatates with non spam it's a 2 and visa versa. - or something else

0 reply

0 recast

0 reaction

Njal pfp

Almost 47% no label 👀 The more you know, the more you realise how less you know

0 reply

1 recast

2 reactions

Njal pfp

In a very specific and very small niche 😎

0 reply

0 recast

2 reactions

Njal pfp

1 reply

0 recast

0 reaction

Njal pfp

Strange. by me it isn't that small

1 reply

0 recast

1 reaction