Mani Mohan pfp

Mani Mohan

@mani

240 Following
356 Followers


Mani Mohan pfp
Mani Mohan
@mani
Interesting feature on Dia - custom skills. You can create your own and run it on the chat window when the website is up. I first tried it on LinkedIn to summarize all the AI slop - but it works on FC too.
0 reply
0 recast
0 reaction

Mani Mohan pfp
Mani Mohan
@mani
tried to use it for a few hours. My brain refuses to switch back to horizontal tabs. Easy switch if they had just put in the vertical tabs + maybe spaces.
1 reply
0 recast
1 reaction

Mani Mohan pfp
Mani Mohan
@mani
forgot to include: the above classification took ~3 seconds (for 100 casts).
0 reply
0 recast
2 reactions

Mani Mohan pfp
Mani Mohan
@mani
Gemini Diffusion is insane fast. It’s not very robust yet, but this shift to diffusion architecture is exciting - looking forward to API release. I wonder if they will pivot to the diffusion models for the v3 flash. fun test: get your feed from neynar and play with it.
1 reply
0 recast
1 reaction

Mani Mohan pfp
Mani Mohan
@mani
horizontal tabs 😦 Does it still support vertical tabs / spaces?
0 reply
0 recast
0 reaction

Mani Mohan pfp
Mani Mohan
@mani
Throwback to Microsoft Longhorn launch
0 reply
0 recast
1 reaction

Mani Mohan pfp
Mani Mohan
@mani
Custom app icons for Pro subs.
0 reply
0 recast
0 reaction

Mani Mohan pfp
Mani Mohan
@mani
Built something useful this week: a morning personalized newsletter for both me and the wife. - provides weather, individual portfolio snapshot - latest news in our individual company/sector - day care lunch menu detection (google drive integration - detects PDF file to parse based on date and finds the lunch menu for that day from the PDF) - ai summary of the day based on calendar (ours + daycare calendar) - whoop data integration (sleep/recovery/strain from previous day) - provides a window for 30 mins workout (based on calendar + toddler duties and all the above data) Runs as a cron job on a local machine, using gemini for all the AI part. the goal was to auto-detect my daughter's lunch menu and send it as email. Just built on top of it. Happy progress.
3 replies
1 recast
10 reactions

Mani Mohan pfp
Mani Mohan
@mani
Yessir!
0 reply
0 recast
1 reaction

Mani Mohan pfp
Mani Mohan
@mani
This is for last week's rewards. I don't have the data for the past weeks though. I just copy pasted usernames from the rewards app and ran the analysis. https://farcaster.xyz/mani/0xb070c7fc
1 reply
0 recast
1 reaction

Mani Mohan pfp
Mani Mohan
@mani
Past week caster rewards insights: Avg Neynar score = 0.87, median = 0.95. Full data and plots here - feel free to play with it. https://observablehq.com/d/8295f460e9adebc1
1 reply
1 recast
3 reactions

Mani Mohan pfp
Mani Mohan
@mani
I think Neynar team is actively working on the scores - I saw it change for some users last few days. I have been running cron jobs everyday for all pro users to track how the scores are changing (it is changing). Problem with the original cast is - pro subscription != high signal at least in its current state. A combination of factors like pro + neynar_score + maybe median score of followers(?) might give a better picture. I don't have a proper answer.
1 reply
0 recast
2 reactions

Mani Mohan pfp
Mani Mohan
@mani
Update on this: Tried imagen4 using whisk (labs) and the results were drastically better. Well formatted text with proper padding, better color choices, much faster for iterating. It did create duplicate texts and missed some items in the image, but iterating was easier with better speed.
0 reply
0 recast
0 reaction

Mani Mohan pfp
Mani Mohan
@mani
4o image generation consistently fails on text-heavy infographics: typos, poor padding, bizarre color choices despite exhaustive prompting. Even had GPT review image + regenerate prompts. Anyone tried Imagen 4 for this use case? Looking for alternatives that actually handle typography and layout properly.
0 reply
0 recast
1 reaction

Mani Mohan pfp
Mani Mohan
@mani
I have it on GitHub as a csv (https://raw.githubusercontent.com/manimohans/fc_pro_analytics/refs/heads/main/addresses_fc.csv), but some pointers: - this csv has around ~11.8K subscribers (first 48 hours, not 24h, use an older commit for 24h) - the number of followers for folks @ 48h will be different from 24h. A few 0s moved to 1-10s and so on. I think the neynar scores change weekly so thats probably ok - unless they update scores on a Thursday. Hope this helps.
0 reply
0 recast
3 reactions

Mani Mohan pfp
Mani Mohan
@mani
Closing the loop. This is the last 1500 transactions (did NOT slice it from the 10001th acct who signed up). FIDs > 1.09M (Newer accounts): 47 (3.1%) FIDs ≤ 1.09M (Older accounts): 1,453 (96.9%) 0 followers: 30 (2.0%) 1-10 followers: 128 (8.5%) 100-1K followers: 551 (36.7%) 1K-10K followers: 400 (26.7%) 10K+ followers: 41 (2.7%) Average Neynar Score roughly about the same. Github repos updated - for anyone interested.
1 reply
1 recast
14 reactions

Mani Mohan pfp
Mani Mohan
@mani
I do and pay for it yearly - for past 3-4 years. Great product, very useful for me.
0 reply
0 recast
1 reaction

Mani Mohan pfp
Mani Mohan
@mani
Some are surely new. You can verify this with google sheets that @dwr.eth shared too. Current max fid is ~1.093mil and there are about ~1200 pro fids > 1.090mil.
1 reply
1 recast
2 reactions

Mani Mohan pfp
Mani Mohan
@mani
Yes. I have the csv in the github repos - I verified manually random usernames (note neynar followers # is usually > what you see on fc here), feel free to verify further.
1 reply
0 recast
2 reactions

Mani Mohan pfp
Mani Mohan
@mani
https://farcaster.xyz/mani/0xde57512a yes, i was seeing the follower count and it matches ~1200
0 reply
0 recast
1 reaction