Idan Mann (@mann_idan) header
Idan Mann (@mann_idan) profile picture

building agents that do real work, and getting labs the data to make them better and cheaper. https://xitter.cf/t.co/zW8CUw79uF · https://xitter.cf/t.co/j0L5WgNgnV · https://xitter.cf/t.co/zTjvIp6i3G

Joined April 2021
Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)Photo by Idan Mann (@mann_idan)
Idan Mann retweeted
we’re buying large company data archives for AI training. priorities: - banks/lenders - healthcare/medical billing - consulting/legal - kernel optimization traces must have 50+ employees dm me if you have this type of data!
12
3
42
4,223
we’re buying large company data archives for AI training. priorities: - banks/lenders - healthcare/medical billing - consulting/legal - kernel optimization traces must have 50+ employees dm me if you have this type of data!
12
3
42
4,223
Idan Mann retweeted
we’re buying traces of engineers modifying weights like this plus related engineering tasks! comment a description of your data, and I’ll dm you if we’re interested.
Got fable to hand-write weights of a 100k param model to recognize real versus fake legal citations. Basically one-shotted a full NN that’s better than training it
16
7
5
122
29,778
we’re buying traces of engineers modifying weights like this plus related engineering tasks! comment a description of your data, and I’ll dm you if we’re interested.
Got fable to hand-write weights of a 100k param model to recognize real versus fake legal citations. Basically one-shotted a full NN that’s better than training it
16
7
5
122
29,778
Idan Mann retweeted
why are data brokers not taking halloween off? like bro I'm sorry but selling six figs of CAD to some lab can wait until Sunday
For those asking what data is most in demand. - CAD data + operational data (must have rights to it) - kernel optimization traces dm me if you have those! (or anything else interesting) feel free to ask questions below, so I don’t need to answer repeat questions in dm’s!
4
1
2
49
5,330
Idan Mann retweeted
For those asking what data is most in demand. - CAD data + operational data (must have rights to it) - kernel optimization traces dm me if you have those! (or anything else interesting) feel free to ask questions below, so I don’t need to answer repeat questions in dm’s!
10
3
1
71
9,428
For those asking what data is most in demand. - CAD data + operational data (must have rights to it) - kernel optimization traces dm me if you have those! (or anything else interesting) feel free to ask questions below, so I don’t need to answer repeat questions in dm’s!
10
3
1
71
9,428
I’m leaking my friends number, spam him. +1 (407) 234-1884
1
4
1,185
this is pretty cool.
Arlo can now keep you updated on your favorite podcasts. Tell it what you’re into. It’ll keep up with new episodes, and text you the parts you’ll care about, with the exact minutes to listen to.
2
1
5
805
who’s the best inference provider right now? looking at @runpod_io, @togethercompute and @FireworksAI_HQ. what’s been your experience with latency, reliability and cost in production? anyone else worth a look?
5
9
877
Idan Mann retweeted
we have $1M+ worth of data suppliers in our dm's. since I can't answer everyone drop questions you have about data brokering below. I'll answer as many as possible.
10
3
34
3,713
we have $1M+ worth of data suppliers in our dm's. since I can't answer everyone drop questions you have about data brokering below. I'll answer as many as possible.
10
3
34
3,713
Instinct is raising at a $30B valuation with a team of 14 people. $2.1 billion in valuation per employee lol what happens when the bubble bursts?
1
6
853
Idan Mann retweeted
labs i’ve talked to often don’t have enough data to work with after over 4k dm’s and tons of sellcodebase.com submissions we have more data than we can handle. if you’re a lab who moves fast and want wants high quality data. my dm’s are open (no data brokers please)
2
4
36
5,296
labs i’ve talked to often don’t have enough data to work with after over 4k dm’s and tons of sellcodebase.com submissions we have more data than we can handle. if you’re a lab who moves fast and want wants high quality data. my dm’s are open (no data brokers please)
2
4
36
5,296
Idan Mann retweeted
people who make tasks are super valuable rn good researchers are paid 5M-20M by Mercor/Micro1 to make rl envs/tasks and ensure data quality this sounds like a lot but they are still underpaid! tasks go for 1-5k+, and some researchers can easily make well over 30 per day. some are choosing to just make tasks on their own time, and revenue split. the problem is that they need high-quality raw data to work with, which is now readily available outside of Mercor/Micro1
selling data to labs is on its exponential curve. labs already trained on the whole internet. What they need now is high quality data, and the only way to make it worth anything is turning it into tasks agents can learn from. we’re looking for more people with experience turning raw data into tasks. 50/50 rev share. If you've built tasks for labs, send me a dm.
20
17
4
593
55,874
Idan Mann retweeted
selling data to labs is on its exponential curve. labs already trained on the whole internet. What they need now is high quality data, and the only way to make it worth anything is turning it into tasks agents can learn from. we’re looking for more people with experience turning raw data into tasks. 50/50 rev share. If you've built tasks for labs, send me a dm.
19
7
3
207
65,308
selling data to labs is on its exponential curve. labs already trained on the whole internet. What they need now is high quality data, and the only way to make it worth anything is turning it into tasks agents can learn from. we’re looking for more people with experience turning raw data into tasks. 50/50 rev share. If you've built tasks for labs, send me a dm.
19
7
3
207
65,308
Idan Mann retweeted
for doctors/lawyers/senior engineers/experts using codex/claude code/openclaw/hermes I profoundly believe this to be true. try selltraces.com and sellcodebase.com to license your local traces and codebases to labs
there's a million dollars trapped in your macbook you just have to sell data to frontier labs to get it
19
15
685
64,392