X Feed Intel beta

individual tinkerer enterprises
Weekly Prep Posts Topics History
670 relevant 263 topics 1841 posts $1.088 cost fetched 02-23T21:39
670
Relevant
263
Topics
1841
Total Posts
$1.088
Cost This Week
$1.088
Total Cost
2026-02-23T21:39
Last Fetch
← Back to Topics
Frontier Models

Synthetic data generation and pipeline scaling for AI training

Synthetic video data generation, scaling internal data pipelines, and cost-effective open-source training approaches

6 posts · First seen 2026-02-23 · Last activity 2026-02-23
TimeAuthorPost
2026-02-23T20:54 @srchvrs @yoavgo This is too little for pre-training, but pre-training nowadays is probably not a bottleneck. For post-training 16M training samples can meaningfully improve performance on a lot of tasks. ↩ reply parent
2026-02-23T19:37 @tanvi_ratna The White House is launching a new Tech Corps & a bundle of novel initiatives in a new AI strategy. I sat down with @mkratsios47 Assistant to POTUS & Director of the White House Office of Science & Tech Policy, to break it down for @FoxNews https://t.co/VOZwIulwh5
2026-02-23T19:13 @1a3orn @kalomaze > training the the mimesis machine on the entirety of human intellectual output: Haha fuck yeah!!! Yes!! > others train the mimesis machine on a few billion tokens from me: Well this fucking sucks. What the fuck. ↩ reply parent
2026-02-23T18:52 @Dorialexander @zephyr_z9 Simultaneously a lot and very few. If you’re serious about synth pipeline you want to mostly scale up internal capacity. Big labs input is for very initial seeding. ↩ reply parent
2026-02-23T18:29 @flowersslop white pill is that now that video models are this good, it is relatively cheap for a company to train an oss model by pretraining on effectively infinite high quality synthetic video thats much more diverse than any pre 2025 video dataset and then just finetune on real video https://t.co/dZukh3FnqN
2026-02-23T17:28 @gizakdag I've been exploring Seedance 2.0, already obsessed. https://t.co/l02TmXDjdf
@srchvrs 2026-02-23T20:54
@yoavgo This is too little for pre-training, but pre-training nowadays is probably not a bottleneck. For post-training 16M training samples can meaningfully improve performance on a lot of tasks.
↩ reply parent
@tanvi_ratna 2026-02-23T19:37
The White House is launching a new Tech Corps & a bundle of novel initiatives in a new AI strategy. I sat down with @mkratsios47 Assistant to POTUS & Director of the White House Office of Science & Tech Policy, to break it down for @FoxNews https://t.co/VOZwIulwh5
@1a3orn 2026-02-23T19:13
@kalomaze > training the the mimesis machine on the entirety of human intellectual output: Haha fuck yeah!!! Yes!! > others train the mimesis machine on a few billion tokens from me: Well this fucking sucks. What the fuck.
↩ reply parent
@Dorialexander 2026-02-23T18:52
@zephyr_z9 Simultaneously a lot and very few. If you’re serious about synth pipeline you want to mostly scale up internal capacity. Big labs input is for very initial seeding.
↩ reply parent
@flowersslop 2026-02-23T18:29
white pill is that now that video models are this good, it is relatively cheap for a company to train an oss model by pretraining on effectively infinite high quality synthetic video thats much more diverse than any pre 2025 video dataset and then just finetune on real video https://t.co/dZukh3FnqN
@gizakdag 2026-02-23T17:28
I've been exploring Seedance 2.0, already obsessed. https://t.co/l02TmXDjdf

Markdown Export

Loading...