seth bloomberg @bloomberg_seth

Investment Partner @UnsupervisedCap | prev: @CrucibleLabs, research @MessariCrypto | Woo Pig 🐽 t.me/CollectionofTh… Joined September 2019

Tweets

1K
Followers

4K
Following

3K
Likes

4K

seth bloomberg @bloomberg_seth

2 days ago

We need more batteries.

2 0 3 456 0

Sami Kassab @Old_Samster

2 days ago

I actually just spoke with both entities, and they confirmed that their deals have been 20% or less, with one exception at 30%. The other thing to keep in mind, is that 20% stake in the Owner's emissions ends up being less than 3% of the fully diluted supply when you factor in…

JJ @JosephJacks_

2 days ago

68 66 442 68K 80

8 7 127 18K 14

templar @tplr_ai

a week ago

Starting in an hour! @Old_Samster and @bloomberg_seth from @UnsupervisedCap join @DistStateAndMe for our TGIF community chat Builder insights meet investment expertise - early crypto x AI pioneers discussing what they're seeing x.com/i/spaces/1LyxB…

13 7 41 2K 5

seth bloomberg @bloomberg_seth

a week ago

I bet I know who did the max/227 commits…

Doug Sillars 🌻 @dougsillars

a week ago

I bet I know who did the max/227 commits…

1 1 2 2K 0

3 0 10 739 0

seth bloomberg @bloomberg_seth

a week ago

Everything is RL environment

Prime Intellect @PrimeIntellect

a week ago

Everything is RL environment

117 394 3K 1.2M 2K

Download Video

1 0 11 638 1

seth bloomberg @bloomberg_seth

a week ago

Very few ways for open source AI companies to survive. Those who do will emphasize brand building and aesthetics as much (if not more) than they do innovative tech. Seems the former is required to endure to generate the latter.

2 0 6 364 0

Jason Weston @jaseweston

a week ago

🪜Introducing: StepWiser🦉 📝: arxiv.org/abs/2508.19229 - Reframes stepwise reward modeling as a reasoning task: outputs CoT + judgment. - Trained by RL using relative outcomes of rollouts. Results: (1) SOTA performance on ProcessBench! (2) Improves policy at train time. (3)…

11 98 487 83K 364

Download Image

clem 🤗 @ClementDelangue

a week ago

The "one model to rule them all" paradigm is officially dead, time to move on to "a model for each use-case"! As we've been saying, ultimately there will be as many AI models as code repositories, as shown by the 3,000 new models shared on HF every day!

41 33 249 17K 39

Download Image

seth bloomberg @bloomberg_seth

2 weeks ago

One of crypto’s core features has always been using tokens as a new mechanism for capital formation. The ICM thesis is a new wrapper around this old idea, and it marks a reflexive swing back towards crypto’s early roots. Great perspective from Sami here.

Sami Kassab @Old_Samster

2 weeks ago

16 50 200 39K 93

0 0 5 548 4

Sami Kassab @Old_Samster

2 weeks ago

Hyperliquid showed the ICM path works for a single app Bittensor is showing it works for an entire ecosystem

Sami Kassab @Old_Samster

2 weeks ago

Hyperliquid showed the ICM path works for a single app Bittensor is showing it works for an entire ecosystem

16 50 200 39K 93

16 13 116 17K 25

Sami Kassab @Old_Samster

2 weeks ago

x.com/i/article/1957…

16 50 200 39K 93

Tommy @yeahfortommy

3 weeks ago

x.com/i/article/1947…

6 6 51 9K 32

Sami Kassab @Old_Samster

3 weeks ago

Nobody is really taking the crypto-AI space seriously at the moment. And I get why. The large majority of projects that raised off the AI hype in 2024 are all irrelevant and never shipped anything worth using. The few that did have removed any crypto terminology from their…

19 16 154 13K 32

seth bloomberg @bloomberg_seth

3 weeks ago

If you allow yourself to believe in something, then you can come to understand that this is the big picture of what Bittensor is doing. Cryptoeconomic protocols with well-designed and aligned incentive mechanisms, coupled with open competition is the path forward.

const @const_reborn

3 weeks ago

4 18 142 13K 15

2 0 5 491 2

Jay @jayendra_ram

3 weeks ago

I usually don’t talk about RL envs on the tl out of respect for our customers but this take won’t age well. Making problems that provide signal for models to get better is pretty hard, and is only going to get harder every year as models improve. The notion that you can vibe…

Rohan Pandey @khoomeik

3 weeks ago

22 9 313 91K 168

7 6 218 40K 198

clem 🤗 @ClementDelangue

3 weeks ago

Fun to think about open-source models and their variants as families from an evolutionary biology standpoint and analyze "genetic similarity and mutation of traits over model families". These are the 2,500th, 250th, 50th and 25th largest families on @huggingface:

Hamidah Oderinwale @didaoh

3 weeks ago

8 33 221 54K 157

Download Image

9 19 107 32K 31

Download Image

seth bloomberg @bloomberg_seth

4 weeks ago

As this unfolds over the next year or so, very clear that these two areas will be in high demand: • RL environment architects/engineers • Custom eval developers

will brown @willccbb

4 weeks ago

As this unfolds over the next year or so, very clear that these two areas will be in high demand: • RL environment architects/engineers • Custom eval developers

114 114 2K 112K 345

2 0 12 1K 2

seth bloomberg @bloomberg_seth

4 weeks ago

The breadth/scale of Google's AI science and research is insane: • Gemini/Gemma models • Veo3 • Alpha series (fold, genome, evolve) • Genie 3 • TPUs Alpha series + Genie 3 are especially fascinating. My biggest question: how do they decide funding priorities?