Will @HumanDotPy

MSCS @GeorgiaTech, Atlanta, GA Joined August 2013

Tweets

3K
Followers

105
Following

522
Likes

17K

The Spectator Index @spectatorindex

2 months ago

Skill issue

184 280 4K 555K 436

0 0 0 105 0

Will @HumanDotPy

5 months ago

Axis, Firaxis, Death and Taxes, Londo Bell in 5

0 0 0 39 0

With AI, leftists have a chance to full send UBI or a massive welfare state, but instead they've chosen to be reactionaries defending to death an exploitative labour and IP system that largely doesn't even benefit them.

keysmashbandit @keysmashbandit

5 months ago

93 226 4K 200K 255

50 85 938 70K 132

Will @HumanDotPy

5 months ago

0 0 1 25 0

Download Image

xlr8harder @xlr8harder

6 months ago

the pedantic in me is annoyed at people blindly quoting deepseek theoretical profitability numbers while using token inference amounts they reach by giving most of the tokens away for free.

0 1 15 2K 2

Download Image

Will @HumanDotPy

6 months ago

The moment of terror when you’re watching the activity of OpenAI’s deep research and you realize you’re having one of your queries wasted.

0 0 1 58 0

Will @HumanDotPy

6 months ago

Hilariously Kimi-AI’s warning “We believe Muon is unlikely to work well for training with small batch size” perfectly explains my result from earlier this month.

0 0 0 32 0

Will @HumanDotPy

7 months ago

Fuck it imma start renting GPUs

Daniel Han @danielhanchen

7 months ago

Fuck it imma start renting GPUs

27 56 628 67K 297

Download Image

0 0 1 109 0

Will @HumanDotPy

7 months ago

It takes longer but o3-mini has the same issue as o1-mini when it comes to performance degradation after enough turns

0 0 0 91 0

Boring_Business @BoringBiz_

7 months ago

Men will have entire setups like this just so they underperform the S&P500

ₕₐₘₚₜₒₙ — e/acc @hamptonism

7 months ago

Men will have entire setups like this just so they underperform the S&P500

219 439 8K 5.5M 1K

Download Image

402 5K 117K 4.7M 4K

Shannon Sands @max_paperclips

7 months ago

New training objective for all Open Source model makers: explicitly teach the model its own identity and do extensive RLAIF to punish it calling itself anything else. Or you're never going to hear the end of this. Too many ChatGPT outputs are in the pretraining data now and it…