@dummerspast39

anti ai lobby china shill big hater

Joined August 2024
  • Tweets4.7K
  • Following276
  • Followers2.8K
800Media
Fireworks AI was breached on October 6. OpenCode on October 17. Railway on October 21. We are accelerating!
3
2
19
1,115
By vote I will now render in 3:4 ratio
1
1
6
658
I would put the Kiblitz (k3.1 checkpoint?) at around gpt 6 Terra (not 6.1) level
Binkg
@dummerspast396h
Ui and colors is very gpt-6 pilled. Everybody switched from Claude to gpt in their distilling pipeline
2
1
9
880
After seeing 50 outputs
250
Happy last day before AGI!🚨😸
12
2
52
3,024
Ui and colors is very gpt-6 pilled. Everybody switched from Claude to gpt in their distilling pipeline
AiBattle
@AiBattle_7h
The Kibitz model claims to be a Kimi model. The model also currently appears at a very high rate in the Code Arena Asking models about their identity in the Code Arena is unreliable. GLM and other models also sometimes identify as Kimi models, so I’m not sure yet what the Kibitz model is
1
1
13
1,671
I have capture maxxed
2
372
Gemini 4 Carbon - SVG of Cola (I asked for svg yes) This is actually a great output when taking into account it was in the Arena. For like 5 minutes, shit harness. Even better than Opus
26
13
6
356
31,065
Kimi K3.1 might ALREADY be tested in the ARENA! 2 Models: Kalmia and Kibitz
4
2
30
6,819
Deepseek v4.1 flash 1017 - ana de armas
4
16
778
Horizon-6.2 - ana de armas svg
2
1
6
454
Gpt 6 Astra - ana de armas (adiposimas)
Binkg
@dummerspast398h
Gemini 4 Carbon - Ana de armas🤮🤮
1
10
691
ARC AGI is for the Chinese labs now.
ARC Prize
@arcprize8h
Claude Sonnet 5.5 from @AnthropicAI on ARC-AGI (Verified): - ARC-AGI-3: 72.4%, $8.2K (provider adapter harness), 24.3%, $7.4K (standard harness) - ARC-AGI-2: 83.9%, $0.55/task - ARC-AGI-1: 96.5%, $0.14/task Sonnet 5.5 is notable for its 72.4% score on ARC-AGI-3.
14
802
Gemini 4 Carbon - Ana de armas🤮🤮
8
1
1
24
1,764
Another one
1
4
262
Gemini 4 Carbon - Hogwarts Voxel Built (best of 4)
Binkg
@dummerspast39Oct 17
I made GPT 6.1 Soul edit this video about the Hogwarts Castle voxel in a single HTML that Fable generated, Fable 5.5 obviously, and holy fuck, this voice is horrible. But yes, this is the... the crazy thing is how, uh, accurate it is and how much detail and taste it has, way better than Opus 5.5.
3
2
28
1,316
I am the only person who even cares enough to shit on this motherfucker @petergostev
1
3
351
Bullish on SSI
1
364
This is the most misleading chart all year: -GB300 ultimate rack vs DeepSeek on GB200 -peak prices vs off-peak prices, so half price for DeepSeek -the tweet calls it 50$ profit but their own chart says 51$ is the revenue Same prices and DeepSeek makes around 78$ per GPU hour on an older chip. Thats 1.5x the best vLLM setup on the GB300 and they framed it the other way round Now same hardware, same off-peak price, revenue vs revenue with their own numbers: -vLLM on GB200: 3.58$ a hour -SGLang on GB200: 5.38$ -DeepSeek on GB200: 39$ Thats 10.9x vLLM. GB200 rent in their chart is 4$ so vLLM loses 0.42$ every hour at DeepSeeks real price
SemiAnalysis
@SemiAnalysis_11h
Replying to @SemiAnalysis_
Assuming 60% utilization, GB300 generates $50/GPU/hr profit at 125 TPS with Openrouter's prices, while DeepSeek makes $39/GPU/hr off-peak. As per Jenson, COMPUTE IS REVENUE. (5/5)
3
1
18
2,266
Load more