DeepSeek-V3, every mention
13 scenes · ← back to DeepSeek-V3
tap a year for its mentions
every year anyone John Coogan 8Sholto Douglas 1Jordi Hays 1
Verbatim, from the transcripts: the passages where DeepSeek-V3 comes up
OpenAI Sets Sights on $750 BILLION Valuation
- ▶ 17:41 unnamed speaker And then December first, Deep Seek releases Deep Seek version 3.2.
Apple Considers New AI partnership
- ▶ 7:11 John Coogan And so you, if you take all the best practices from deep seek V three and R one, like the reasoning model, and you spend the money to go do the training and you pull all the data together and, and maybe you're not completely stealing the…
Weekly Recap - Elon Vs Trump, Ukraine's Drone Attack, Cluely Update & OpenAI CRO
- ▶ 1:31:05 Sholto Douglas Um, I think you have like deep seek V three and this kind of stuff like R one, um, which means that with that's like at least two rooms just to get to the scale of GPT four and GPT four was two years ago.
How Does GROK Compare? (Full Analysis)
- ▶ 0:12 John Coogan Across math, science, and coding benchmarks, Grok III beats Google Gemini, DeepSeq VIII, Anthropic Cloud, and GPT's four
The Future of Artificial Intelligence
- ▶ 4:15 unnamed speaker uh H 800 has the same compute as the H H 100 but less network bandwidth uh the misleading six million figure was widely circulated as the cost of v three but that doesn't include R and D hardware depreciation uh repeated experiments actual… 3 times in the scene
- ▶ 9:40 unnamed speaker So, you know, V three surpasses GPT four Oh, but GPT four
- ▶ 19:50 unnamed speaker These training pre and post V three utilizes multi-token prediction at a scale not seen before. 2 times in the scene
DeepSeek Update, Market Crash, Timeline in Turmoil, Is VC Cooked, Zero Cope Policy
- ▶ 36:43 John Coogan And so, uh, the last quarter of the article talks about the seismic waves rocking the industry right now caused by DeepSeq V-III and R-I. 2 times in the scene
- ▶ 52:12 Jordi Hays And the result is that they have something like 13 individuals working on the Lama stuff who each individually earn more per year in total compensation than the combined training cost for deep seek V three models, which outperform it.
- ▶ 53:57 John Coogan So at the bottom, yes, you need a robust model and that's what their deep seek V three is.
- ▶ 1:18:08 John Coogan So, uh, Dylan Patel says deep seek V three and R one discourse boils down to this.
DeepSeek, Stargate Update, Bees Do Enjoy Honey, Golden Retriever Maxxing, Bring on the Trilly
- ▶ 24:49 John Coogan And so last December they made waves in the global AI industry after benchmark tests showed that it's DeepSeq version three LLM, which again is not the reasoning model.
The END of Ski Season? (E26)
- ▶ 2:17:14 John Coogan While DeepSeqs VIII looks to be a stronger model at only 2.8 million GPU hours, a hundred, uh, 10 X less compute or 11 X less compute.