News for hackers. Stuff that matters.
Re:More Agents Is All You Need: LLMs performance scales with... by trash_cat on Saturday April 06, 2024 @11:05PM (#39957826)
Is this not an incredibly expensive/unsustainable method? I agree with the sentiment that MoE is the way to go as the newer models will probably see diminishing returns. But the compute for a single prompt will suddenly increase 7-15 fold?
Re:More Agents Is All You Need: LLMs performance scales with... by bearjaws on Saturday April 06, 2024 @11:51PM (#39958062)
"all you need is a six figure OpenAI bill"
Re:More Agents Is All You Need: LLMs performance scales with... by Garlef on Sunday April 07, 2024 @06:11AM (#39959739)
if it replaces 2-3 experts that might be a good deal
Re:More Agents Is All You Need: LLMs performance scales with... by bigEnotation on Saturday April 06, 2024 @11:20PM (#39957904)
Yeah, see GPT 3.5 vs GPT 4 pricing.
Re:More Agents Is All You Need: LLMs performance scales with... by ukuina on Sunday April 07, 2024 @12:46AM (#39958242)
If GPT4 is 20x the price of GPT3.5, but it only takes 10x GPT3.5 runs to get similar quality of response (and likely faster), you'll still come out ahead.
Re:More Agents Is All You Need: LLMs performance scales with... by dimask on Sunday April 07, 2024 @05:24AM (#39959555)
I doubt that 10xGPT3.5 > GPT4. There are a lot of tasks that GPT4 can do and GPT3.5 just cannot. Also, in such cases I find that GPT3.5's hallucinations are quite consistent, so such a method is probably not gonna help.
Re:More Agents Is All You Need: LLMs performance scales with... by imtringued on Sunday April 07, 2024 @03:21AM (#39958893)
So what? It's not like GPUs are compute starved.
Re:More Agents Is All You Need: LLMs performance scales with... by ZachSaucier on Sunday April 07, 2024 @08:55AM (#39960440)
Also the amount of non-renewable resources used and emissions goes up 7-15x