[flagged]
Sakana Fugu
131–140 of 141 posts
Re: Sakana Fugu
#132For anyone finding this, I used this during the beta. Beats GPT-5.5 xhigh on complex tasks. Since it’s expensive and difficult to subsidize, use it for the most challenging problems.
OAI/ANT can subsidize their own subscriptions, so it’s hard to compete there. But the results I got from fugu-ultra were impressive.
Re: Sakana Fugu
#133Imho there are two dimensions here: Firstly different LLMs and secondly the strategy in which you break down the problem in an agentic fashion (e.g. break up to separate agents with own persona and then judge evaluates across all agents). You can of course mix-up the dimensions as well and that's what I have been tinkering* with for a good few months with some success. This was all done using home-brew setup running…
This is interesting. Would you share a few ways in which you're using this in your workflow? What about if you were to start a new project and test and built it out from scratch - how do you work this approach in without bogging everything down(including the simple things) down with overanalysis?
I only use this for high value problems/challenges. A lot related to life decisions including work, where to live, finances etc. It gives me a small army that can break down and slice, dice the problem in different ways then someone to reconcile it all and present it back to me. The variance in their opinions is the most interesting part of this project so f ar.
Re: Sakana Fugu
#134Is this just OpenRouter Fusion but from Japan?
Re: Sakana Fugu
#135Earlier quoted context omitted.
This is interesting. Would you share a few ways in which you're using this in your workflow? What about if you were to start a new project and test and built it out from scratch - how do you work this approach in without bogging everything down(including the simple things) down with overanalysis?
I only use this for high value problems/challenges. A lot related to life decisions including work, where to live, finances etc. It gives me a small army that can break down and slice, dice the problem in different ways then someone to reconcile it all and present it back to me. The variance in their opinions is the most interesting part of this project so f ar.
Thanks. I like the idea of doing this, myself, and glad you see a value from this.
Do you not worry about giving away your most intimate data to for-profit companies who have not signed to protect your data in a dignified fashion?
Re: Sakana Fugu
#136Earlier quoted context omitted.
I wish I only paid $200/mo for Anthropic! Multiply that by 20x.
What are you getting out of it at $4000/month?
Redesigned and rearchitected our company's backend and mobile apps in about 2 months with 1 engineer and 1 PM trained to work with Claude. We used a custom-built multi-agent harness to crank through lots of parallel tasks, which necessitated the API usage.
Re: Sakana Fugu
#137I learned about this today because one HNer linked to this post, otherwise I wouldn't have noticed it. Can I just say, what a horrible title. This is probably why I didn't notice it, because it doesn't tell me anything. At least mention its some sort of a new Agent that coordinates mutliple models and acts as one, which achieves impressive scores, I don't know, something at least. Not just saying "Sakana Fugu". That's so vague. Is being mysterious supposed to catch more attention?
Re: Sakana Fugu
#138This is a joke, right?
Not necessarily. There were some tests last year-ish from hf that showed that simply alternating (randomly) between claude and gpt (whatever their versions were at the time) on a task produced better results than either of them individually. So during a task, the first call was sent to one, then the other and so on. There's also the concept of "smart routing" requests based on some heuristics / embeddings. You'd get…
> There were some tests last year-ish from hf that showed that simply alternating (randomly) between claude and gpt (whatever their versions were at the time) on a task produced better results than either of them individually. So during a task, the first call was sent to one, then the other and so on.
Where can I read more about these tests?
Re: Sakana Fugu
#139Isn't this what perplexity is?
Perplexity is just a RAG search engine using one model at a time, from my understanding