Viewing profile — colingauvin

16 DGX Sparks can run K3 at a reasonable TPS. So that's $64k. 2 DGX Sparks can run DS4 at 1 million context with 50 TPS so that's $8k. 1 A4500 can run 35A3B. Those are about $1200 …

Claude 4.5/4.6 - absolutely agree. Fable 5? From my (limited) testing, also reasonable to interact with. Opus 4.7/4.8/5? Absolutely smug and antagonistic and preachy. I'm constantl…

DS4 Flash Q2/Q4 mixed quant fits on a DGX Spark (a $4000 device which is not particularly unheard of expense for Apple customers), and is indistinguishable for me from Opus for my …

Prefill is survivable if you cache well. But what kills me is the context. Qwen 27 needs a ton of room for KV Cache. I guess not an issue on a 128 GB Halo or Spark, but if you are …

>1) Back then, even as a free user you'd be able to use the strongest model (even if with tight limits). Now, you need to pay to use Sol, and you need to pay to use Opus or Fable. …

They must really be feeling the commoditization pressure. I'm not sure what the way out of this is, ChatGPT and Claude are still good products, but they are not necessarily premium…

Is there any niche yet capitalism has not exploited to its fullest? Why do we think it wouldn't here?

Came here to comment this.

I had read that they were skipping 5.3 for 5.5 which was going to be ~3 T param.

I have found 27 to just be so much more coherent than 35: https://humanparadox.org/local-vs-frontier-benchmarks-for-my... It can complete multi-step tasks much better, and has a bi…

OK really cool. I've been working on something similar but I started from the opposite direction. I first developed a database that pulls in data from as many personal sources as p…

Curious what you mean by proactive? Could you share a bit more?

This has nothing to do with what I am talking about.

There's plenty of information https://en.wikipedia.org/wiki/CASP

>DeepMind started developing AlphaFold in 2018. In 2020, it was recognized as a solution to humanity's 50-year-old "protein folding problem," which sought to answer how amino acids…

I'm guessing that there are storage vendors that already do that - maybe like Liqid. But then you've got to traverse the NIC as well.

The problem is you'd have to traverse the PCI-E bus every transfer. Even with DMA it still has to physically get off the drive and onto the card. Even with the mythical PCI-E 6, if…

There's a Python API that's "unofficial" but has worked well for me for ages. I've got it hooked up to NextCloud tables and it imports everything once a day. I've now got years of …

Your comment is a stark reminder that I take for granted that the single biggest killer of human beings in history is simply something I don't have to think about. For some reason,…

Yes, which is why the immunological tests done in 2006 were not seen as conclusive. However, this time they found actual plasmodium DNA which suggests active infection at time of d…

>In order to use Ask Sage, the Army had access to 100,000,000 tokens as part of an annual subscription to an “enterprise pack.” Tokens represent a unit of output, either in text or…

Great piece. Was wondering whether to get an NVLink bridge for $500+ for my two A4500s for inference. This pretty much confirms it only really rescues you if you have nonsensical P…