_6wbq

Currently burning money quickly on official deepseek api. They are also increasing pricing starting today. V4 Flash 0731 still feels like the most outstanding model of the past few months and probably to come.

So not worth it over flash? Even at ~7x the size it isn't worth the price hike. Flash may be a monster of a model due to all the RL it received from free usage everywhere.

These LiquidAI models have never worked well for me in practice.

Makes me feel hopeful. Things felt more positive around the llama 3 era. Now it’s like a dark, dreadful race.

You personally? Just curious. Context window is also a factor and ram isn’t really cheap. Sparks are assembled units which I like.

There have been discussions on language specific not really being a relevant change to reduce size.

What I think would be perfect is a model that could run on a single DGX spark and be competitive with DSV4 Flash 731. Flash is already a game changer. Hopefully meta plans on this, like the old 70b. V4 flash is smart enough for any use but slightly too big. 27b-30b isn’t intelligent enough.

Dr. Tom Murphy seems highly relevant here. https://dothemath.ucsd.edu/2011/10/the-energy-trap/ also: https://escholarship.org/uc/energy_ambitions

Disagree. It’s all about the plot.

Have already seen countless major ad spots using ai generated videos; big name companies. One major use case.

You can easily tune how it responds? This isn’t really a problem am I wrong?

Ha there were a few buttons on the page that might’ve been distracting lol

I thought the clips were real footage until they were dancing in a flooded room.

I know this is true and all but I think a lot of people never really write anything anyways. Texting, random notes, emails. Strange to say but it feels kind of hipster and performative with the technology we have, which kind of makes it hard to enjoy. What I mean is I understand the theory but never have a chance to implement it or even want to.

Now when I see new software projects I just think of how much of them are made with agents and the unreliability/easy abandonment that comes with it. Ten years ago this product would have been of a known quality. Not saying anything about buzz specifically.

I read this as a soft let down to not expect too much from 3.5 Pro.

> We have started our most ambitious pre-training run yet, for Gemini 4, and are excited by the progress.

I think they have a compute shortage and this wouldn’t look very good to investors.

Great concept but buggy so not a daily option imo

What would be the point? Kimi is kind of forcing their hand.

If their claim is accepted the rest follows logically

I was also about to go back to CDs but realized it’s a pain, and streaming is a completely different, fluid experience. CDs are expensive, if you listen to new artists, it’s a whole process just to buy one. On the other hand, streaming apps now have recommendation algorithms and top charts and paid promotions, etc. My solution has been using Apple Music’s web player with uBlock origin tweaks. Yes I know it’s not lossless, but it’s the best I’ve come up with.

It’s rage bait

Sol came out (public access restrictions that Chinese models don’t have to worry about) just a week ago.

This would be very compelling. Can anyone share more details on how it would work? Only issue is that you are stuck at a certain point in time but that’s not a huge deal. Even just a good 27b model would be useful.

Looks like open models being months behind is a thing of the past. Now more like weeks.