This stuff is a nightmare for new manufacturers and is usually lobbied-for by large OEMs or to keep startups out of the market or as a patent trap
The most recent regulatory disaster that blew up a bunch of startups was mandatory lane keep assist for trucks in overseas western markets, which meant all new startups needed fancy steering racks which are very much not off-the-shelf, and it virtually tripled the cost of the software stack too
I don’t think this is true. All the models prior to Fable were honestly dumb as rocks, and Fable is too sometimes, but at least it’s helpful now and not a hindrance.
The future of AI most definitely involves making something twice as good as Fable that is virtually its own employee, and not on reducing inference costs because to be honest Fable isn't actually that expensive.
The real utility behind an AI model (imagining that it can be made twice as good as it is now) would be being able to scale a small business up and down instantly without hiring (to implement a new feature or whatever), which is costly and time consuming these days.
Is it as good as Fable..? Fable is the first model that mostly writes without the AI slop format for me, and so I can comfortably actually copy and paste most of what it spits out.
OpenAI models have always been the worst in my experience for verbose, slop formatted responses, with each generation increasing in sloppiness.
Just because something doesn’t perceive to be lightweight doesn’t mean that it isn’t. Many modern web frameworks are extremely lightweight (ie Svelte), but it’s complicated how they achieve it, and the source code looks like sludge.
It makes sense to me that distributing across more parameters results in models that can be quant more heavily (information theory - more bits available)
I wonder if anyone has figured out how the information is compressed and calculated the amount of information an LLM can hold depending on its size
80 tok/s which is kind of a lot for GLM. My experience running 80 tok/s on other LLM is that it ~seems faster than cloud inference, but that obviously depends what you use, in my case ChatGPT.
It’s like an angsty book written by a teenager with only a middle-school level understanding of the world: throwing a few completely random words in there to sound smart.
Hard to believe that something that writes so terribly is so good at mathematics, given that writing non-slop must be at least some part formulaic.
I don’t know about the advice on market size… nearly every 2nd founder says go into beach head markets and stay away from markets with high competition.
The ones in beach head markets seem to succeed more often on average (anecdotal)
Ah yes but because it’s a MoE 48GB active model, then it’s possible that we might be able to run it locally in specialised setups such as 256GB unified memory.
Many MoE models (seem?) to only require enough memory to load the active expert.
More detailed response below, but I believe the best way is to use a router that can filter DNS requests (ie Unifi), and set the upstream of that to your ISP.
You can set the DNS server of your entire LAN using Network options without setting it on the individual devices. It’s up to the end-device whether it uses it or not, but 99%-ish do.
Heck. I wonder if they surveil Outlook in western countries.
Now that Cloudflare Email is out of beta someone should build a self hostable serverless email stack based on it. Then we can deploy our own email with reliable delivery, but own our own data and only pay for actual usage as well.
A good way to work around this is to have a small server in your house as the first hop for ad blocking, and use the second hop (set the DNS servers of your DNS blocker) as your ISP.
If you use a Unifi router you can use it for this purpose (and caching) by broadcasting its IP as the default DNS server using Network options. It caches/forwards/filters any requests according to how you setup DNS in Firewall settings.