Consumer-centric customizations and non-free package options.
It’s somewhat akin to answering “How can I add stuff like OneDrive to Linux and not have to configure/self-host everything?”
I used to like Ubuntu and Canonical products and these days I’m more indifferent towards them. Their products have a place, but I wish Canonical would stop killing off some of the more interesting side projects.
I wouldn’t say it’s better, because that’s very subjective, but it is more “feature complete”. Alacritty is very minimal and rejects community PRs for features often driven by their general ethos. libghostty supports a lot of features I don’t foresee making their way into Alacritty.
I use Alacritty and GNOME Terminal. I don’t have a great reason for not using or trying Ghostty other than I have a minor aversion to Zig. (Again, not a great reason and arguably a not good reason)
Well they did pull back on ZDR a few times and at least Anthropic has made their arguments (that I disagree with) that model outputs, no matter how much user context was fed, do not fall under their definition of "user data".
Are these prices subsidized by user generated outputs?
Can I even trust any answer to not be carefully worded and rephrase the response to "user data"?
Even so, what's to prevent some "oops" event that seems to be all too common?
At least some months back Bedrock seemed to be properly isolated, but I've heard of many getting non-ZDR notices for new models.
> > one NVIDIA GB200 NVL72
That’s around ~$4-6 million in hardware cost (one rack, 72 GPUs + accompanying hardware).
that’s a good clarification, skimming some of the comments here I was thinking they trained it on a single GB200 gpu. THAT would be a huge win for indie AI
Edit: Nvidia GB200 does not refer to a single gpu, Nvidia’s naming scheme is very confusing. Considering past gpus were called H100, H200 etc, I thought GB200 would just be a newer GPU, but it does actually refer to their 30+ gpu rack scale systems. The claim of “we trained this on a single rack of gpus” is definitely much less impressive than “we trained this on a single gpu”
> Nvidia GB200 does not refer to a single gpu, Nvidia’s naming scheme is very confusing. Considering past gpus were called H100, H200 etc, I thought GB200 would just be a newer GPU, but it does actually refer to their 30+ gpu rack scale systems.
It's even more confusing than that :)
GB200 = 1 CPU + 2 B200 GPUs (packaged together)
NV72 = A rack of 72 B200 GPUs (or 32 GB200 superchips)
Think of GB200 as something like Apple's 'Ultra' chip. Not exactly, but similar. They fuse the chips together with a very fast interconnect (connection).
Also a speech to speech model it can't do any tool use or you can't guide it much am I mistaken? How will it be used for agentic uses is the part I don't get. Edit: this is response from Argon if anyone was also curious: You are not mistaken about Dessa: because it was trained solely on raw audio with zero text data, it cannot read text system prompts or output code for tool calls on its own. To use it as an agent, developers must either steer it purely through spoken voice instructions or pair it as a conversational "voice front-end" alongside a separate text LLM that listens to the audio and executes the tools in the background.
> they didn’t have to worry about hosting/billing/subscriptions or any of that yet
Or marketing, which they kinda get for free. So much research goes unseen, there’s huge competition for attention, and most research is very limited in that regard
I would say that’s the biggest factor that makes the team “not indie”
At the same time, like you said, still amazing accomplishment, very cool model, and great work
I find it so weird the wall mounted version is a different product, it seems like a lost opportunity to do more interesting things with a speaker base.
The problem is if they made it battery powered with a speaker dock, people would criticise them for just making a nerfed ipad with no app store, which is what it is.
Reading the “paper”, this is just a job runner built on Ollama.
Exo, and the teams at Apple that built the foundations Exo uses, actually use modern distributed compute.
Apple has released some small models previously that seemed genuinely great, along with some quality research/efforts. I’m just not seeing the same level of depth here, am I missing something?
Other commenters are suggesting you can’t blame the spec for end implementation mistakes, except that’s one of the many issues - JWT being so error-prone is a problem.
I use JWT just for handling of tokens, because it’s so well supported, but I won’t use it for anything more than token storage _because_ it is so vulnerable to mistakes.
The fact that mistakes are so easy to make is indicative of poor design in the spec itself.
This is a bit nuanced, because if there’s redundant constructs then that does impact maintainability and efficiency with both runtime and LLMs working with the code.
I’d push this more towards personal preference of how code is expressed matters much less now than how maintainable it is.
There is the aspect of long type names, they often don’t have much impact when tokenized. The character count of words is nearly negligible - they often become one or two tokens anyways. But, the choice of words may have a greater impact on how the word choice weights an LLMs contextual processing of that word (a human may be able to ignore an inaccuracy in naming a bit more flexibly than some LLMs).
I used to like Ubuntu and Canonical products and these days I’m more indifferent towards them. Their products have a place, but I wish Canonical would stop killing off some of the more interesting side projects.
reply