They're certainly welcome to do whatever they're like, and for a microkernel based OS it might make sense--I think there's probably pretty "Meh" output from a lot of LLMs.
I think part of the battle is actually just getting people to identify which LLM made it to understand if someones contribution is good or not. A javascript project with contributions from Opus 4.6 will probably be pretty good, but if someone is using Mistral small via the chat app, it's probably just a waste of time.
I have a recurring problem where I can't even read one of my favorite recipe websites (seriouseats.com) from my phone because the series of popups completely blocks the page, and can't be dismissed.
But if I ask Claude or Gemini for a nice version of the recipe, it works perfectly. I think there's a lot of own goals out there.
For what it's worth, most people already are doing this! Some of the subagents in Claude Code (Explore, I think even compaction) default to Haiku and then you have to manually overwrite it with an env variable if you want to change it.
Imagine the quality of life upgrade of getting compaction down to a few second blip, or the "Explore" going 20 times faster! As these models get better, it will be super exciting!
I think the fast inference options have historically been only marginally more expensive then their slow cousins. There's a whole set of research about optimal efficiency, speed, and intelligence pareto curves. If you can deliver even an outdated low intelligence/old model at high efficiency, everyone will be interested. If you can deliver a model very fast, everyone will be interested. (If you can deliver a very smart model, everyone is obviously the most interested, but that's the free space.)
But to be clear, 1000 tokens/second is WAY better. Anthropic's Haiku serves at ~50 tokens per second.
Do you guys all think you'll be able to convert open source models to diffusion models relatively cheaply ala the d1 // LLaDA series of papers? If so, that seems like an extremely powerful story where you get to retool the much, much larger capex of open models into high performance diffusion models.
(I can also see a world where it just doesn't make sense to share most of the layers/infra and you diverge, but curious how you all see the approach.)
I think there's clearly a "Speed is a quality of it's own" axis. When you use Cereberas (or Groq) to develop an API, the turn around speed of iterating on jobs is so much faster (and cheaper!) then using frontier high intelligence labs, it's almost a different product.
Also, I put together a little research paper recently--I think there's probably an underexplored option of "Use frontier AR model for a little bit of planning then switch to diffusion for generating the rest." You can get really good improvements with diffusion models! https://estsauver.com/think-first-diffuse-fast.pdf
I think the closest that this has come is in the form of GitLab, which pretty famously did a ton of the corporate work in the format of a very open Handbook (https://handbook.gitlab.com/)
In the early years, it was extremely, extremely open and comprehensive. I've definitely looked through it when I wasn't sure how to handle something at work.
I have a hard time believing that the right move for most organizations that aren't already bought into an OpenAI enterprise plan is going to be building their entire business around something like this. This ties you to one model provider that has been having issues keeping up with the other big labs and provides what looks like superficially some extremely useful tools but with unclear amounts of rigor. I don't think I would want to build my business on this if I was an AI-native company that was just starting right now unless they figure out how to make this much more legible and transparent to people.
I think it's actually already there. It's definitely possible to make these sorts of explainers with something like a Claude Code, you just have to spend a fair amount of time making sure that it's actually doing what you expect it to do. I think the biggest danger with something like a Claude Code is that you get something that looks functionally correct but that the details are suddenly wrong on. I wrote a little bit about this on my blog for some of the places that I've done visualizations actually, and I think it's remarkably easy to iterate on them now.
This is the equivalent of setting up a developer environment for charging a car. Once you have a car that's working, and you know how to connect to the app and charge it, almost all these problems go away. If you're in a place that has a lot of public chargers near your destination that you're already going to, then it's even easier, and it just becomes trivial.
That being said, I don't think I would want to rent a car that didn't have a place to charge it or a very easy-to-use fast charger nearby.
> So you are saying that these special hybrid seeds that are the first generation of combining two strains are the only ones that can perform well?
For a lot of crop systems, yes! There are obviously crop systems where you can do clones and some exceptions are always present in biology, but basically yes. Yes for all the big staple crops except Canola.
> And that using any other seeds, even the second generation of that same strain, is so bad and so easy to confuse that it should be outright illegal?
I probably wouldn’t make it illegal, I think farmers should be allowed to do whatever they want to! (My completely out of the loop guess is the government is trying to help small holder farmers who are reporting that they’re being scammed by these groups and that they lack the resources to do genetic testing to prosecute them for the fraud.)
> That is very hard to believe.
Such laws are in place to protect the IP of these special seed producers, to make their business model viable. That does have merit to a degree, you do want such companies to exist, but they should also have to contend with competition from other, perhaps less effective but cheaper, sources of seed.
It’s not really an IP protection thing, it’s an extremely difficult many year process to recover genetics on most hybrid crop systems. I don’t think most seed companies care about folks using saved seed, they know almost all farmers will buy good seed if they can.
> This doesn’t have much to do with protecting the farmers from being cheated into planting bad seed. And I am skeptical of the fact that even second generation seeds are that bad, or that these hybrids are really such a life-changing upgrade.
I think well answered by a parent comment, but the book The Wizard and The Prophet is pretty good reading on Borlaug and the green revolution. If you look at global food capacity vs population, it’s probably the single most important life upgrade for everyone of modernity.
(Small Edit: I should note that I’m not an agronomist, I’m just a guy who codes okay sometimes and that I’ve gotten to spend a lot of time talking to agronomists and smallholder farmers trying to make agriculture for small farmers work better.)
Hybrid seeds are ~100 years old and are nearly universally adopted across developed agricultural markets. They’re as controversial as “you should probably use source control” is in programming. You may be confusing hybrid seeds with GMOs.
To state once again, I don’t know much about this law or how the government believes it’s preventing counterfeit seed, but bad seed is a huge problem for farmers. I personally want farmers to be able to do whatever they want to with their farms!
I think part of the battle is actually just getting people to identify which LLM made it to understand if someones contribution is good or not. A javascript project with contributions from Opus 4.6 will probably be pretty good, but if someone is using Mistral small via the chat app, it's probably just a waste of time.