Sounds nice except that these are 1 very small scale model, 1 reranker, and 1 embedding model that are far from frontier LLM level. And they're not open sourced.
As much as I agree with the message, this reads like marketing copy trying to make a big deal out of a tiny model being hosted privately.
> Sounds nice except that these are 1 very small scale model, 1 reranker, and 1 embedding model that are far from frontier LLM level.
We've tried to take a first-principles approach to our end goal of 'legal superintelligence' that has involved identifying the areas of our domain most in need of improvement and releasing models that raise the bar on quality in those areas.
We've been around for a couple months now and ended up starting with retrieval and enrichment. The models we've released to tackle those problems have indeed been smaller in size than their competitors, yet they still rank ahead on open-source benchmarks.
Them being so small also helps with their accessibility — as I mention in our post, our models can be deployed on ordinary hardware, not a supercomputer.
Next on our roadmap is reasoning and research, which will require more infrastructure to support, but again, we aim to be judged by performance at the time of release.
> As much as I agree with the message, this reads like marketing copy trying to make a big deal out of a tiny model being hosted privately.
The point of this point is really just to reaffirm our commitment to sovereignty and accessibility and contrast our approach with that of major AI labs. It _is_ possible to commercialize LLMs while still keeping them accessible. A customer using a self-hosted deployment today does not need to worry about our models no longer being available tomorrow. We think that's a good thing. And moving forward, we want to keep that option available for anything we do, instead of trying to pull up the ladder while we're ahead.
> ... this reality only further cements our position that frontier artificial intelligence ought to be accessible to everyone, rather than concentrated in the hands of a select few frontier labs, ourselves included.
> To that end, every single model we’ve released has, from day one, been available for air-gapped self-hosting. We have no plans on changing that. In fact, we’ve doubled down on AI sovereignty, ...
Good thing about LLMs is that they can't put the genie back in the bottle, and long after OpenAI and Anthropic bite the dust (not wishing that but just saying given their trajectories), there will continue to be people, engineers and startups working on open source LLMs.
My hope is that we're able to somehow repurpose all of the GPU chiplets currently sitting in warehouses and in massive datacenters for broader consumer, academic, educational and non-profit consumption. It will create such great value and ripple effect creating and spreading hardware + computing literacy far and wide. Ugh, hope that happens.
Despite the fact that it was available only till 22nd June I largely call it a marketing stunt. But the fact that government took extreme measures to contain the danger that model violation exposes I think this move reflects towards more broader issue of accessibility. As of my knowledge not just US but major leading countries have been keeping knowledge classified. Major example is defence sector which heavily keeps information under legal and penalizing guards. It is tough to identity the misusing entities of this assets they go into heavy licensing and red tapism. Maybe AI would also meet the same fate in coming future is what I sense from this move.
There is a lot of information out there that can make one person have a huge impact that used to be gated by the difficulty of acquiring the knowledge.
When it took years and years of training to learn how to master something enough to use it with great impact, that effort was strongly correlated with the discipline and lifestyle to not misuse those abilities.
Completely gate-free LLMs would have very "useful" answers to a nutjob prompting "How do I cause as much destruction as possible with only what I can find at the hardware store" or similar. Sophisticated guards are still very hard to do to avoid giving a whole lot of power to someone not very friendly and not very disciplined.
Our models can be deployed on premises as well, though that is more of a bespoke offering at the moment. We've also been fortunate enough to have trained most of our models on our own private infrastructure. Your question raises a broader question, though, about sovereign risk attached to cloud services in general. For some, particularly governments, sensitivity is so high that certain workloads must run fully on their own hardware. I don't see that changing for now, and, in fact, in certain jurisdictions, the trend is turning against cloud services.
Well that's the neat thing about this, it paves the way towards that yes. We may be finally able to smash Silicon Valley + the SF VC mind rot for good, these entities have been far too damaging to society and they should have never been controlled by individuals consolidated in a single city (mostly).
If I found out my lawyer was using an LLM, I would fire them immediately. Why is everyone using this slop. An LLM can't even center a fkin div reliably.
Lmao, imagine doing all this instead of just centering the div yourself. This explains why everyone I work with that uses llms is slow af. Its all pseudo-productivity, its kinda similar to how people sit around customizing their nvim config feeling productive but not actually accomplishing anything. The difference is you pay money for this junk.
As much as I agree with the message, this reads like marketing copy trying to make a big deal out of a tiny model being hosted privately.
We've tried to take a first-principles approach to our end goal of 'legal superintelligence' that has involved identifying the areas of our domain most in need of improvement and releasing models that raise the bar on quality in those areas.
We've been around for a couple months now and ended up starting with retrieval and enrichment. The models we've released to tackle those problems have indeed been smaller in size than their competitors, yet they still rank ahead on open-source benchmarks.
Them being so small also helps with their accessibility — as I mention in our post, our models can be deployed on ordinary hardware, not a supercomputer.
Next on our roadmap is reasoning and research, which will require more infrastructure to support, but again, we aim to be judged by performance at the time of release.
> As much as I agree with the message, this reads like marketing copy trying to make a big deal out of a tiny model being hosted privately.
The point of this point is really just to reaffirm our commitment to sovereignty and accessibility and contrast our approach with that of major AI labs. It _is_ possible to commercialize LLMs while still keeping them accessible. A customer using a self-hosted deployment today does not need to worry about our models no longer being available tomorrow. We think that's a good thing. And moving forward, we want to keep that option available for anything we do, instead of trying to pull up the ladder while we're ahead.
> To that end, every single model we’ve released has, from day one, been available for air-gapped self-hosting. We have no plans on changing that. In fact, we’ve doubled down on AI sovereignty, ...
Extremely commendable.
Good thing about LLMs is that they can't put the genie back in the bottle, and long after OpenAI and Anthropic bite the dust (not wishing that but just saying given their trajectories), there will continue to be people, engineers and startups working on open source LLMs.
My hope is that we're able to somehow repurpose all of the GPU chiplets currently sitting in warehouses and in massive datacenters for broader consumer, academic, educational and non-profit consumption. It will create such great value and ripple effect creating and spreading hardware + computing literacy far and wide. Ugh, hope that happens.
When it took years and years of training to learn how to master something enough to use it with great impact, that effort was strongly correlated with the discipline and lifestyle to not misuse those abilities.
Completely gate-free LLMs would have very "useful" answers to a nutjob prompting "How do I cause as much destruction as possible with only what I can find at the hardware store" or similar. Sophisticated guards are still very hard to do to avoid giving a whole lot of power to someone not very friendly and not very disciplined.