Comments - Dean Ball on open models and government control
SubscribeSign in
Dean Ball on open models and government…
Subtle precedents on the future of open models set by the unfolding Anthropic v. Department of War case. Listen →
10 Comments
Top first
Interconnects AI reply rules
Liked by Nathan Lambert
Really nice. My morning commuteReally nice. My morning commute listen. listen.
Reply
how does average non tech person understand all this? technolog? funding?
Reply
Wouldn't stronger closed models help with developing harnesses for open models? Given the chants of "coding is solved", it's rather opaque how solved they are. But it seemed like an advantage for keeping the open models close to the capabilities of the closed models, as much the outside world is aware of the capabilities of the closed models.
Reply
Seeing 'government' as the threat to liberty is cartoonish. Our founders understood that tyranny could come from any concentration of power, public or private. Madison warned explicitly that wealthy factions were the most durable threat to self-governance. They designed a system so that no single concentration of power could dominate, and crucially, they gave the people the means to direct where power could be applied. That's what democratic governance is: not a threat to liberty but the mechanism through which citizens collectively defend it. Right now, that mechanism is being dismantled at the will of a handful of people with more wealth than most nations, while those same people control the media that tells us government is the problem. The answer isn't less government. It's reclaiming the one institution we actually have a voice in.
Reply
I always wonder, where does funding come from for your models?
Reply (1)
Author
Allen institute is from the late Paul Allen (MSFT)
Reply
I admire the fast development of models, but I've wondered why the efficiency of use has been rather neglected. My educated guess, based on e.g. what we have done with episodic memory use in our Venho OS, is that innovation on this side can be a significant lever and even alleviate model limitations. The question is of course, what are these "aspects of use" that could have such power. I have my experimental psychologist and HCI hat and see quite a lot of powerful opportunities. Thank you for your inspiring views on this journey.
Reply
Nathan, have you used GPT-OSS-120b? How does it compare to Opus 4.6 from a feeling perspective.
Reply (1)
Author
not in a good apples to apples way no not really
Reply (1)
Mar 6Edited
Hmm, sad days. I am transferring my OpenClaw from Opus4.6 to a DGX Spark.
$100/day is a bit much for me for API costs. 😣
Do you know of any non-Chinese OSS models that feel like Opus yet?
Reply