Thanks a ton Ryan! i think it’s still early with them. they are just getting to the point where they are reliable. honest assessment is they probably need another 6 months or so for most people, but their getting really good. add in a new open source harness from
the CC leak, and i think end of the year local models will be usable for the average user.
i think we will get a harness from the CC leak, and an few general purpose/ public and trust MCP servers. so for most people looking to automate, study, research, write etc…
it will be essentially free. at the yearly trends, SaaS LLM providers will move more towards coding / security. their tokens will be “smarter” but more expensive.
when local models can answer most people’s questions and do their tasks with a low hallucination rate, a lot of people will want to save money.
Suggestion: make a post about how to navigate Huggingface and track a model from official base, through fine tuned, and down to quants. Explain popular methods to post-train and compress.
If you're running local models, make sure they are secure.
5 pro tips on staying secure.
Feel free to AMA!
I had a question regarding running claw in a sandbox on an Ubuntu box. What do you recommend? ToxSec thank you so much!
yeah that would be very secure. i run openclaw on a separate dedicated machine, but if i didn’t have that, your set up is how i would go!
thank you!!! 👍🙏
of course! thank you as well :)!
Layer 2 is what I need, so practical and useful. Thanks a ton ToxSec!
awesome to hear 😁🔥🔥🔥
So practical Tox. Local models are on 2026 roadmap, will need a rain check on that AMA for another quarter.
Thanks a ton Ryan! i think it’s still early with them. they are just getting to the point where they are reliable. honest assessment is they probably need another 6 months or so for most people, but their getting really good. add in a new open source harness from
the CC leak, and i think end of the year local models will be usable for the average user.
Great article! I really appreciate the specific steps you gave us to verify a local model!
thanks a ton Erich, appreciate that. Local models are going to be a powerhouse by the end of the year.
Sure thing!
What do you think the most common usage will be?
i think we will get a harness from the CC leak, and an few general purpose/ public and trust MCP servers. so for most people looking to automate, study, research, write etc…
it will be essentially free. at the yearly trends, SaaS LLM providers will move more towards coding / security. their tokens will be “smarter” but more expensive.
when local models can answer most people’s questions and do their tasks with a low hallucination rate, a lot of people will want to save money.
Money is often the best motivation after all! Thank you for your opinion. Definitely an interesting one!
Suggestion: make a post about how to navigate Huggingface and track a model from official base, through fine tuned, and down to quants. Explain popular methods to post-train and compress.
i should totally do this. great suggestion. i’ll start looking into this. screenshot or maybe a video could help.
DONT
hahaha. 😂