jonahbenton

(2012)

I buy hardware from a us provider ( thinkpenguin), who have something like this, and from a european provider (turris) who also do. It is a potential business with discriminating customers. If the goal is dd-wrt, I would banish the thought of anything near mass market (Walmart/Best Buy)- you have to be doing millions of units to even get into that conversation and dd-wrt is not the vehicle to sell to millions. But if you are looking for opportunity, local llm friendly home device integration of which the edge router is a small part, and of which dd-wrt is irrelevant, is a big wide open space.

Sympathetic to other strategies but that writeup is wholly unconvincing. It comes across as astroturf.

Long ago I ordered an indoor algae system from a couple of CMU PhDs. It is very cool, it looks cool and has a nice water flow backgroundy sound, the algae grows quickly, allegedly doing the co2 consumption equivalent of 20 houseplants. Space efficient! But immaterial relative to human co2 production.

Junk. The plain language of the executive order differs from what this piece claims. There is no conflict with BIS work.

The permanence and effectiveness of every policy Trump declares is mediated by his attention span, its inherent legality, and his opportunity to profit from gatekeeping. I would be very surprised if this particular whim impacts model development in practice at all.

With what process, spec driven? Having AI produce all the code is pretty common. An agentic workflow operating at a higher level of abstraction especially for brown field is not.

OpenCode is a "harness" that manages conversations/interactions with LLMs. It does not run a model itself. An extended interaction can include- many turns of conversation, providing files for the LLM to process, having the LLM direct the running of programs called tools by the harness. The privacy aspect is determined by which LLM you configure OpenCode to use. If you configure it to use an LLM you host yourself, then you maintain your privacy. If you have it use an LLM provider you do not host yourself then your conversation and file and tool results and whatever else is sent to the provider and your privacy is dependent on the provider. There have been no reports of OpenCode communicating out of band with user data to any servers- it does do things like check for version updates and can download plugins and the like. But in that sense OpenCode is reasonably configurable for privacy. The default however is to use an external provider for LLM, not a local model, so you have to change that and not use it oob.

To have your own LLM provider, you have to run different software. Ollama is one option (though it also by default uses external provider) and llama.cpp is another. Jan and LMStudio are other options for actually running a model on your own hardware, to which you can direct OpenCode.

Run llama.cpp on your machine and point opencode to it

Way to put the work in!

Tyler is an academic economist. He is accustomed to thinking about the magic of the markets. No consciousness there. Sure.

But the "market" is a human invention around which there also exists governance / regulation / accountability / responsibility. That these structures exist- even when people try to evade them- is the existence proof of the thing that the word intent refers to, and that there is a feedback loop incorporating the modeling of intent in human behavior is what consciousness refers to.

The existence proofs are there no matter how poorly Tyler argues against them.

Any formulation of consciousness requires learned/stored distillations/compressions of reality, some form of object constancy, so, yes.

I thought this reference was to skill files, which do rapidly become technical debt- so what was the rotting inventory metaphor? But no this is in reference to human skill acquisition around each of the different agentic platforms.

(2024)