ThirdConsul

I hate AI too

? Never said that.

Anthropic claims that Mythos has low low low hallucination rate of just 40% in its System Card.

LLMs outputting code solving this exact task is a compound function of luck, with non determined a priori chance of success and unknown a priori cost.

This is my main disappointment with agentic coding. Prompting is fine, quality assurance is there from the start. Greenfield, couldn't care less. Established enterprise code? This is a minefield.

If the tokens were 10 to 100 times cheaper, then it would be a maybe.

I also hate how it makes half of my senior engineers dumber.

You mean LLM, not agent?

Or are you using agent to mean "instance of specific LLM model hosted on X"? Your description is also a bit weird offline + firewall (firewall against what? Local hosting environment?)

kill months

It's called passage of time sir. :P

Don't. They had 180 million usd income, with operating cost of 3 million usd, and they currently have over quarter BILLION is net assets. They (Wiki foundation) are at the point where they become cancerous to the Wikipedia - e. g. they are currently union busting.

It’s 90% right 90% of the time. But the most important part is the last 10%.

No LLM model system card claims such high rate as 90%. Fable claims in its system card, what, 40% hallucination rate?

I am also curious what AnBee tried to edit and was refused.

You think building an EV is straightforward?

While I don't necessarily agree or disagree on the broadsr point, building EVs is straigthforward, albeit a shitton of EV parts are still manually crafted before they arrive in the "dark factories" for the final assembly (or after, dealers choice).

Last time I checked a Xiaomi dark factory was utlising like a 1k sub components production lines from all over China delivering shit.

I literally direct you again to fucking google that, because the first response from Google literally tells you why that will not fucking work.

Are you so entitled that you demand we copy paste it for your pleasure.

The google will include the over-the-air elevated fibreglass that costs 20x more per mile than anywhere else, why it would cost one third of that number a year to maintain, why it cannot go underground (because of permafrost melting when being dug up and turning into mud and bog and sinking the installation and million other things), thst there are no roads in many places at all, etc. etc.

Your laziness is offending.

Ah. I see. You're thinking to let the fiberglass cables lose on top of permafrost like it's a hose from a shed.

If you're able, you can learn why that is a bad idea online. There is plethora of reasons why fiberglass cables usually go underground.

Which part of permafrost do you not understand?

LLMs generally work in one way. They get the prompt and give an answer. CLAUDE.md, system promp, rules, memory, tool defintions, mcps are different ways to prefix your prompt with extra information or context.

Skills, or plugins, are a way to inject less information until is needed (you can think about them as prefixing your prompt with "if you are asked about pizzas, add to context separate file pizza.md").

What you could add to CLAUDE.md depends on what you're doing. Generally it should be context LLM cannot infer relevant to all/most task performed in given project.