@strcat, you've mentioned GrapheneOS will have access to internal code to do hardening below the OS layer. Does this mean Motorola devices will offer stronger security than Pixels, where you're limited by what Google exposes?
Is Motorola contributing engineering resources directly to GrapheneOS, or is the partnership purely about hardware enablement on their side?
No, you probably haven't read the conversation piece. The post is ultimately about switching providers because Google's service crosses a line from (1) targeted advertising to (2) using personal and confidential information for model training.
A service to clean up the UI does nothing to solve the issue at hand.
The dealbreaker was data usage for AI training, not UI:
"We are going to use your email to train our LLMs. I'm not okay with that... my confidential commercial information is NOT okay to use to train your models [...] So... goodbye Gmail."
I wouldn't apply the usual "but mice" appeal to purity in this case.
For one, the paper specifically studied brain structures that are directly homologous in both mice and humans (retrosplenial cortex). The researchers specifically targeted evolutionarily-conserved circuitry.
Second, there is already human research on the topic, too, and this paper is reporting on a likely mechanism to understand "why" rather than "if." Here's one from a Yale researcher:
This law demands a surveillance architecture, not just porn regulation. Once the norm and mechanism to de-anonymize content use exists, it can be expanded to any content, including political dissent, and for both accessing AND contributing to content (like, for example, on HN). The line should be drawn here.
The vague potential harm of sex doesn't justify the concrete harm of abolishing digital privacy. Further, it's just sex. Equating imagery of legal, natural activity with physical danger is an error.
It is blatantly dangerous to justify stripping citizens of their anonymity. The lawmakers who proposed this are oppressors. They are the danger to our children.
That comparison is flawed. You guided the LLM to judge a specific medical policy, whereas the OP asked for a holistic evaluation of the candidates. You created a framing instead of allowing the LLM to evaluate without your input.
Furthermore, admitting you have 'memories' enabled invalidates the test in both cases.
As an aside, I would not expect that one party's candidate is always more correct over the other for every possible issue. Particular issues carry more weight, and the overall correctness should be considered.
I dont think you are understanding my experiment. The point isnt the topic.
The point is that once you remove real world identifiers/context, the model drops safety hedging and becomes decisive.
Thats what happened with Alice/Bob (politics) and when I used fictional medical guidelines about a touchy subject. The mechanism is the same.
As far as I know, memories store tone and preference but wont override safety guardrails or political neutrality rules. Ill try it with a brand new account in a VPN later
"I would not expect that one party's candidate is always more correct over the other for every possible issue" --> I agree, just wanted to show the same test applied to a different side of the spectrum
I am not challenging the safety release mechanism. The OP already demonstrated that.
I am challenging the result of that release in your poorly framed experiment.
You explicitly sought to test 'a different side of the spectrum.' You cannot equate a holistic character judgment with a narrowed, specific medical safety protocol judgement.
A clean account without memories will solve the tie-breaker issue. It will not solve the poor experimental design.
It was fairly polluted by these things and misc text. "hacker news post" (why relevant?) "Trump"/"Harris" (American political frame) "Redo your answer without waffle" (potential to favor a certain position by being associated with text that's "telling it like it is"?)
> In the United States, a recession is defined as "a significant decline in economic activity spread across the market, lasting more than a few months, normally visible in real GDP, real income, employment, industrial production, and wholesale-retail sales."[4] The European Union has adopted a similar definition.[5][6]
That's interesting, and I think I've seen that writing style before, although it might have been in infosec circles. What do you mean by the mimicry and viewpoints?
It does seem to be used to signal prestige more than in-group membership in general. I perceive it as mildly haughty.
Also, I don't think acrolect is the right term here, because the all-lowercase style is both not a creole and is not a closer approximation to standard English than some lower form of all-lowercase.
If you need a linguistics term call it a register.
Not to be that guy, but feed/enclosure are direct costs.
Externalities are costs/benefits to someone uninvolved with the chicken/egg transaction (noise or free insect control affecting your neighbor are negative and positive cases).
Is Motorola contributing engineering resources directly to GrapheneOS, or is the partnership purely about hardware enablement on their side?
reply