What are ways that y’all avoid/manage goodharting?
Do you have a process to change goals if getting politician’s signatures seems less valuable?
Beckeck
Karma: 359
The initial release is just in the US. It might get an international release if sufficiently successful, but we don’t know. Similarly we don’t know about streaming, but i’d bet it will get onto some platform eventually.
Also have this issue on galax s24. (And not on other parts of the website)
Hope this goes well!
random pitch but—maybe add anki integrations for extra nutritious content?
Yes, see PauseAI (even if I disagree with some of their positions, i’m glad they exist and hope that soon there exist multiple such orgs (but don’t donate to StopAI, they don’t appear serious imo))
upvoted for topic importance.
thanks, I appreciate the reply.
It sounds like I have somewhat wider error bars but mostly agree on everything but the last sentence, where I think it’s plausibly but not certainly less worrying.
If you felt like you had crisp reasons why you’re less worried, I’d be happy to hear them, but only if it feels positive for you to produce such a thing.
we might disagree some. I think the original comment is pointing at the (reasonable as far i can tell) claim that oracular AI can have agent like qualities if it produces plans that people follow
yeah, if the system is trying to do things I agree it’s (at least a proto) agent. My point is that creation happens in lots of places with respect to an LLM, and it’s not implausible that use steps (hell even sufficiently advanced prompt engineering) can effect agency in a system, particularly as capabilities continue to advance.
“Seems mistaken to think that the way you use a model is what determines whether or not it’s an agent. It’s surely determined by how you train it?”
---> Nah, pre training, fine tuning, scaffolding and especially RL seem like they all affect it. Currently scaffolding only gets you shitty agents, but it at least sorta works
Top post claims that while principle one (seek broad accountability) mightbe useful in a more perfect world, but that here in reality it doesn’t work great.
Reasons include that the pressure to be held in high standards by the Public tend to cause orgs to Do PR, rather then speak truth.
know ” sentence needs an ending
“ARC (they just changed names to METR, but I will call them ARC for this post)”—almost but not quite—
ARC Evals (the evaluation of frontier models people, led by Beth Barnes, with Paul on board/ advising) has become METR, ARC (alignment research center, doing big brain math and heuristic arguments, led by Paul) remains ARC.
″ football, hockey, rugby, boxing, kick-boxing and MMA to be amongst the worst sports for this stuff.” - - I’m not up to date on the current literature but I’m pretty sure this list is rather wrong. I don’t remember all the details of the study I do remember (and I don’t have time for a lit review) but in it women’s high school soccer actually had the highest concussion rate (idk if it was per participant season or hour or per game minute or...).
I accept that mayonaise is an evolution of allioli (but maintain that the historical fact is that its american ubiquity routes through french chefs).
Wikipedia also doesn’t say that it’s not a mother sauce, if you scroll down you’ll find this:
“Auguste Escoffier wrote that mayonnaise was a French mother sauce of cold sauces,[27] like ’espagnole or velouté. “
I originally wrote “controversially a mother sauce” because the most common listing of mother sauces on the internet is ~wrong- The youtube video i linked includes primary source scholarship on the topic that has begun to update the general understanding in the direction that the quote supports.
This is as much a nitpick with Zvi’s article as with this one, but french food just seems hard to find because its easy to misidentify. french technique is the bedrock of american food—both as the history of fine dining(/haute cuisine) routes directly through french chefs, restaurants, systems, and techniques and as french food has been repurposed into american food.
Some examples: mayonnaise, the delicate, challenging-to-make emulsion of flavored fats and vinegars,controversially a mother sauce* becomes ‘mayo’ the white stuff that goes on sandwiches; charcuterie becomes the deli isle; boeuf bourguignon becomes stew.
so in your example you can probably (haven’t researched the restaurant, but from general knowledge as a processional chef) count at least the “new american” restaurant as french as “new american” is the (new(ish)) American take on a fine dining tradition that comes from france. ‘Chef’ just means ‘chief’ in french (like the military rank or the man in charge) and comes from the brigade system (https://en.wikipedia.org/wiki/Brigade_de_cuisine)
((https://en.wikipedia.org/wiki/French_mother_sauces) https://www.youtube.com/watch?v=tcDk-JcAnOw )
If it’s the case that the game theory here is correct I’m sad why it can’t be simply explained as such, if the game theory here isn’t correct I’m sad it’s curated.
plus one for “stop worrying about what people will say in response so much, get the actual information out there, stop being afraid.”
see also Anna Salamon’s takes on ‘not doing PR’ that someone else might find and link?
I appreciate your work on many issues, but find myself pretty strongly disagreeing.
finding minor criticism of a sub point that is technically valid (often only if you interpret things noncentrally) doesn’t mean the broader claim is wrong.
Said imposed more cost then provided value, and refused to cooperate with extensive efforts to find third options. Banning that seems productive.