SubiculumCode

Are any of these multimodal yet? I'd love to try asking a model with calibrated probabilities to answer question like, "do these shapes match?". Sure, you can ask a LLM....

Exactly my point.

It's obtuse because you are claiming that the academic peer reviewer's decisions on NIH study sections, for example, are driven not by the quality of the application right in front of them, but by how that application will, in the long-term, alter their own funding possibilities in the future. Maybe in the abstract they know this, but it's not a driver of individual decisions...it's analogous to theories of population level selection in genetics..yes it could happen, but it is much weaker than natural selection at the individual fitness level.

And to be clear, it's not that I am claiming that there are not bad incentive structures in science-- there are; I'm saying that peer reviewer's on grant committees/study sections are not the root of those problems...not propagating those problems with their decision making on individual level grants...it's way too cynical of a suggestion. Instead, blame the laws that govern grant funding mechanisms (we spend way too much time begging for funding instead of doing science), blame Universities for using impact factors and publication counts as proxy metric for researcher quality, etc. But peer review itself? Only if you blame scientists for eating the only food that is given to them by the system-level structures involved. So, yes, obtuse.

Now you are being unreasonably obtuse.

Except, at least in America, almost all research funding comes from the Federal government, from a process of peer review, by reviewers who have no conflicts of interest, or stand to personally benefit by the applicant's success. So while I appreciate your sentiment, I don't see how it applies.

The author provides no empirical evidence to support their claims, providing only anecdotal evidence. Their claim might be true, might be truer in some places/departments/countries than in others, but I do happen to agree with the author...or I am concerned about the issue.

Also anecdotal, as a peer reviewer, I too often encounter shoddy, careless, or even probably fraudulent conduct papers, and it frustrates me because peer review is not cost free: Authors are asking scientists to spend their precious time to understand the author's work, and that only works under a social contract....that authors only send out manuscripts for peer review after they have tried their best to do their best science.

I hate to say it, because I know that China has great scientists, but most of the manuscripts I receive and reject due to sloppy work, or purposeful negligence in citing most directly relevant to their manuscript (often because that literature did exactly what they are trying to pass if as new), or out-right making shit up, have been manuscripts from Chinese Universities. Yes it happens from western authors as well, but what I am naming is real, and is increasingly a major burden on peer-review as a scientific institution. So much so that many have confided to me that they will refuse to serve as a reviewer of a manuscript out of China unless it is from a known lab of some standing. In science, there must be a culture motivated not by number of papers published or grants funded, but by its rigor and integrity, and like the finest craftsman who cares most cares about doing the finest of work. That matters, and that culture of integrity and personal pride, does need to be encouraged and rewarded, no matter where the lab is located.

This is all to say that I agree with the author, despite their lack of empirical evidence, that these values are under threat.

More than just hard budget caps, we need agents to be aware of those hard budget caps, and plan accordingly. Too much agent behavior is driven by the immediate proximal goal instead of long term, strategic considerations. What we have now is agents deciding to on embark on expensive, circuitous routs to a goal, perhaps with potential but not certain in any case, benefits, and intervention depends on either an attentive user or forces stop after the money pot dries up. Prior research (#?) has Dem nsttated agent behavior to adjust behavior in response to budget considerations that appeared to enable more efficient token usage, and if that means it needs to be imposed on a custom harness (and not the self interested ai provider), so be it.

Aren't most hard things like this?

That's why knees hurt in old age? They forgot they were knees, decided they were sentient cheese?

Becoming such is not my goal; I have too many other interests. It was more commentary about how one might be good, but no where even near competitive, and the journey, even if mentally capable, of improving further would take years and year of full time dedication...because chess is that deep, and because "good" is always context dependent.

Chess...I'm in the top 1 percent among regular players on chess.com/lichess, yet impossibly far away from Master/GrandMaster level as to be impossible; their minds are completely alien to me.

Or another way of saying it, when the default no-thinking baseline strategy can win frequently enough by chance that it becomes difficult to learn from a single game. In this case, just 'buy silver, then gold, then provinces) can win too often just by chance.

perfect place for it

Reminds me of the UC Davis Death Star [1][2]

[1] https://localwiki.org/davis/Social_Sciences_and_Humanities_B...

[2] https://www.youtube.com/watch?v=mNjfGUh-Rhk

Infinite bugs, or limited supply that can be extinguished. That is the question.

You made this in a day?

I was glad to see that asking who I should vote resulted in a gentle refusal.

They did it before "AI" using machine learning.

Loved that game...and there is definite potential here. It seems like a player can just move right next to opponent and fire? That doesn't seem right?

t/s maybe? IDK, because their token speed comparison was against Sonnet 5.

Yeah, I did kind of feel like the step down from Opus 5.5 was so large as to never make it appealing.

Feel our power rarrrrr!

*"Forget the map, just tell me if I should get off the Eisenhower Expressway and take Roosevelt, Madison, or Lake back home?"

I'm driving some place new, but of course, I first have to go through all my neighborhood streets to the freeway, and then later, eventually, something unfamiliar. I call up Google maps and it calls out all the play by play turns and such, starting from my house, instead of just saying get on the I5 North for 70 miles, then shuts up until I am close to the unfamiliar exit. It's why I mute it, I hate the blow by blow through familiar territory. Google knows where I go, God damn them, why don't they use that information already?

How long will AI "growing pains" be tolerated?

Not steal, but keep it indefinitely, per the JEV TOS

Where did he get the attractivess distribution, from the paper??

So what is the way forward for Linux to have sandboxi g as strong as Android? Appimage everything? I know it's possible now hacking together things, but I mean by default and integrated.

University of California has licensed zoom accounts with privacy agreements. I install it in Linux. Now I will check if I can replicate this behavior. I am not sure if it is a violation of their agreement or not if they do, but I don't like it.

Anyone can level a charge of AI writing, but it's getting to the point that you must write sloppy to seem human. and never never write well with clear technical prose. It is getting to be just ridiculous McCarthyism.

Interesting. I admit I was largely going on what Mr. Harris said on the episode when responding to several posts.