xpct

You can definitely try to regularize against ruleset changes by generating a bunch of cards and making the agent play in randomized subsets of those cards.

I didn't look for prior work on this, but my estimate is that it's probably within 2-3 orders of magnitude of additional training compared to a static game. (Still a lot!)

They already cap the actions per second a bot can take, sometimes to very low numbers. I don't remember if it was AlphaStar that limited how often you can move your camera or not, but some work definitely did.

I personally don't see why vision is important, if anything I'd frame vision as useful to humans, rather than being the baseline.

Oh, I like the point about not reaching 100% even with benchmaxxing. Do you know what the problems are that they fail?

This is actually why I've been reluctant to setup my own degradation trackers. I'm afraid it might be too much of a time investment for something that's much easier for them to detect.

"A couple more step functions" is doing a lot of heavy lifting here

Thank you for sharing :) good food for thought

Because we're afraid students might lack self control, or be unable to evaluate how well they absorbed the material.

That's entirely fine. Have midterms that are also interviews. Have labs where TAs interview students vocally every few weeks, etc.

The AI would just output that it can't be solved..

I'm in the same boat. When I started thinking about LLM development as multi-year iterative training adjustments, it felt like the majority of shortcomings can be addressed.

I'm sorry, I just don't buy the hyperbole. Something with your previous approach was deeply wrong if you condensed 20 years of learning into 2 months.

I don't entirely understand the point you're making, but I disagree with the last part. People not only have different needs from the media they consume, they also aren't equally rational with how they decide what to consume.

I never found a real life use case for memory tricks. Would be curious to hear if someone has.

I think it's half culture, half being text-based. HN is heavily self-correcting when it comes to low quality content, trolls, spammers, clickbait titles.

It feels very healthy how I can choose what to engage with, and what to ignore.

I have a bad habit of asking my LLMs such questions repeatedly..

I think we could agree that everyone already does this up to a point. This is basically just an exploration-exploitation dilemma, where everyone tries to maximize how much value they gain from the things they read, and a lot of people have strategies to pick what material to read.

Reading trusted authors is likely to be valuable, but they may devalue over time. We don't know if our chosen authors maximize our value, nor whether exploratory reading does, etc.

https://en.wikipedia.org/wiki/Exploration%E2%80%93exploitati...

Very sorry to hear this, I thought phone theft was becoming rare nowadays. Did it get stolen abroad?

Isn't that weird? The full knowledge of how to make such chips may one day be accessible to anyone, yet only the entrenched companies will remain the makers.

If we imagine machines being able to do the full process end-to-end, and the quality of that process only dependent on capital spent on tokens, I don't see how new companies could ever enter the market.

I don't work in the business, but my understanding was that even with these companies' budgets, it's still too expensive to do any kind of verified performance optimality.

Aw, I was expecting more details but this just seems to be a rehash of what they unveiled a month ago.

I think making these type of games simply becomes an unviable source of income in the current era.

I'm sure there's a named concept for this: the idea of the game is its selling point, and has to be revealed to consumers to market itself. But the market has low friction (via Steam distribution, and by copies being cheap to make w/ LLMs), so you can't reliably establish your product before masses take over, especially if you don't have reputation built up from previous games.

You have to be bringing in something else for this not to happen, trade secrets not present to the consumer being one such factor.

I feel like this is on a different trajectory than what LLM-based tech does. This type of individualistic extrapolation is one of the things they're really bad at, in my experience.

I am genuinely worried about violence, especially if people start losing their jobs due to automation.

I think it's quite possible for a reversal to happen, where nerds will start meeting IRL while everybody else flocks to web spaces.

Wow, thank you for sharing! It's not quite clear to me why number of projects seem to be oscillating every few years.

It's the same as before, dev tools ranking higher up. For better or worse, developers like to spend time on meta-development.

Could we compare this to HN growth?

My understanding was that the Ad business was starting to suffer because of how much bot activity there's been both in social media and indexed search (Google).

But I wouldn't write off ads yet. It seems plausible to me that LLMs will be able to slightly alter their output to casually mention paying brands and to profile users better than we could before.

Is it wrong to presume they tried to run a similar prompt on all ciphers that come before this one in search results, and this was the only one that worked?

I'm not seeing the same kind of resentment towards non-technologists as you are claiming. Overall in my life I feel like I haven't seen as much gatekeeping as people claim there is in tech.