ip26 · HackerTrans

Is public perception the greater problem, or delivery? My outsider read has been relatively few diseases can be targeted right now due to payload delivery obstacles

ip26

Hôm kiadiscuss

Phase 3 seems closer to doing a soft production launch and finding customers completely reject it despite early focus groups liking it.

Prototypes, when they fail, are also often tweaked into something that works. A lot of phase 3 failures appear to have been complete dead-ends.

ip26

4 ngày trướcdiscuss

It wasn’t long ago at all that the chief problem was “can AI even help me with this” (cost be damned). Until a time when the answer is an unmitigated “yes, obviously”, the frontier labs have everything to lose and nothing to gain on routing, because if they screw it up you might incorrectly decide “nope, it can’t help yet” due to a poor routing decision.

ip26

6 ngày trướcdiscuss

The alternative to the rationalist one is “be suspicious of anyone who claims to know the unarguable truth”

ip26

8 ngày trướcdiscuss

It's like modeling a living cell in a computer building upon the basic maths of physics. Yes, it is a thing that is "possible". However it is not practicable to capture everything that matters and run that simulation.

ip26

13 ngày trướcdiscuss

The “affordable” housing thing seems like such a misdirection. I can’t help but daydream that some moneyed interest somewhere fans those flames, as it looks like a dead end that can absorb endless fervor.

You know what you do if you want an affordable car? You buy used. I think most people understand Ford is never going to build another a car that costs $10k brand new, and the last new car near that price barely sold because it was so stripped.

ip26

16 ngày trướcdiscuss

If we’re just spitballing, it certainly could be that, but it’s also obvious that publishing papers can help communicate your existing work. AI can definitely help at that, and even great work needs to be communicated.

If this same article had said “adoption of word processors over typewriters increases publishing”, would we still be thinking of the babble hypothesis?

ip26

17 ngày trướcdiscuss

You describe it like it’s some kind of a treadmill chasing new. But if you have kids at 35, they start school when you’re 40.. were you thinking about the quality of the neighborhood schools when you bought your house at 25?

Or suppose you meet your spouse when you are 30, after you bought a house.

There’s inherently much less moving-around if you get married at 18 and have kids straight away - the plan is settled from the start.

ip26

18 ngày trướcdiscuss

My assumption is somewhere in the harness toolchain there's a prompt inject that says:

    - Always sign commit messages

In which case, Claude is being given conflicting instructions, which is a different class of problem than too many instructions.

ip26

25 ngày trướcdiscuss

The basic purpose of patents is to make it worthwhile to develop something new and valuable. They are unnatural - but useful. Working backwards, if we want people to spend time and money developing new breeds of plants, you arrive at these kinds of rules.

ip26

tháng trướcdiscuss

[dead]

ip26

tháng trướcdiscuss

Regardless his actual private beliefs, one of his "jobs" is to create new unicorns ycombinator can invest in. So evangelizing startups in general and spurring new entrepreneurs is very much in his interest, and spreading this kind of mythology seems like a useful step to that end. He only needs a small number of people to buy into it.

ip26

tháng trướcdiscuss

You don't need to project to the heat death of the universe to get Facebook without Zuckerberg.

ip26

tháng trướcdiscuss

Parallel invention would like a word. Elisha Gray registered a patent for a telephone the same day as Alexander.

It's impossible to fully prove a counterfactual, but few things "wouldn't exist at all" if "that one person" hadn't done it.

Netflix is a decent example. Many people saw the coming of video streaming. We would still be able to stream videos today even if Hastings had stayed at Rational.

ip26

2 tháng trướcdiscuss

Infinite tokens rate-limited to 10 tok/s is 26MTok per month.

ip26

2 tháng trướcdiscuss

A never had the authority to contract the land as a park indefinitely ... once C owns it they own it

That's not what a deed restriction is.

are you stuck with passing that obligation forward

Generally yes. Which is why the deed restriction can affect the market value of the property.

ip26

2 tháng trướcdiscuss

Hard to assign that a value before you are a parent though. You can only understand it in the abstract.

ip26

2 tháng trướcdiscuss

I probably interpret the message incorrectly, because to me "the hardest thing" is usually some class of some famous unsolved problem. The thing is, to be famously unsolved, many of the world's most brilliant people must have already tried and failed to solve it. There's a chance I have an edge they don't, but it's probably not wise to track your career path to the darkest of dead ends purely on the merits of it being particularly dark.

Example: you know what would be harder than anything we're talking about here? Quadrupling the performance of the best compression algorithms. It's hard. In fact, maybe there's some information theory that even says it's mathematically impossible. That makes it really hard, which makes it what all of us should immediately start working on.

The author writes elsewhere that you also have to have an edge, but that's frequently omitted from this "hardest thing" advice.

ip26

2 tháng trướcdiscuss

You’d think Berkshire would be at least passingly aware of that principle, though.

ip26

2 tháng trướcdiscuss

Depends on the quality of your validation loop. Can the agent find the bug in a five second unit test, or does it have to run the full deployment test?

It also presents tradeoffs in compute budget. Cycles spent executing large arrays of tests could mean less tokens spent debugging.

comments