skissane

19 ms·

331.
by skissane 10mo ago
Western AIs are trained to defend the “party line” on certain topics too. It is even possible that the damage to general reasoning ability is worse for Western models, because the CCP’s most “sensitive” topics are rather geographically and
332.
by skissane 10mo ago
I’ve seen people go with IBM Cloud because their salespeople were willing to discount more heavily than AWS/GCP/Azure were. Tier 2 players can be hungrier for your business than tier 1 are. And here I’m talking about completely ma
333.
by skissane 10mo ago
But, did you actually give the agent access to a tool to measure code coverage? If it can't measure whether it is succeeding in increasing code coverage, no wonder it doesn't do that great a job in increasing it. Also, it can help
334.
by skissane 10mo ago
Also the UK gallon is different from the US gallon. And the same applies to all the other non-metric fluid measurements such as pints and fluid ounces. Historically the UK gallon was used throughout the former British Empire (Australia, Can
335.
by skissane 10mo ago
Models have some limited means of refinement available to themselves already: augment a model with any form of external memory, and it can learn by writing to its memory and then reading relevant parts of that accumulated knowledge back in
336.
by skissane 10mo ago
When we think, our thoughts are composed of both nonverbal cognitive processes (we have access to their outputs, but generally lack introspective awareness of their inner workings), and verbalised thoughts (whether the “voice in your head”
337.
by skissane 10mo ago
> For LLMs, there is an added benefit. If you can formally specify what you want, you can make that specification your entire program. Then have an LLM driven compiler produce a provably correct implementation. This is a novel programmin
338.
by skissane 10mo ago
Give a coding agent a codebase with no tests, and tell it to write some, it will - if you don’t tell it which framework to use, it will just pick one. No denying you’ll get much better results if an experienced developer provides it with so
339.
by skissane 10mo ago
Give coding agent some software. Ask it to write tests that maximise code coverage (source coverage if you have source code; if not, binary coverage). Consider using concolic fuzzing. Then give another agent the generated test suite, and as
340.
by skissane 11mo ago
You may be joking but I think seriously that may be a decent idea. The universal RPC/automation mechanism which everybody implements-whatever its warts may be-is superior to a better-designed one which few actually end up supporting. Y
341.
by skissane 11mo ago
It needn’t be a “free lunch” at all. An O(n^1000) algorithm for an NP-complete problem would constructively prove that P=NP yet be completely useless for solving any NP problems in practice
342.
by skissane 11mo ago
It was still there up to Windows XP, although by XP it had become a legacy component lacking support for newer clipboard formats. Vista removed it. The NT/2000/XP version supported sharing clipboard data across your local network
343.
by skissane 11mo ago
> A new "math" might be needed, but an answer (affirming or not) will be found. What if there exists a proof that P!=NP, but the shortest possible proof of that proposition is a googolplex symbols that long? Then P!=NP would be
344.
by skissane 11mo ago
Someone might eventually provide a non-constructive proof that P=NP - a proof that such an algorithm must exist but which fails to actually produce one. Or even a galactic algorithm-an algorithm for solving an NP-complete problem that is te
345.
by skissane 11mo ago
Couldn’t you equally say “The fact that thousands of people have failed to prove that P!=NP indication that it is probably not true”? My completely unscientific hunch is someone will eventually prove that P=?=NP is independent of ZF(C). May
346.
by skissane 11mo ago
> ServiceNow is so terrible I genuinely wonder how it is ever deployed anywhere. Seriously, do the purchasers never look at it? When I first saw ServiceNow, I was impressed - because my point of comparison (I worked for a university at t
347.
by skissane 11mo ago
If we look at afterlife beliefs-and their secular substitutes such as life extension, cryonics, mind uploading, simulationism, quantum immortality-I don’t think they all have the same motivation-two people may adopt the same belief with dif
348.
by skissane 11mo ago
No one has ever done it is not in itself evidence it can’t be done, only that it is hard. And my point was-no matter how hard human life extension is, mind uploading is many orders of magnitude harder. The first, it seems likely in principl
349.
by skissane 11mo ago
We know other species have different maximum lifespans-some shorter, some longer. Obviously this is determined by genetics-as our knowledge of that subject continues to improve, why wouldn’t we eventually work out how to alter it? We can al
350.
by skissane 11mo ago
Life extension to make people live several times longer? Seems plausible we’ll get there eventually, simply by extrapolating current trends in science and biotechnology, and observing what is possible in other species. Mind uploading? That’
351.
by skissane 11mo ago
Life extension research isn’t going to make anyone immortal - it can’t prevent deaths from accidents or foul play, and after a few thousand years the odds you will succumb to one or the other becomes quite high. Suicide is likely to be anot
352.
by skissane 11mo ago
The article claims to prove no computer could accurately simulate quantum gravity. Suppose they are right, and as a result our simulators are forced to make quantum gravity experiments (if that were a thing) give “incorrect” answers, becaus
353.
by skissane 11mo ago
> One problem that I have with trying to understand "time" is that we can't measure how quickly it "flows" or at least how quickly we travel through it. Our experience of time passing is heavily influenced by the
354.
by skissane 11mo ago
> Existing for zero time would imply it never existed. In the mathematics of infinity, something can have exactly zero probability yet still happen, and exactly one probability yet nonetheless fail to happen (hence the standard term “alm
355.
by skissane 11mo ago
The Chinese Communist Party is obsessed with social stability. Do you think they’ll allow AI to take all the jobs, destroying China’s domestic economy in the process? Or will they enact human protectionism regulations? What Would Xi Jinping
356.
by skissane 11mo ago
B2C businesses need consumers. If AIs take all the jobs, then most of the population-minus the small minority who are independently wealthy and can live off their investments-go broke, and can’t afford to buy anything any more. Then all the
357.
by skissane 11mo ago
Addendum: from what I understand, the article’s proof relies on computational quantum gravity having a Gödel sentence. Now, quantum gravity is in practice, as far as we know, experimentally untestable-the distinctive phenomena it predicts o
358.
by skissane 11mo ago
Does the article propose anything empirically testable? I mean, suppose we are actually in a computer simulation-what observations could we perform, which according to the rules of this article, would show that we were in one, and not the “
359.
by skissane 11mo ago
> The idea in the article would refute the inductive step. No it doesn't. The article describes a proof that it is impossible for a computer to simulate this physical universe with perfect accuracy; but, that's not actually a p
360.
by skissane 11mo ago
Someone I know is a high school English teacher (being vague because I don’t want to cause them trouble or embarrassment). They told me they were asking ChatGPT to tell them whether their students’ creative writing assignments were AI-gener

More ›