A'ight.
-
@AVincentInSpace @munin "i'm sure we can improve the accuracy only after we've gained enough developer mindshare"
-
@AVincentInSpace @munin "here you can see our incredibly-rapid DAU growth, and here's another chart showing our industry-leading estimation error per unit of compute"
-
@AVincentInSpace @munin "here you can see our incredibly-rapid DAU growth, and here's another chart showing our industry-leading estimation error per unit of compute"
@AVincentInSpace @munin not sure how many VCs can be bamboozled this way, but it certainly feels like "more than one"
-
If you're one of the people who knows why this hasn't been done before, let the vibecoders figure it out for themselves.
@munin@infosec.exchange someone should halt them before they find the keys to the kingdom
-
A'ight. The whole vibecoding thing deserves a chance to prove itself.
So here's something that I haven't seen done sufficiently well by regular coders; if AI is truly that much more innovative, it shouldn't have any problem.
One of the problems with compute is the whole billing and scheduling thing - a lot of places have specific cost-per-hour to run batch processing; a lot of large enterprises have complex pipelines that need scheduling in order to interleave things that need processing with resources available to process them.
So a vibe coder who's confident they can prove themselves could create a utility that can look at a given program's binary, analyze it, and determine how long it will take to run, and calculate the cost to run it. Do this within 1% of actual and you'll have a truly innovative new product.
@munin this sounds like a great idea and I think that vibecoders should definitely look into it
-
@munin this sounds like a great idea and I think that vibecoders should definitely look into it
That's why I posted it, yes. It's a perfect problem for them to really, comprehensively prove the superiority of their approach - nice and legible, obvious utility, and a clear measure of success.
And it's truly innovative; no other product on the market is capable of this.
-
A'ight. The whole vibecoding thing deserves a chance to prove itself.
So here's something that I haven't seen done sufficiently well by regular coders; if AI is truly that much more innovative, it shouldn't have any problem.
One of the problems with compute is the whole billing and scheduling thing - a lot of places have specific cost-per-hour to run batch processing; a lot of large enterprises have complex pipelines that need scheduling in order to interleave things that need processing with resources available to process them.
So a vibe coder who's confident they can prove themselves could create a utility that can look at a given program's binary, analyze it, and determine how long it will take to run, and calculate the cost to run it. Do this within 1% of actual and you'll have a truly innovative new product.
@munin simply append sleep(1000000000) and predict the program will finish within about 11.6 days
-
That's why I posted it, yes. It's a perfect problem for them to really, comprehensively prove the superiority of their approach - nice and legible, obvious utility, and a clear measure of success.
And it's truly innovative; no other product on the market is capable of this.
@munin @wuest If you want an answer that will usually be right for typical things people run as batch jobs (as opposed to for every possible program), the problem I think you're trying to point at without naming isn't really relevant. There's lots of difficulty there caused by randomness coming from timing of events across different machines coupled with design of batch pipelines that often amplify changes in running time[1].
[1] For example, a version of MapReduce would have shufflers query data directly from mappers' memory, unless they fell far enough behind so that this data was no longer stored in a cyclic buffer there, in which case they would read it from some disk-backed scratch storage. Obviously, once a shuffler fell that far behind it would never catch up again. -
@AVincentInSpace @r @munin Well, since I just found that an ex-DEA agent raised $10 million in VC money with provably false background and achievements in conjunction with software that I wrote and maintained on Github for free. The falsehoods can be proven because they do not match up with either OIG's audit on the very issue at the exact time she claimed to have accomplished things that did not occur, cross-referenced with Courtlistener's PACER database. In fact the services advertised on her site would be obtained in a manner that violates the 4th Amendment and kill any case it's involved in. Luckily, there's no docket that indicates that her startup or her was ever involved in any legal proceedings, which was only a handful, and I had RSS feeds on all of them and paid to populate the Courtlistener DB. All this came about from a Bloomberg article that was so outlandishly false that it was obviously a PR plant to rehab ICE-HSI and the DEA's reputation, which is a widely known practice. Except most people don't know this, except I practiced criminal defense and dealt with computer crimes and the DEA a plenty. and so, can do my due diligence between ordering Doordash and the food showing up.
So basically this is, well, potential fraud, which due diligence could have uncovered during a lunch break, but instead have netting ten million dollars of investment into services that literally exist for free and relies on a general lack of technical expertise in criminal law practitioners. Why vibe code when you can just make stuff up, right?
-
R relay@relay.infosec.exchange shared this topic
-
A'ight. The whole vibecoding thing deserves a chance to prove itself.
So here's something that I haven't seen done sufficiently well by regular coders; if AI is truly that much more innovative, it shouldn't have any problem.
One of the problems with compute is the whole billing and scheduling thing - a lot of places have specific cost-per-hour to run batch processing; a lot of large enterprises have complex pipelines that need scheduling in order to interleave things that need processing with resources available to process them.
So a vibe coder who's confident they can prove themselves could create a utility that can look at a given program's binary, analyze it, and determine how long it will take to run, and calculate the cost to run it. Do this within 1% of actual and you'll have a truly innovative new product.
@munin had me in the first half, won't lie
-
A'ight. The whole vibecoding thing deserves a chance to prove itself.
So here's something that I haven't seen done sufficiently well by regular coders; if AI is truly that much more innovative, it shouldn't have any problem.
One of the problems with compute is the whole billing and scheduling thing - a lot of places have specific cost-per-hour to run batch processing; a lot of large enterprises have complex pipelines that need scheduling in order to interleave things that need processing with resources available to process them.
So a vibe coder who's confident they can prove themselves could create a utility that can look at a given program's binary, analyze it, and determine how long it will take to run, and calculate the cost to run it. Do this within 1% of actual and you'll have a truly innovative new product.
@munin@infosec.exchange Wow, nice one.
