vidarh

27 ms·

721.
by vidarh 7mo ago
I'm surprised it's only 2-3x times quicker w/Truffle? Is that because it only encodes/decodes a single image at the time and incurs higher startup costs? Or do you mean 2-3x vs. an MRI alternative that calls into a nativ
722.
by vidarh 7mo ago
I exchanged some e-mails with him because I was writing a series on writing a Ruby compiler in Ruby at the time (now self-hosting but woefully buggy and incomplete still; I pick it up now and again, but it's very much a slow-burn hobby
723.
by vidarh 7mo ago
Isaac Asimov vote a huge amount of popular science book as well. They just have a shorter shelf life.
724.
by vidarh 7mo ago
A lazy copy has of course always been lame, but it also depends on when in the 90s, and the platform. There were plenty of images that amazed me back then even though I was perfectly aware of the source material. It depends on the platfor
725.
by vidarh 7mo ago
Copyright law exists exactly because it is universally accepted that ideas are not property: Copying an idea or expression of it does not deprive you of your ideas. The entire notion of "intellectual property" is the creation of a
726.
by vidarh 7mo ago
Even if you live somewhere where it does, that is not remotely "almost free", and lots of places the payback period is more in the range of 10-15 years even with subsidies.
727.
by vidarh 7mo ago
The amortised cost including the panels and labour is nowhere near "almost free".
728.
by vidarh 7mo ago
You'll only lose the skill of critical thought if you don't spend the time it frees up solving other problems. This is the lazy part of the argument: The assumption that people stop applying themselves when they get a tool to repl
729.
by vidarh 7mo ago
We are not automating humans. We are automating some things that have previously required humans to carry out, just like vast numbers of things before.
730.
by vidarh 7mo ago
Open models are available that while they seem primitive to current frontier models lag only 1-2 years behind.
731.
by vidarh 7mo ago
> This is clearly not like any other tool How is it not? How is this not yet another lazy argument. I'm old enough to have lived through multiple "this is the end of civilization" waves that people insisted were different.
732.
by vidarh 7mo ago
This is only a problem if regulators and/or courts and/or consumers all fail to recognise that said 99% car isn't safe enough.
733.
by vidarh 7mo ago
More realistically, a company that fails to properly evaluate this in ways that reflect actual market needs will fail in the marketplace.
734.
by vidarh 7mo ago
So vibe coding won't be sufficient to replace a skilled Java developer, and won't obsolete that skill, and if there aren't alternatives that more completely replaces a skilled Java developer, this then isn't a relevant
735.
by vidarh 7mo ago
All of human history.
736.
by vidarh 7mo ago
This is such a lazy argument. Every tool that displaces old tools causes skills to be lost when those skills are no longer needed. To the extent that people still need to be able to critically assess what AI delivers to achieve their goals,
737.
by vidarh 7mo ago
> Perhaps these things aren't well represented in the training data for these open models I know from first-hand experience that at least a couple of the SOTA providers use third-party providers for supervised finetuning with instru
738.
by vidarh 7mo ago
While I like these models, if you're getting similar results to SOTA models from 6 months ago, I have to question how far you pushed those models 6 months ago. It is really easy to find scenarios were these models really underperform
739.
by vidarh 7mo ago
There are desk attachments that will allow your cat to lord it over you from on high.
740.
by vidarh 7mo ago
Putting "cat desk" into Amazon didn't show any custom desks , but did come up with a wide range of desk attachments for cats, as well as "cat laptop" (a scratching board shaped like a laptop) so this doesn't
741.
by vidarh 7mo ago
I get decent results with Kimi, but I agree with your overall premise. You do need to realise that while you can save money on a lot of tasks with those models, for the hardest tasks the "sticker price" of cost per million tokens
742.
by vidarh 7mo ago
Not my experience at all. If anything, they make it cheap enough to deal with tech debt that it is far easier to justify being strict. EDIT: I'll add that you can't expect it to guess what you want , but you can let it manage ho
743.
by vidarh 7mo ago
Yes, they can, and they do a reasonably good job at it. Hand them playwright or similar, and point them at it. The caveat is that they're often "lazy", and it takes some practice to coax them into being thorough (hot tip: hav
744.
by vidarh 7mo ago
I prefer to have my agents review my agents output and progress, and have them improve the prompts for future runs.
745.
by vidarh 7mo ago
It's generally not worth it worrying about it too much other than at a very high level vs. letting them fight it out, as long as your test suite is good enough and your orchestrator is even moderately prepared to handle retries.
746.
by vidarh 7mo ago
I've come to the opposite conclusions: The big limitation of systems like this is starting and ending with human involvement at the same level, instead of directing at a higher level. You end up quibbling over detail the agents can han
747.
by vidarh 7mo ago
Maybe 5% of what I use Ruby for is on the server. I'd suggest those of us who use Ruby client side are likely to outnumber Haiku users by magnitude or two.
748.
by vidarh 7mo ago
They have several, but the others won't show unless you have showdead turned on, as they've already been flagged.
749.
by vidarh 7mo ago
Branding/marketing monitoring companies would be all over this.
750.
by vidarh 7mo ago
The problem is that it is still to make vague promises of future income to buy them. E.g. speaking fees.

More ›