1,823 karma · joined May 22, 2020
Nobody, and I mean nobody beats physicians at the Arrogance game.
The very second that ML models begin to consistently beat humans at diagnostics, it would be a moral imperative, as well as an absolute requirement by the terms of the Hippocratic Oath, to replace the humans.
Introverts and extroverts will never understand each other. This entire thread is pointless.
Fortunately, with function-calling (and recently, with guaranteed data structure), we've had access to application interoperability with LLMs for a while now.
Don't get mad at a company for developing for the masses - that's what they are expected to do.
But they built for us, first.
Aider (with the older Claude models) is already a semi-competent junior developer, and it will produce 1kloc of decent code for the equivalent of 50 cents in API costs.
Sure, you still have to review the commits, but you have to do that anyway with human junior developers.
Anthropic could charge 20x more and we would still be happy to pay it.
But eventually they (RNNs, likely) will. And we won't know when.
This can't possibly be a serious solution. A quart of acetone costs $2.
The metric is useless in the first place, this is "per licensed drivers", not "per driver-miles".
It's a certainty that men drive more miles than women, primarily for job purposes. The NHTSA doesn't give enough data to determine this. My guess is that it's split almost exactly evenly per-mile, but who knows, without the data?
I suspect that insofar as this exists, it's because when junior engineers question the utility of the requirements or user stories, they are told specifically to stfu because nerds don't understand business. Over time, the message get received?
That said, I've very rarely seen an engineer like this. In fact, I frequently encounter the opposite: the sterotypical asperger's who doesn't know when they're being rude with their probing.
This is an incredible take, and I have no idea where you have seen this, as it's entirely counter to my own personal experience. I have never, in my life, met a finance bro or management bro who had any non-work curiosity about anything beyond sports, cars, and sex.
Most engineers I know are immensely curious about how the world works, how the universe works, etc, and are constantly trying to learn and understand.
Anthropic takes safety to mean "let's not teach people how to build thermite bombs, engineer grey goo nanobots, or genome-targeted viruses", which is the traditional futurist concern with AI safety.
OpenAI and Google safety teams are far more concerned with revising history, protecting egos, and coddling the precious feelings of their users. As long as no fee-fees are hurt, it's full speed ahead to paperclip maximization.
Did I miss a memo? This is one of the largest [citation needed] I've seen on this site in some time. Did he kick a puppy?
- If you're scared to perform upgrades, this is a good indicator that you lack confidence in your backup/recovery/reversion policies and procedures, probably for good reason, so maybe fix that.
AMD is not taking ML applications seriously, outside of their marketing hype.
Local and state taxes generally actually help you. Your local infrastructure is maintained, and your kids go to schools funded by your taxes.
Federal taxes, however, are largely transfer payments from the productive to the unproductive, as well as funding wars in areas that have absolutely nothing to do with you.
It's pretty easy to make the distinction between "good" and "bad" taxes.
Generally untrue, since most rich people hold their assets in what's being sold. There are a handful value investors left, who bother holding cash equivalents when PE ratios get absurd, but they are few and far between. Tech billionaires, in particular, are very unlikely to be sitting on much cash.
The entire dental industry is only in business because of sugar and ignorance. Fortunately for them, the overall vanity level of society is increasing, so they'll likely make up for it with whitening sessions and veneers.
It probably is, actually (and sadly).
Looking around at the percentage of drivers around me on the road with their face looking at a phone, while moving, is probably in excess of 5%. These people won't see the kids crossing the road, either.
The left is very clearly attempting to curtail the right to free speech, free association, and economic freedom.
The right is very clearly attempting to curtail the reproductive rights of women, and freedom of movement across borders.
Sure, if you're inured to the "my side is angelic, and the other side is evil", it's easy to tell. But stepping outside of that, it's pretty clear that both sides are gunning for their own specific versions of tyranny.
Sure, if you speak to actual carnivores, they will make it clear that fat, offal, sinews, marrow, etc are all critical parts of their diet, but if you just employ selective redefinition that has no bearing on actual reality, the argument stands strong in your own mental vacuum.
I remember doing the same thing as a thirteen year old at local Lincoln-Douglas debate tournaments. Winning debates by stretching dictionaries is a solid move, even if it doesn't get you very far at the nonlocal level.
It's always critically important that our Straw Men are constructed perfectly, to avoid pesky and annoying limitations of the real world.
A lot of people here have fallen for the cloud marketing efforts. Nobody is saving any time.
It takes 5 minutes to set up a Docker Swarm Mode cluster. It takes maybe 15 minutes for k3s or microk8s. After that, auto-scaling is dead simple, and no MORE complex than some shitty vendor locked-in cloud solution.
> that needs to have both a fail-over node and backup solution
ZFS pool replication, Ceph, GlusterFS, etc. Lots of options here. These are long-solved problems.
> A disaster can be as simple as a failed disk
Right, which is why you design your on-prem cluster with N+2 redundancy in the first place, and with a locked cabinet with spare parts. Cattle not pets, and all that. Do you think your EBS storage never fails? You'd need to do exactly the same thing in the cloud, anyway.
> but spending 55k/yr for a sysadmin on-prem to maintain
First, if you're paying only 55k for sysadmins, you should be planning to fail anyway. Competence is compensated quite a bit north of there.
Second, assuming the context is small business, you're going to have role crossover anyway, it's inevitable - chances are that your developer(s) is(are) administering this. Not every business is Facebook.
I've had far less downtime (read: zero) from my custom power redundancy solutions on my homelab than from power issues with my colo provider.
Small companies would be far better off with local power redundancy than some gargantuan complex industrial Liebert solution with a 4 hour SLA.
My personal experience, as someone who has lived in, and worked from, rural areas with limited bandwidth, is that latency (for SSH connections) is the only thing that matters for learning and productivity.
But OP clearly knows better, because if we just gave everyone gigabit fiber, the access to UHD Pornhub, Netflix, Amazon Video, etc, will instantly correct the "digital divide". And OP has a point. I know someone who started designing > 500k qubit quantum computers with > 5s coherence after spending two weeks straight watching all seasons of My 600Lb Life.
He kept mumbling something about "It's not in the box, it's in the band"
Sure, 7mbps may not be good enough to supply the demands of your multi-screen 4k UHD loli goon cave, but it's more than enough to send a < 1kB message to your family that you're safe in a disaster area.
Try to get some perspective.
Hire a team to build a project, when it's finally satisfied most of your requirements, you progressively cut staff until only Jim is left, and Jim spends the next two decades maintaining the system, growing out his hair and beard, piling kludge on top of kludge, and drinking heavily until retirement.
The HN bubble, focused so intently on BigTech and FAANG, is woefully unaware of how things work basically everywhere else.
Greenlighting training a new foundation model is very expensive, but is also a human decision that can be postponed based on available capital.
The market demands models that don't fail constantly with HAL-like "Sorry, Dave, I cannot do that for you" moralizing responses.
An API that is used for mission-critical purposes and that randomly fails with "HTTP/1.1 406 You Are A Terrorist And/Or Hitler" is a BUG, and the market will coalesce around models that don't have this bug.