ctoth

20 ms·

421.
by ctoth 1y ago
Who decides what technologies are too dangerous? You, apparently. AI isn't nukes - anyone can train a model at home. There's no centralized thing to restrict. So what's your actual ask? That nobody ever trains a model? That w
422.
by ctoth 1y ago
Wait, you're literally advocating for handicapping everyone because some people can't handle the tools as well as others. "The disciplined minority can use AI well, but the lazy majority can't, so nobody gets to use it&q
423.
by ctoth 1y ago
And the billions of years of evolution and the language that you use to explain the task to him and and the schooling he needs to understand what you're saying it and... and and and?
424.
by ctoth 1y ago
One thing I keep not understanding is I see colleagues or AIs put f-strings in logger calls! Doesn't the logger do lazy interpolation? Doesn't this lose that nice feature?
425.
by ctoth 1y ago
I saw the phrase "goalposts aren't just moving, they're doing parkour" recently and I do love that image. It does seem to capture the state of things quite well.
426.
by ctoth 1y ago
We've hit the limit of 'our current training techniques'? This result literally used newly developed techniques that surprised researchers at OpenAI. Noam Brown: 'This result is brand new, using recently developed techni
427.
by ctoth 1y ago
Pre-registering a prediction: When (not if) AI does make a major scientific discovery, we'll hear "well it's not really thinking, it just processed all human knowledge and found patterns we missed - that's basically chea
428.
by ctoth 1y ago
Obviously this is clearly and demonstrably because those courts are odd.
429.
by ctoth 1y ago
One way to not be in the dangerous situation of being a hated goon is to avoid being a hated goon. Hated goon advice there.
430.
by ctoth 1y ago
Modafinil, not Adderall.
431.
by ctoth 1y ago
I asked my contractor to install a door over the back stairs opening outward, came back, and it was installed opening inward. He told me he tried to figure out a way he could do it in-code, but there wasn't one, so that's what he
432.
by ctoth 1y ago
Oh do let me tell you how much effort I put into tending my non-deterministic garden or relationships or hell even the contractors I am using to renovate my house! A few small markdown documents and putting in the time to understand somethi
433.
by ctoth 1y ago
Sometimes, you'll just have a really productive session with Claude Code doing a specific thing that maybe you need to do a lot of. One trick I have gotten some milage out of was this: have Claude Code research Slash commands, then mak
434.
by ctoth 1y ago
> In simpler terms, CCusage is a relatively-accurate barometer of how much you are costing Anthropic at any given time, with the understanding that its costs may (we truly have no idea) be lower than the API prices they charge, though I
435.
by ctoth 1y ago
The concept you are searching for is CoT faithfulness, and there are lots and lots of open questions around it! It's very interesting!
436.
by ctoth 1y ago
The paper isn't proposing "monitor chain of thought for safety" - that's already standard practice. The core finding is that training on CoT monitors destroys them. They provide empirical evidence that doing the obvious
437.
by ctoth 1y ago
Might it be as simple as: before, the LCD viewer who didn't get it had no platform, now they do? Responsiveness to Rotten Tomatoes instead of Roger Ebert?
438.
Beyond "Abolish the FDA"(astralcodexten.com)
12 points by ctoth 1y ago|1 comments
439.
by ctoth 1y ago
> It can be both, but there's usually no place for differentiation Fool me once, shame on you. Fool me 153,927,861 times, shame on me. The place for differentiation, the place for "oh this is probably fine", the benefit of
440.
by ctoth 1y ago
I had some very technical friends be incredibly surprised by the Edge form thing, I think that is not sufficiently called out! They send any text you type in a form to their AI cloud and hold on to it for 30 days. Any form. On any website.
441.
by ctoth 1y ago
Frankly I don't care if we figure it out after I'm dead, and I'm not sure why you would either. I want to know now
442.
by ctoth 1y ago
Weird. Back in 2019 were all the coders just better? Never hardcoded keys?
443.
by ctoth 1y ago
The formula is so consistent: cherry-pick some disappointments, interpret any pivot as retreat, claim the hype is unsustainable, predict imminent collapse, position yourself as the sober realist. Rinse and repeat every 6-12 months, adjustin
444.
by ctoth 1y ago
What is the reason you wrote your comment? Like, if life were a narrative, who would you be?
445.
by ctoth 1y ago
I'm just a regular developer - not rich, not running a company. I used to maintain accessibility apps for blind users (QRead for ebooks, various Twitter clients back when their site was barely screen-reader compatible). Had to abandon
446.
by ctoth 1y ago
Reinforcement learning from self-play/AlphaWhatever? Nah must just be datasets. :)
447.
by ctoth 1y ago
The other kind of ChatGPT psychosis?
448.
by ctoth 1y ago
The laws of robotics were literally designed to cause conflict and facilitate strife in a fictional setting--I certainly hope no real goddamn system is built like that,. > To ensure robots behave safely, Gemini Robotics uses a multi-laye
449.
Putting Brakes on the Singularity(maxmore.substack.com)
2 points by ctoth 1y ago|0 comments
450.
Using AI: A Quick Guide(oneusefulthing.org)
1 points by ctoth 1y ago|0 comments

More ›