Mainly: true sandbox separation. I don't want the model having full access to my machine. With a dump format that Claude understands, I'm able to pass only the files I want Claude to see, and he can't break any of them. I don't care about setting up access lists and so on. I don't trust that the cli product will be properly sanboxed and it's quite clear their software offerings are largely aigen code, and I catch bugs from Claude every day. I also get useful stuff, so it's worth it, but definitely not worth it, imo, to grant it any access to my machine.
I've yet to see any masterpieces. I've had a lot of good enough code come out, and some that was blatantly wrong. You still need to be an expert in the domain if you want to use ai codegen effectively. I've been using Claude for a few months now.
cool idea - i can't tell from the page if there's any choice of model (though the name and voices suggest gpt only); neat idea though. Perhaps this is the solution: devs should have a device like this, and pedal to power it. Solves the power problem, also solves the typical "devs need exercise" problem.
and I just upvoted a comment, refreshed, and it's still upvoted - perhaps check your network stack in your browser? Sounds like something isn't getting through and the UI is unaware.
and then, after 6 months with a huge codebase that doesn't work, don't forget to hire a real human to rewrite it - you, personally, don't have to write the code, but someone will have to turn that ball of mud into a product that can be extended.
Recognising this has also made me recognise the real problem of using an LLM without expert knowledge in the domain - if you don't know when it's wrong, you won't know when to push back. So whilst I can be very critical of code outputs, for example, my knowledge of quantum mechanics is so rudimentary that it's practically impossible to push back on conversations about QM. So who knows if my hobby knowledge actually holds any value? The best I can do here, when repeating the information, is to say "at least, that's what the machine says".
I agree largely, but hold up a bit on the statement that getting an LLM to write the code means I never solved the problem. That may be true for a lot of vibe-coders - loops of "write this code" "no, not that", "fix plz", etc. But what I've found is that Claude (at least) does best on small problems that are very well defined - in other words, I have to solve the problem _before_ I can get Claude to produce the code for me, if for no other reason than that the model will model along, and needs an expert on the problem to be able to push back when it's wrong.
So I agree that leaving an LLM churning for a week or two and then claiming that you have a product to sell is tenuous, but I disagree that one can't both use an LLM _and_ understand the solution - it just takes active participation towards that result.
Whilst there's no remuneration involved, I view projects I put up on github (that I also tell other people about) in the same light - so, for example, the largely-ai-written https://github.com/fluffynuts/nestray, whilst designed to scratch an itch of mine, may be useful for others because I've aimed at some level of genericism. Is it a perfect product? NO! But it does what I want, and probably what someone else might want with respect to showing email notifications without requiring any account access (simply reads from local Thunderbird files)
you could enable prompts on download - and set the filename yourself (or just rename after download). Ultimately, naming it yourself is the only way you're going to get names you're happy with.
And, for an encore - stop all the other stupid shit. The rest of the world (and the US) is paying the price for little trump-tantrums, like the one against Iran. He's not a good international leader. He's not even a reasonable at-home president.
functionally the same - and more accurate to use the original title, as Cuba is the one doing the blaming. I don't know why you're standing up for this - it's more bad behavior from a country that sells itself as the savior, and it's not new - they've been doing this (whatever they need to, to change regimes) in other countries for decades. It's shameless bullying, and completely contravenes "the rules" about how to interact with other countries.
1. the logs hopefully tell you where to look
2. the logs hopefully give you enough context to construct a physical test (ie, replicate)
3. the physical test hopefully gives enough context to write an automated test
4. we write the automated test, submit to staging, do the physical test again
5. PROFIT!
either:
1. you spend all the effort and time (and money - it's not free to drive around, and there are often join fees) to switch banks and it makes no difference whatsoever, or
2. the overall energy consumption remains the same, and everyone levels out, because that's just how much energy banks require per customer
I'm all for saving the earth but I don't see how making my life more difficult will change this situation one iota - meanwhile, I have limited time to live, limited resources of my own.
Good distros should backport updates like this - the patch has been merged into several downstreams. Whilst I prefer a rolling release, this isn't the reason to do it. You could even be on a rolling release that is behind (like arch - my Gentoo box was patched for this vuln before I even understood what it was, and I could run the exploit on my arch machine days laterz until eventually the update came). Speed of response by the upstream here is more important than anything else.
the value of the _US_ has weakened significantly in the eyes of the rest of the world due to the great orange muppet - if y'all don't like it, get rid of the useless skinflap.
spoken like a true non-software-dev. If your ai is making the actual decisions behind your code - you (a) don't know what you're doing anyway and (b) will end up with a ball of mud. The whole point of ai _assistants_ is to fill in gaps - gaps that, eg, I used to fill with stackoverflow. The sheer number of times I have to push back against claude's stupid ideas - coding is 100% _NOT_ solved. Claude is well useful, sure, but rabbit-holes down paths that are obviously wrong to anyone with the experience; is quite happy to speculate on the finest details with very little information (so is often wildly wrong), and works best under the guidance of someone who actually does know what they're doing and can stop a disaster before it happens.
If you're not experiencing this, it's simply a signal of your lack of skill in the area. These tools are useful, but they do not run the show, thank goodness, because we'd need to level up our power game on this planet to deal with their klunky shit code.
yeah, the gmail dots-alias thing is a fucking nightmare
I haven't been scammed like this, but I know to check for stray dots in incoming mail's To field because there's someone who was granted my email (with a period in it) and I keep getting their mails (holiday photos, amazon emails, and just recently, a belgian second-hand item-selling platform). So it doesn't surprise me in the least that someone is taking advantage of this. If google had even an iota of one good programmer left, they'd change this. But apparently not, because this great big hole has been there since the beginning, and will likely stay there until the end.
A QR code to defeat bots? what's next? spaghetti to defeat italians? what's stopping ai agents from reading, decoding, and following the QR code? Nothing in the article gives me any confidence whatsoever.
"""
Use USB charging for the heart rate monitor, step tracker and notifications. Time display is powered by solar charging alone when the battery runs low.
"""
I have a gshock already (GM-B2100D-1A) and I love it - I especially love that it should never be opened, always just works, and it looks ok too (:
the best idea I can suggest is to set up a virtual debit/credit card, if your bank permits it - for example, I have one which I keep at a level that's enough to pay for my monthly patreons, etc, but which will not pay when it hits zero. Then I don't have to worry (that much) about, eg, a data breach.